CN112307287B - Method and device for data classification and processing of power Internet of things based on cloud-edge collaborative architecture - Google Patents

Abstract

Translated fromChinese

本发明提出一种基于云边协同架构的电力物联网数据分类处理方法及装置，其中方法包括在边缘层汇总收集原始数据并根据数据的来源，将数据分为上载数据和待处理数据；利用随机森林算法对待处理数据进行分类和处理，得到分类结果数据；边缘层将所述上载数据和分类结果数据上传至云端层，在云端层利用LSTM‑FCN数据分类模型对所述上载数据和分类结果数据进行分类存储。上述基于云边协同架构的电力物联网数据分类处理方法及装置可以避免数据上传和等待数据返回的延迟，达到了对数据的快速响应，并且提升了故障诊断分类的准确性。

The present invention provides a method and device for classifying and processing power Internet of Things data based on a cloud-edge collaborative architecture, wherein the method includes collecting and collecting raw data at the edge layer and dividing the data into upload data and data to be processed according to the source of the data; using random data The forest algorithm classifies and processes the data to be processed, and obtains classification result data; the edge layer uploads the uploaded data and classification result data to the cloud layer, and uses the LSTM-FCN data classification model in the cloud layer to classify the uploaded data and classification result data. Classified storage. The above-mentioned method and device for classifying and processing data of the power Internet of Things based on the cloud-edge collaboration architecture can avoid the delay of data uploading and waiting for data return, achieve fast response to data, and improve the accuracy of fault diagnosis and classification.

Description

Translated fromChinese

基于云边协同架构的电力物联网数据分类处理方法及装置Method and device for data classification and processing of power Internet of things based on cloud-edge collaborative architecture

技术领域technical field

本发明涉及数据分类技术领域，尤其涉及一种基于云边协同架构的电力物联网数据分类处理方法及装置。The invention relates to the technical field of data classification, and in particular, to a method and device for classifying and processing power Internet of Things data based on a cloud-edge collaboration architecture.

背景技术Background technique

随着电力物联网技术的不断发展，电网中的数据也在不断增多。电力行业的数据来源十分复杂，有来自于发电厂、变电站的设备状态信息，有来自于众多地区的电力用量信息，还有来自于偏远地区的监测各类设施的传感器的信息等等。现有的基于云计算架构的电力物联网会将这些数据全部传入云端进行数据分析和处理，这不但会使云端背负较大的计算压力，也会造成数据传输路线的巨大宽带负担。这种模式已经无法满足行业日益増长的数据处理需求。With the continuous development of power Internet of Things technology, the data in the power grid is also increasing. The data sources in the power industry are very complex, including equipment status information from power plants and substations, power consumption information from many areas, and information from sensors monitoring various facilities in remote areas. The existing power Internet of Things based on cloud computing architecture will transfer all these data to the cloud for data analysis and processing, which will not only burden the cloud with a large computing pressure, but also cause a huge bandwidth burden on the data transmission route. This model has been unable to meet the growing data processing needs of the industry.

发明内容SUMMARY OF THE INVENTION

为了解决现有技术中存在的问题，本申请提出了一种基于云边协同架构的电力物联网数据分类处理方法及装置，以避免数据上传和等待数据返回的延迟，达到了对数据的快速响应，并且提升了故障诊断分类的准确性。In order to solve the problems existing in the prior art, the present application proposes a method and device for classifying and processing power Internet of Things data based on a cloud-edge collaborative architecture, so as to avoid the delay of data uploading and waiting for data return, and achieve fast response to data , and improve the accuracy of fault diagnosis and classification.

为了实现上述目的，本申请的一方面提出了一种基于云边协同架构的电力物联网数据分类处理方法，包括以下步骤：In order to achieve the above purpose, an aspect of the present application proposes a method for classifying and processing data of the Internet of Things in electric power based on a cloud-edge collaborative architecture, which includes the following steps:

步骤1、在边缘层汇总收集原始数据并根据数据的来源，将数据分为上载数据和待处理数据；Step 1. Collect raw data at the edge layer and divide the data into uploaded data and pending data according to the source of the data;

步骤2、利用随机森林算法对待处理数据进行分类和处理，得到分类结果数据；Step 2, using the random forest algorithm to classify and process the data to be processed to obtain classification result data;

步骤3、边缘层将所述上载数据和分类结果数据上传至云端层，在云端层利用LSTM-FCN数据分类模型对所述上载数据和分类结果数据进行分类存储。Step 3: The edge layer uploads the uploaded data and the classification result data to the cloud layer, and uses the LSTM-FCN data classification model to classify and store the uploaded data and the classification result data in the cloud layer.

在一些实施例中，在所述步骤2中，还包括以下步骤：步骤21、对待处理数据进行数据预处理，获得特征数据集；步骤22、利用装袋算法从所述特征数据集中随机抽取训练数据子集训练随机森林模型；步骤23、利用测试数据集对训练后的随机森林模型进行测试，随机森林模型的输出结果为综合T棵CART树的分类结果数据。In some embodiments, the step 2 further includes the following steps: step 21, performing data preprocessing on the data to be processed to obtain a feature data set; step 22, randomly extracting training from the feature data set using a bagging algorithm The data subset trains the random forest model; Step 23 , using the test data set to test the trained random forest model, and the output result of the random forest model is the classification result data of the comprehensive T CART trees.

在一些实施例中，在所述步骤2中还包括以下步骤：步骤24、将所述分类结果数据下传到设备层，并打包等待发送到云端层。In some embodiments, the step 2 further includes the following steps: Step 24: Download the classification result data to the device layer, and package the data to wait for sending to the cloud layer.

