CN115438897A - Industrial process product quality prediction method based on BLSTM neural network - Google Patents
Industrial process product quality prediction method based on BLSTM neural networkDownload PDFInfo
- Publication number
- CN115438897A CN115438897ACN202210537012.2ACN202210537012ACN115438897ACN 115438897 ACN115438897 ACN 115438897ACN 202210537012 ACN202210537012 ACN 202210537012ACN 115438897 ACN115438897 ACN 115438897A
- Authority
- CN
- China
- Prior art keywords
- data
- quality
- model
- product quality
- training
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Withdrawn
Links
Images
Classifications
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q10/00—Administration; Management
- G06Q10/06—Resources, workflows, human or project management; Enterprise or organisation planning; Enterprise or organisation modelling
- G06Q10/063—Operations research, analysis or management
- G06Q10/0639—Performance analysis of employees; Performance analysis of enterprise or organisation operations
- G06Q10/06395—Quality analysis or management
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q50/00—Information and communication technology [ICT] specially adapted for implementation of business processes of specific business sectors, e.g. utilities or tourism
- G06Q50/04—Manufacturing
Landscapes
- Business, Economics & Management (AREA)
- Human Resources & Organizations (AREA)
- Engineering & Computer Science (AREA)
- Strategic Management (AREA)
- Economics (AREA)
- Development Economics (AREA)
- Theoretical Computer Science (AREA)
- Marketing (AREA)
- Entrepreneurship & Innovation (AREA)
- Educational Administration (AREA)
- Tourism & Hospitality (AREA)
- Physics & Mathematics (AREA)
- General Business, Economics & Management (AREA)
- General Physics & Mathematics (AREA)
- General Health & Medical Sciences (AREA)
- Manufacturing & Machinery (AREA)
- Health & Medical Sciences (AREA)
- Primary Health Care (AREA)
- Game Theory and Decision Science (AREA)
- Operations Research (AREA)
- Quality & Reliability (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
Description
Translated fromChinese技术领域technical field
本发明涉及一种工业过程预测方法，特别是涉及一种基于BLSTM神经网络的工业过程产品质量预测方法。The invention relates to an industrial process prediction method, in particular to a BLSTM neural network-based industrial process product quality prediction method.
背景技术Background technique
随着国民经济的快速发展，工业生产规模不断扩大，现代工业过程向着非线性、非稳态、高噪声和高延迟等复杂化方向发展，产品质量越来越不能得到保证，人力、物力损失及其巨大，工业过程安全也直接关系到国家经济发展和人民生命财产安全，过程安全、产品质量以及节能减排增效逐渐成为现代工业的核心目标。虽然复杂工业过程关键变量在线分析和监控的新方法和新理论不断被提出，但是这些技术过度依赖于精确的模型识别和可靠的测量，特别是对关键过程变量的在线分析和过程监控。近年来，质量预测以其响应速度快、维护成本低、预测结果准确等优点被广泛应用于关键产品质量的在线评估。With the rapid development of the national economy, the scale of industrial production continues to expand, and the modern industrial process is developing in the direction of complexity such as nonlinearity, instability, high noise, and high delay. It is huge, and