在一些实施例中，在所述步骤21中，所述数据预处理包括对数据提取平均值、标准偏差、峰值、峰度、脉冲指数和偏度特征以得到特征数据集。In some embodiments, in the step 21, the data preprocessing includes extracting mean value, standard deviation, peak value, kurtosis, pulse index and skewness features from the data to obtain a feature dataset.

在一些实施例中，所述步骤23中，所述CART树的构建如下：每颗树的每个节点将其对应的数据集

依据其属性，通过自适应基尼系数分割为两个数据子集X₁，X₂，其中x_i表示第i个数据，其表达式为x_i＝{d₁,d₂,...,d_m}，n为数据个数，m为特征数量，则数据集X对应的自适应基尼系数公式为：

其中函数f(d_j)为统计d_j出现次数；数据集X依据第j个属性划分的自适应基尼系数公式为：

其中α为自适应阈值，选择对应基尼系数最小的第j个属性作为其数据集分裂的依据进行分割数据集。In some embodiments, in the step 23, the construction of the CART tree is as follows: each node of each tree converts its corresponding data set

According to its properties, it is divided into two data subsets X₁ , X₂ by the adaptive Gini coefficient, where_xi represents the ith data, and its expression is_xi ={d₁ ,d₂ ,...,d_m }, n is the number of data, m is the number of features, the adaptive Gini coefficient formula corresponding to the data set X is:

The function f(d_j ) is the number of occurrences of d_j ; the adaptive Gini coefficient formula of the data set X divided according to the jth attribute is:

Among them, α is the adaptive threshold, and the jth attribute with the smallest corresponding Gini coefficient is selected as the basis for its data set splitting to divide the data set.

在一些实施例中，所述步骤3包括以下步骤：步骤31、在云端层，接收来自边缘层的上载数据和分类结果数据；步骤32、对接收到的来自边缘层的上载数据和分类结果数据进行识别分类，将接收到的数据分成上载数据和分类结果数据；步骤33、对于分类结果数据依据其处理结果做出决策；步骤34、对于上载数据，采用LSTM-FCN算法进行分类存储。In some embodiments, the step 3 includes the following steps: step 31, in the cloud layer, receiving the upload data and classification result data from the edge layer; step 32, comparing the received upload data and classification result data from the edge layer Carry out identification and classification, and divide the received data into upload data and classification result data; step 33, make a decision on the classification result data according to its processing result; step 34, adopt LSTM-FCN algorithm to classify and store the uploaded data.

在一些实施例中，所述步骤34中，采用LSTM-FCN算法进行分类的过程如下：步骤341、将所述上载数据传入时间卷积块，数据在块中先通过时间卷积层，再使用批量归一化，之后通过ReLU激活函数得到该块的输出数据，所述输出数据再作为输入数据传送到下一个卷积块，重复两次上述过程；与此同时，将所述上载数据送入维度混洗层，之后将变幻后的数据输入到由Basic LSTM和Attention LSTM组成的LSTM块中，之后经过Dropout；步骤342、经过三个堆叠的时间卷积块的数据进入全局平均池化层；步骤343、将全局平均池化层和LSTM块的输出数据进行串联，发送到softmax分类层进行分类，得到分类结果。In some embodiments, in the step 34, the classification process using the LSTM-FCN algorithm is as follows: Step 341, the uploaded data is passed into the temporal convolution block, the data first passes through the temporal convolution layer in the block, and then Batch normalization is used, and then the output data of the block is obtained through the ReLU activation function, and the output data is then transmitted to the next convolution block as input data, and the above process is repeated twice; at the same time, the upload data is sent to Enter the dimension shuffling layer, and then input the transformed data into the LSTM block composed of Basic LSTM and Attention LSTM, and then go through Dropout; step 342, after three stacked temporal convolution blocks The data enters the global average pooling layer ; Step 343 , concatenate the output data of the global average pooling layer and the LSTM block, and send them to the softmax classification layer for classification to obtain a classification result.

本发明的另一方面提出了一种基于云边协同架构的电力物联网数据分类处理装置，包括以下模块：Another aspect of the present invention proposes a data classification and processing device for the Internet of Things based on cloud-edge collaboration architecture, including the following modules:

采集模块，用于在边缘层汇总收集原始数据并根据数据的来源，将数据分为上载数据和待处理数据；The acquisition module is used to aggregate and collect raw data at the edge layer and divide the data into uploaded data and pending data according to the source of the data;

处理模块，用于利用随机森林算法对待处理数据进行分类和处理，得到分类结果数据；The processing module is used to classify and process the data to be processed by using the random forest algorithm to obtain the classification result data;

执行模块，用于通过边缘层将所述上载数据和分类结果数据上传至云端层，在云端层利用LSTM-FCN数据分类模型对所述上载数据和分类结果数据进行分类存储。The execution module is configured to upload the uploaded data and the classification result data to the cloud layer through the edge layer, and use the LSTM-FCN data classification model to classify and store the uploaded data and the classification result data in the cloud layer.

在一些实施例中，所述处理模块具体用于：对待处理数据进行数据预处理，获得特征数据集；利用装袋算法从所述特征数据集中随机抽取训练数据子集训练随机森林模型；利用测试数据集对训练后的随机森林模型进行测试，随机森林模型的输出结果为综合T棵CART树的分类结果数据。In some embodiments, the processing module is specifically configured to: perform data preprocessing on the data to be processed to obtain a feature data set; randomly extract a subset of training data from the feature data set using a bagging algorithm to train a random forest model; use a test The data set tests the trained random forest model, and the output of the random forest model is the classification result data of the comprehensive T CART trees.

在一些实施例中，所述处理模块具体用于：将所述分类结果数据下传到设备层，并打包等待发送到云端层。In some embodiments, the processing module is specifically configured to: download the classification result data to the device layer, and package the data to wait for sending to the cloud layer.

在一些实施例中，所述处理模块具体用于：所述数据预处理包括对数据提取平均值、标准偏差、峰值、峰度、脉冲指数和偏度特征以得到特征数据集。In some embodiments, the processing module is specifically configured to: the data preprocessing includes extracting mean value, standard deviation, peak value, kurtosis, pulse index and skewness features from the data to obtain a feature data set.

在一些实施例中，所述处理模块具体用于：所述CART树的构建如下：每颗树的每个节点将其对应的数据集

依据其属性，通过自适应基尼系数分割为两个数据子集X₁，X₂，其中x_i表示第i个数据，其表达式为x_i＝{d₁,d₂,...,d_m}，n为数据个数，m为特征数量，则数据集X对应的自适应基尼系数公式为：

其中函数f(d_j)为统计d_j出现次数；数据集X依据第j个属性划分的自适应基尼系数公式为：

其中α为自适应阈值，选择对应基尼系数最小的第j个属性作为其数据集分裂的依据进行分割数据集。In some embodiments, the processing module is specifically configured to: construct the CART tree as follows: each node of each tree converts its corresponding data set

According to its properties, it is divided into two data subsets X₁ , X₂ by the adaptive Gini coefficient, where_xi represents the ith data, and its expression is_xi ={d₁ ,d₂ ,...,d_m }, n is the number of data, m is the number of features, the adaptive Gini coefficient formula corresponding to the data set X is:

The function f(d_j ) is the number of occurrences of d_j ; the adaptive Gini coefficient formula of the data set X divided according to the jth attribute is:

Among them, α is the adaptive threshold, and the jth attribute with the smallest corresponding Gini coefficient is selected as the basis for its data set splitting to divide the data set.

在一些实施例中，所述执行模块具体用于：在云端层，接收来自边缘层的上载数据和分类结果数据；对接收到的来自边缘层的上载数据和分类结果数据进行识别分类，将接收到的数据分成上载数据和分类结果数据；对于分类结果数据依据其处理结果做出决策；对于上载数据，采用LSTM-FCN算法进行分类存储。In some embodiments, the execution module is specifically configured to: at the cloud layer, receive the upload data and classification result data from the edge layer; identify and classify the received upload data and classification result data from the edge layer, and receive The received data is divided into upload data and classification result data; for classification result data, decisions are made based on its processing results; for upload data, the LSTM-FCN algorithm is used for classification and storage.

在一些实施例中，所述执行模块具体用于：将所述上载数据传入时间卷积块，数据在块中先通过时间卷积层，再使用批量归一化，之后通过ReLU激活函数得到该块的输出数据，所述输出数据再作为输入数据传送到下一个卷积块，重复两次上述过程；与此同时，将所述上载数据送入维度混洗层，之后将变幻后的数据输入到由Basic LSTM和AttentionLSTM组成的LSTM块中，之后经过Dropout；经过三个堆叠的时间卷积块的数据进入全局平均池化层；将全局平均池化层和LSTM块的输出数据进行串联，发送到softmax分类层进行分类，得到分类结果。In some embodiments, the execution module is specifically configured to: pass the uploaded data into a temporal convolution block, where the data first passes through a temporal convolution layer in the block, then uses batch normalization, and then obtains through the ReLU activation function The output data of this block is sent to the next convolution block as input data, and the above process is repeated twice; at the same time, the uploaded data is sent to the dimension shuffling layer, and then the transformed data is Input into the LSTM block composed of Basic LSTM and AttentionLSTM, and then go through Dropout; the data after three stacked time convolution blocks enter the global average pooling layer; the global average pooling layer and the output data of the LSTM block are concatenated, It is sent to the softmax classification layer for classification, and the classification result is obtained.

本申请的该方案的有益效果在于上述基于云边协同架构的电力物联网数据分类处理方法及装置通过在设备终端进行数据处理，避免了将数据上传和等待数据返回的延迟，达到了对数据的快速响应；其利用随机森林模型进行数据分类可以达到更高的准确率；利用LSTM-FCN模型也可以提升准确率。The beneficial effect of the solution of the present application is that the above-mentioned method and device for classifying and processing data of the power Internet of Things based on the cloud-edge collaborative architecture avoids the delay of uploading data and waiting for the data to return by performing data processing at the equipment terminal, and achieves the realization of data integrity. Fast response; using the random forest model for data classification can achieve higher accuracy; using the LSTM-FCN model can also improve the accuracy.

附图说明Description of drawings

图1示出了实施例中基于云边协同架构的电力物联网数据分类处理方法的架构图。FIG. 1 shows an architecture diagram of a method for classifying and processing data of the Internet of Things in electric power based on a cloud-edge collaboration architecture in an embodiment.

具体实施方式Detailed ways

下面结合附图对本申请的具体实施方式作进一步的说明。The specific embodiments of the present application will be further described below with reference to the accompanying drawings.

如图1所示，本申请所涉及的基于云边协同架构的电力物联网数据分类处理方法，包括以下步骤：As shown in FIG. 1 , the method for classifying and processing data of the power Internet of Things based on the cloud-edge collaborative architecture involved in the present application includes the following steps:

步骤1、在边缘层汇总收集原始数据并根据数据的来源，将数据分为上载数据UpD和待处理数据PeD。具体的，边缘层不对上载数据UpD进行处理，直接将其向上传输到云端层；待处理数据PeD则需要在边缘层进行处理，处理完毕后边缘层将结果下传到设备层，同时，将处理之后的数据上传到云端层，在云端层等待进一步分析。Step 1. Collect and collect raw data at the edge layer and divide the data into upload data UpD and pending data PeD according to the source of the data. Specifically, the edge layer does not process the uploaded data UpD, but directly transmits it to the cloud layer; the data to be processed PeD needs to be processed at the edge layer, and after the processing is completed, the edge layer downloads the result to the device layer, and at the same time, the processing Subsequent data is uploaded to the cloud layer, where it awaits further analysis.

步骤2、利用随机森林算法对待处理数据PeD进行分类和处理，得到分类结果数据PeD’。Step 2. Use the random forest algorithm to classify and process the data PeD to be processed, and obtain the classification result data PeD'.

步骤3、边缘层将所述上载数据UpD和分类结果数据PeD’上传至云端层，在云端层利用LSTM-FCN数据分类模型对所述上载数据UpD和分类结果数据PeD’进行分类存储。Step 3, the edge layer uploads the uploading data UpD and the classification result data PeD' to the cloud layer, and utilizes the LSTM-FCN data classification model in the cloud layer to classify and store the uploading data UpD and the classification result data PeD'.

其中在所述步骤2中，还包括以下步骤：Wherein, in the step 2, the following steps are also included:

步骤21、对待处理数据PeD进行数据预处理，获得特征数据集。所述数据预处理包括对数据提取平均值、标准偏差、峰值、峰度、脉冲指数和偏度特征以得到特征数据集。Step 21: Perform data preprocessing on the data PeD to be processed to obtain a feature data set. The data preprocessing includes extracting mean value, standard deviation, peak value, kurtosis, impulse index and skewness features from the data to obtain a feature dataset.

步骤22、利用装袋算法(Bagging算法)从所述特征数据集中随机抽取训练数据子集训练随机森林模型。Step 22: Use a bagging algorithm (Bagging algorithm) to randomly extract a subset of training data from the feature data set to train a random forest model.

步骤23、利用测试数据集对训练后的随机森林模型进行测试，随机森林模型的输出结果为综合T棵CART树的分类结果数据PeD’。Step 23, using the test data set to test the trained random forest model, and the output result of the random forest model is the classification result data PeD' of the comprehensive T CART trees.

步骤24、将上述分类结果数据PeD’下传到设备层，并打包等待发送到云端层。Step 24, the above-mentioned classification result data PeD' is downloaded to the device layer, and packaged and waited to be sent to the cloud layer.

具体的，所述步骤23中，所述CART树的构建如下：每颗树的每个节点将其对应的数据集

依据其属性，通过自适应基尼系数分割为两个数据子集X₁，X₂，其中x_i表示第i个数据，其表达式为x_i＝{d₁,d₂,...,d_m}，n为数据个数，m为特征数量，则数据集X对应的自适应基尼系数公式为：

其中函数f(d_j)为统计d_j出现次数；数据集X依据第j个属性划分的自适应基尼系数公式为：

其中α为自适应阈值，选择对应基尼系数最小的第j个属性作为其数据集分裂的依据进行分割数据集。Specifically, in the step 23, the construction of the CART tree is as follows: each node of each tree converts its corresponding data set

According to its properties, it is divided into two data subsets X₁ , X₂ by the adaptive Gini coefficient, where_xi represents the ith data, and its expression is_xi ={d₁ ,d₂ ,...,d_m }, n is the number of data, m is the number of features, the adaptive Gini coefficient formula corresponding to the data set X is:

The function f(d_j ) is the number of occurrences of d_j ; the adaptive Gini coefficient formula of the data set X divided according to the jth attribute is:

Among them, α is the adaptive threshold, and the jth attribute with the smallest corresponding Gini coefficient is selected as the basis for its data set splitting to divide the data set.

所述步骤3包括以下步骤：The step 3 includes the following steps:

步骤31、在云端层，接收来自边缘层的上载数据UpD和分类结果数据PeD’。Step 31: In the cloud layer, receive the upload data UpD and the classification result data PeD' from the edge layer.

步骤32、对接收到的来自边缘层的上载数据UpD和分类结果数据PeD’进行识别分类，将接收到的数据分成上载数据UpD和分类结果数据PeD’。Step 32: Identify and classify the received upload data UpD and classification result data PeD' from the edge layer, and divide the received data into upload data UpD and classification result data PeD'.

步骤33、对于分类结果数据PeD’依据其处理结果做出决策。Step 33: Make a decision for the classification result data PeD' according to its processing result.

步骤34、对于上载数据UpD，采用LSTM-FCN算法进行分类存储，以便于之后的大数据分析。Step 34: For the uploaded data UpD, use the LSTM-FCN algorithm to classify and store, so as to facilitate subsequent big data analysis.

具体的，所述步骤34中，采用LSTM-FCN算法进行分类的过程如下：Specifically, in the step 34, the classification process using the LSTM-FCN algorithm is as follows:

步骤341、将所述上载数据UpD传入时间卷积块，数据在块中先通过时间卷积层，再使用批量归一化，之后通过ReLU激活函数得到该块的输出数据，所述输出数据再作为输入数据传送到下一个卷积块，重复两次上述过程；与此同时，将所述上载数据UpD送入维度混洗层(dimension shuffle layer)，之后将变幻后的数据输入到由Basic LSTM和AttentionLSTM组成的LSTM块中，之后经过Dropout。Step 341: Pass the uploaded data UpD into the time convolution block, the data first passes through the time convolution layer in the block, and then uses batch normalization, and then obtains the output data of the block through the ReLU activation function, and the output data It is then transmitted to the next convolution block as input data, and the above process is repeated twice; at the same time, the uploaded data UpD is sent to the dimension shuffle layer, and then the changed data is input to the Basic In the LSTM block composed of LSTM and AttentionLSTM, it goes through Dropout.

步骤342、经过三个堆叠的时间卷积块的数据进入全局平均池化层。Step 342: The data of the three stacked temporal convolution blocks enter the global average pooling layer.

步骤343、将全局平均池化层和LSTM块的输出数据进行串联，发送到softmax分类层进行分类，得到分类结果。Step 343 , concatenate the output data of the global average pooling layer and the LSTM block, and send them to the softmax classification layer for classification to obtain a classification result.

本申请所涉及的基于云边协同架构的电力物联网数据分类处理装置，包括以下模块：采集模块，用于在边缘层汇总收集原始数据并根据数据的来源，将数据分为上载数据UpD和待处理数据PeD；处理模块，用于利用随机森林算法对待处理数据PeD进行分类和处理，得到分类结果数据PeD’；执行模块，用于通过边缘层将所述上载数据UpD和分类结果数据PeD’上传至云端层，在云端层利用LSTM-FCN数据分类模型对所述上载数据UpD和分类结果数据PeD’进行分类存储。The power Internet of Things data classification and processing device based on the cloud-edge collaborative architecture involved in this application includes the following modules: a collection module, which is used to aggregate and collect raw data at the edge layer and divide the data into upload data UpD and pending data according to the source of the data. processing data PeD; processing module, for using random forest algorithm to classify and process the data PeD to be processed, to obtain classification result data PeD'; execution module, for uploading the upload data UpD and the classification result data PeD' through the edge layer To the cloud layer, the LSTM-FCN data classification model is used in the cloud layer to classify and store the uploaded data UpD and the classification result data PeD'.

其中所述处理模块具体用于：对待处理数据PeD进行数据预处理，获得特征数据集，所述数据预处理包括对数据提取平均值、标准偏差、峰值、峰度、脉冲指数和偏度特征以得到特征数据集；利用装袋算法从所述特征数据集中随机抽取训练数据子集训练随机森林模型；利用测试数据集对训练后的随机森林模型进行测试，随机森林模型的输出结果为综合T棵CART树的分类结果数据PeD’；将所述分类结果数据PeD’下传到设备层，并打包等待发送到云端层。The processing module is specifically used to: perform data preprocessing on the data to be processed PeD to obtain a feature data set, and the data preprocessing includes extracting the average value, standard deviation, peak value, kurtosis, pulse index and skewness characteristics of the data to obtain a characteristic data set. Obtain a feature data set; use the bagging algorithm to randomly extract training data subsets from the feature data set to train the random forest model; use the test data set to test the trained random forest model, and the output result of the random forest model is a comprehensive T tree The classification result data PeD' of the CART tree; the classification result data PeD' is downloaded to the device layer, and is packaged and waiting to be sent to the cloud layer.

所述CART树的构建如下：每颗树的每个节点将其对应的数据集

依据其属性，通过自适应基尼系数分割为两个数据子集X₁，X₂，其中x_i表示第i个数据，其表达式为x_i＝{d₁,d₂,...,d_m}，n为数据个数，m为特征数量，则数据集X对应的自适应基尼系数公式为：

其中函数f(d_j)为统计d_j出现次数；数据集X依据第j个属性划分的自适应基尼系数公式为：

其中α为自适应阈值，选择对应基尼系数最小的第j个属性作为其数据集分裂的依据进行分割数据集。The construction of the CART tree is as follows: each node of each tree associates its corresponding data set

According to its properties, it is divided into two data subsets X₁ , X₂ by the adaptive Gini coefficient, where_xi represents the ith data, and its expression is_xi ={d₁ ,d₂ ,...,d_m }, n is the number of data, m is the number of features, the adaptive Gini coefficient formula corresponding to the data set X is:

The function f(d_j ) is the number of occurrences of d_j ; the adaptive Gini coefficient formula of the data set X divided according to the jth attribute is:

Among them, α is the adaptive threshold, and the jth attribute with the smallest corresponding Gini coefficient is selected as the basis for its data set splitting to divide the data set.

所述执行模块具体用于：在云端层，接收来自边缘层的上载数据UpD和分类结果数据PeD’；对接收到的来自边缘层的上载数据UpD和分类结果数据PeD’进行识别分类，将接收到的数据分成上载数据UpD和分类结果数据PeD’；对于分类结果数据PeD’依据其处理结果做出决策；对于上载数据UpD，采用LSTM-FCN算法进行分类存储。The execution module is specifically used to: at the cloud layer, receive the upload data UpD and the classification result data PeD' from the edge layer; identify and classify the received upload data UpD and the classification result data PeD' from the edge layer, and receive The received data is divided into upload data UpD and classification result data PeD'; for the classification result data PeD', a decision is made according to its processing results; for the upload data UpD, the LSTM-FCN algorithm is used for classification and storage.

所述执行模块还具体用于：将所述上载数据传入时间卷积块，数据在块中先通过时间卷积层，再使用批量归一化，之后通过ReLU激活函数得到该块的输出数据，所述输出数据再作为输入数据传送到下一个卷积块，重复两次上述过程；与此同时，将所述上载数据送入维度混洗层，之后将变幻后的数据输入到由Basic LSTM和Attention LSTM组成的LSTM块中，之后经过Dropout；经过三个堆叠的时间卷积块的数据进入全局平均池化层；将全局平均池化层和LSTM块的输出数据进行串联，发送到softmax分类层进行分类，得到分类结果。The execution module is also specifically used for: passing the uploaded data into the time convolution block, the data first passes through the time convolution layer in the block, and then uses batch normalization, and then obtains the output data of the block through the ReLU activation function. , the output data is then transmitted to the next convolution block as input data, and the above process is repeated twice; at the same time, the uploaded data is sent to the dimension shuffling layer, and then the transformed data is input to the Basic LSTM In the LSTM block composed of the Attention LSTM, and then Dropout; the data of the three stacked time convolution blocks enter the global average pooling layer; the output data of the global average pooling layer and the LSTM block are connected in series and sent to the softmax classification The layers are classified and the classification results are obtained.

本申请所涉及的基于云边协同架构的电力物联网数据分类处理方法及装置通过在设备终端进行数据处理，避免了将数据上传和等待数据返回的延迟，达到了对数据的快速响应；其利用随机森林模型进行数据分类可以达到更高的准确率；利用LSTM-FCN模型也可以提升准确率。The method and device for classifying and processing power Internet of Things data based on the cloud-edge collaborative architecture involved in this application avoids the delay of uploading data and waiting for data to return by performing data processing at the equipment terminal, and achieves a quick response to the data; The random forest model can achieve higher accuracy for data classification; using the LSTM-FCN model can also improve the accuracy.

本领域内的技术人员应明白，本申请的实施例可提供为方法、系统、或计算机程序产品。因此，本申请可采用完全硬件实施例、完全软件实施例、或结合软件和硬件方面的实施例的形式。而且，本申请可采用在一个或多个其中包含有计算机可用程序代码的计算机可用存储介质(包括但不限于磁盘存储器、CD-ROM、光学存储器等)上实施的计算机程序产品的形式。As will be appreciated by those skilled in the art, the embodiments of the present application may be provided as a method, a system, or a computer program product. Accordingly, the present application may take the form of an entirely hardware embodiment, an entirely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present application may take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) having computer-usable program code embodied therein.

本申请是参照根据本申请实施例的方法、设备(系统)、和计算机程序产品的流程图和/或方框图来描述的。应理解可由计算机程序指令实现流程图和/或方框图中的每一流程和/或方框、以及流程图和/或方框图中的流程和/或方框的结合。可提供这些计算机程序指令到通用计算机、专用计算机、嵌入式处理机或其他可编程数据处理设备的处理器以产生一个机器，使得通过计算机或其他可编程数据处理设备的处理器执行的指令产生用于实现在流程图一个流程或多个流程和/或方框图一个方框或多个方框中指定的功能的装置。The present application is described with reference to flowchart illustrations and/or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the present application. It will be understood that each flow and/or block in the flowchart illustrations and/or block diagrams, and combinations of flows and/or blocks in the flowchart illustrations and/or block diagrams, can be implemented by computer program instructions. These computer program instructions may be provided to the processor of a general purpose computer, special purpose computer, embedded processor or other programmable data processing device to produce a machine such that the instructions executed by the processor of the computer or other programmable data processing device produce Means for implementing the functions specified in a flow or flow of a flowchart and/or a block or blocks of a block diagram.

这些计算机程序指令也可存储在能引导计算机或其他可编程数据处理设备以特定方式工作的计算机可读存储器中，使得存储在该计算机可读存储器中的指令产生包括指令装置的制造品，该指令装置实现在流程图一个流程或多个流程和/或方框图一个方框或多个方框中指定的功能。These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing apparatus to function in a particular manner, such that the instructions stored in the computer-readable memory result in an article of manufacture comprising instruction means, the instructions The apparatus implements the functions specified in the flow or flow of the flowcharts and/or the block or blocks of the block diagrams.

这些计算机程序指令也可装载到计算机或其他可编程数据处理设备上，使得在计算机或其他可编程设备上执行一系列操作步骤以产生计算机实现的处理，从而在计算机或其他可编程设备上执行的指令提供用于实现在流程图一个流程或多个流程和/或方框图一个方框或多个方框中指定的功能的步骤。These computer program instructions can also be loaded on a computer or other programmable data processing device to cause a series of operational steps to be performed on the computer or other programmable device to produce a computer-implemented process such that The instructions provide steps for implementing the functions specified in the flow or blocks of the flowcharts and/or the block or blocks of the block diagrams.

最后应当说明的是：以上实施例仅用以说明本发明的技术方案而非对其限制，尽管参照上述实施例对本发明进行了详细的说明，所属领域的普通技术人员应当理解：依然可以对本发明的具体实施方式进行修改或者等同替换，而未脱离本发明精神和范围的任何修改或者等同替换，其均应涵盖在本发明的权利要求保护范围之内。Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit them. Although the present invention has been described in detail with reference to the above embodiments, those of ordinary skill in the art should understand that: the present invention can still be Modifications or equivalent replacements are made to the specific embodiments of the present invention, and any modifications or equivalent replacements that do not depart from the spirit and scope of the present invention shall be included within the protection scope of the claims of the present invention.

Claims (10)

Translated fromChinese

1.一种基于云边协同架构的电力物联网数据分类处理方法，其特征在于：包括以下步骤：1. A method for classifying and processing power Internet of Things data based on cloud-edge collaborative architecture, characterized in that: comprising the following steps:步骤1、在边缘层汇总收集原始数据并根据数据的来源，将数据分为上载数据和待处理数据；Step 1. Collect raw data at the edge layer and divide the data into uploaded data and pending data according to the source of the data;步骤2、利用随机森林算法对待处理数据进行分类和处理，得到分类结果数据；Step 2, using the random forest algorithm to classify and process the data to be processed to obtain classification result data;步骤3、边缘层将所述上载数据和分类结果数据上传至云端层，在云端层利用LSTM-FCN数据分类模型对所述上载数据和分类结果数据进行分类存储；具体包括以下步骤：Step 3. The edge layer uploads the uploaded data and the classification result data to the cloud layer, and uses the LSTM-FCN data classification model to classify and store the uploaded data and the classification result data in the cloud layer; specifically, the following steps are included:步骤31、在云端层，接收来自边缘层的上载数据和分类结果数据；Step 31, in the cloud layer, receive the upload data and classification result data from the edge layer;步骤32、对接收到的来自边缘层的上载数据和分类结果数据进行识别分类，将接收到的数据分成上载数据和分类结果数据；Step 32, identify and classify the received upload data and classification result data from the edge layer, and divide the received data into upload data and classification result data;步骤33、对于分类结果数据依据其处理结果做出决策；Step 33, make a decision for the classification result data according to its processing result;步骤34、对于上载数据，采用LSTM-FCN算法进行分类存储；具体的，采用LSTM-FCN算法进行分类的过程如下：Step 34: For the uploaded data, use the LSTM-FCN algorithm to classify and store; specifically, the process of using the LSTM-FCN algorithm to classify the data is as follows:步骤341、将所述上载数据传入时间卷积块，数据在块中先通过时间卷积层，再使用批量归一化，之后通过ReLU激活函数得到该块的输出数据，所述输出数据再作为输入数据传送到下一个卷积块，重复两次上述过程；与此同时，将所述上载数据送入维度混洗层，之后将变幻后的数据输入到由Basic LSTM和Attention LSTM组成的LSTM块中，之后经过Dropout；Step 341: Introduce the uploaded data into the time convolution block, the data first passes through the time convolution layer in the block, and then uses batch normalization, and then obtains the output data of the block through the ReLU activation function, and the output data is then processed. The above process is repeated twice as input data to the next convolution block; at the same time, the uploaded data is sent to the dimension shuffling layer, and then the transformed data is input to the LSTM composed of Basic LSTM and Attention LSTM In the block, after Dropout;步骤342、经过三个堆叠的时间卷积块的数据进入全局平均池化层；Step 342, the data of the three stacked time convolution blocks enter the global average pooling layer;步骤343、将全局平均池化层和LSTM块的输出数据进行串联，发送到softmax分类层进行分类，得到分类结果。Step 343 , concatenate the output data of the global average pooling layer and the LSTM block, and send them to the softmax classification layer for classification to obtain a classification result.2.根据权利要求1所述的基于云边协同架构的电力物联网数据分类处理方法，其特征在于：在所述步骤2中，还包括以下步骤：2. The method for classifying and processing power Internet of Things data based on cloud-edge collaborative architecture according to claim 1, characterized in that: in the step 2, further comprising the following steps:步骤21、对待处理数据进行数据预处理，获得特征数据集；Step 21: Perform data preprocessing on the data to be processed to obtain a feature data set;步骤22、利用装袋算法从所述特征数据集中随机抽取训练数据子集训练随机森林模型；Step 22, using the bagging algorithm to randomly extract a training data subset from the feature data set to train a random forest model;步骤23、利用测试数据集对训练后的随机森林模型进行测试，随机森林模型的输出结果为综合T棵CART树的分类结果数据。In step 23, the trained random forest model is tested by using the test data set, and the output result of the random forest model is the classification result data of the comprehensive T CART trees.3.根据权利要求2所述的基于云边协同架构的电力物联网数据分类处理方法，其特征在于：在所述步骤2中还包括以下步骤：3. The method for classifying and processing power Internet of Things data based on a cloud-edge collaborative architecture according to claim 2, wherein the step 2 further comprises the following steps:步骤24、将所述分类结果数据下传到设备层，并打包等待发送到云端层。Step 24: Download the classification result data to the device layer, package it and wait to send it to the cloud layer.4.根据权利要求2所述的基于云边协同架构的电力物联网数据分类处理方法，其特征在于：在所述步骤21中，所述数据预处理包括对数据提取平均值、标准偏差、峰值、峰度、脉冲指数和偏度特征以得到特征数据集。4. The method for classifying and processing power Internet of Things data based on a cloud-edge collaborative architecture according to claim 2, wherein in the step 21, the data preprocessing comprises extracting the average value, standard deviation, peak value of the data , kurtosis, impulse index, and skewness features to obtain a feature dataset.5.根据权利要求2所述的基于云边协同架构的电力物联网数据分类处理方法，其特征在于：所述步骤23中，所述CART树的构建如下：每颗树的每个节点将其对应的数据集

依据其属性，通过自适应基尼系数分割为两个数据子集X₁，X₂，其中x_i表示第i个数据，其表达式为x_i＝{d₁,d₂,...,d_m}，n为数据个数，m为特征数量，则数据集X对应的自适应基尼系数公式为：

其中函数f(d_j)为统计d_j出现次数；数据集X依据第j个属性划分的自适应基尼系数公式为：

其中α为自适应阈值，选择对应基尼系数最小的第j个属性作为其数据集分裂的依据进行分割数据集。5. The method for classifying and processing power Internet of Things data based on cloud-edge collaborative architecture according to claim 2, characterized in that: in the step 23, the construction of the CART tree is as follows: each node of each tree corresponding dataset

According to its properties, it is divided into two data subsets X₁ , X₂ by the adaptive Gini coefficient, where_xi represents the ith data, and its expression is_xi ={d₁ ,d₂ ,...,d_m }, n is the number of data, m is the number of features, the adaptive Gini coefficient formula corresponding to the data set X is:

The function f(d_j ) is the number of occurrences of d_j ; the adaptive Gini coefficient formula of the data set X divided according to the jth attribute is:

Among them, α is the adaptive threshold, and the jth attribute with the smallest corresponding Gini coefficient is selected as the basis for its data set splitting to divide the data set.6.一种基于云边协同架构的电力物联网数据分类处理装置，其特征在于：包括以下模块：6. A power Internet of Things data classification and processing device based on a cloud-edge collaborative architecture, characterized in that it comprises the following modules:采集模块，用于在边缘层汇总收集原始数据并根据数据的来源，将数据分为上载数据和待处理数据；The acquisition module is used to aggregate and collect raw data at the edge layer and divide the data into uploaded data and pending data according to the source of the data;处理模块，用于利用随机森林算法对待处理数据进行分类和处理，得到分类结果数据；The processing module is used to classify and process the data to be processed by using the random forest algorithm to obtain the classification result data;执行模块，用于通过边缘层将所述上载数据和分类结果数据上传至云端层，在云端层利用LSTM-FCN数据分类模型对所述上载数据和分类结果数据进行分类存储；所述执行模块具体用于：在云端层，接收来自边缘层的上载数据和分类结果数据；对接收到的来自边缘层的上载数据和分类结果数据进行识别分类，将接收到的数据分成上载数据和分类结果数据；对于分类结果数据依据其处理结果做出决策；对于上载数据，采用LSTM-FCN算法进行分类存储；所述执行模块具体用于：将所述上载数据传入时间卷积块，数据在块中先通过时间卷积层，再使用批量归一化，之后通过ReLU激活函数得到该块的输出数据，所述输出数据再作为输入数据传送到下一个卷积块，重复两次上述过程；与此同时，将所述上载数据送入维度混洗层，之后将变幻后的数据输入到由Basic LSTM和Attention LSTM组成的LSTM块中，之后经过Dropout；经过三个堆叠的时间卷积块的数据进入全局平均池化层；将全局平均池化层和LSTM块的输出数据进行串联，发送到softmax分类层进行分类，得到分类结果。The execution module is used to upload the uploaded data and the classification result data to the cloud layer through the edge layer, and use the LSTM-FCN data classification model to classify and store the uploaded data and the classification result data in the cloud layer; the execution module specifically Used for: in the cloud layer, receive the upload data and classification result data from the edge layer; identify and classify the received upload data and classification result data from the edge layer, and divide the received data into upload data and classification result data; For the classification result data, make decisions according to its processing results; for the uploaded data, use the LSTM-FCN algorithm for classification and storage; Through the temporal convolution layer, batch normalization is used, and then the output data of the block is obtained through the ReLU activation function, and the output data is transmitted to the next convolution block as input data, and the above process is repeated twice; , send the uploaded data into the dimension shuffling layer, and then input the transformed data into the LSTM block composed of Basic LSTM and Attention LSTM, and then go through Dropout; after three stacked time convolution blocks The data enters the global Average pooling layer; the output data of the global average pooling layer and the LSTM block are concatenated and sent to the softmax classification layer for classification, and the classification result is obtained.7.根据权利要求6所述的基于云边协同架构的电力物联网数据分类处理装置，其特征在于：所述处理模块具体用于：对待处理数据进行数据预处理，获得特征数据集；利用装袋算法从所述特征数据集中随机抽取训练数据子集训练随机森林模型；利用测试数据集对训练后的随机森林模型进行测试，随机森林模型的输出结果为综合T棵CART树的分类结果数据。7 . The power Internet of Things data classification and processing device based on the cloud-edge collaborative architecture according to claim 6 , wherein the processing module is specifically used to: perform data preprocessing on the data to be processed to obtain a feature data set; The bag algorithm randomly selects a subset of training data from the feature data set to train the random forest model; uses the test data set to test the trained random forest model, and the output result of the random forest model is the classification result data of the comprehensive T CART trees.8.根据权利要求7所述的基于云边协同架构的电力物联网数据分类处理装置，其特征在于：所述处理模块具体用于：将所述分类结果数据下传到设备层，并打包等待发送到云端层。8 . The power Internet of Things data classification and processing device based on cloud-edge collaborative architecture according to claim 7 , wherein the processing module is specifically used for: downloading the classification result data to the device layer, and packaging and waiting. 9 . sent to the cloud tier.9.根据权利要求7所述的基于云边协同架构的电力物联网数据分类处理装置，其特征在于：所述处理模块具体用于：所述数据预处理包括对数据提取平均值、标准偏差、峰值、峰度、脉冲指数和偏度特征以得到特征数据集。9 . The device for classifying and processing power Internet of Things data based on a cloud-edge collaborative architecture according to claim 7 , wherein the processing module is specifically configured to: the data preprocessing includes extracting the average value, standard deviation, Peak, kurtosis, impulse index, and skewness features to obtain a feature dataset.10.根据权利要求7所述的基于云边协同架构的电力物联网数据分类处理装置，其特征在于：所述处理模块具体用于：所述CART树的构建如下：每颗树的每个节点将其对应的数据集

依据其属性，通过自适应基尼系数分割为两个数据子集X₁，X₂，其中x_i表示第i个数据，其表达式为x_i＝{d₁,d₂,...,d_m}，n为数据个数，m为特征数量，则数据集X对应的自适应基尼系数公式为：

其中函数f(d_j)为统计d_j出现次数；数据集X依据第j个属性划分的自适应基尼系数公式为：

其中α为自适应阈值，选择对应基尼系数最小的第j个属性作为其数据集分裂的依据进行分割数据集。10. The device for classifying and processing data of power Internet of Things based on cloud-edge collaborative architecture according to claim 7, wherein the processing module is specifically used for: the construction of the CART tree is as follows: each node of each tree the corresponding dataset

According to its properties, it is divided into two data subsets X₁ , X₂ by the adaptive Gini coefficient, where_xi represents the ith data, and its expression is_xi ={d₁ ,d₂ ,...,d_m }, n is the number of data, m is the number of features, the adaptive Gini coefficient formula corresponding to the data set X is:

The function f(d_j ) is the number of occurrences of d_j ; the adaptive Gini coefficient formula of the data set X divided according to the jth attribute is:

Among them, α is the adaptive threshold, and the jth attribute with the smallest corresponding Gini coefficient is selected as the basis for its data set splitting to divide the data set.

Images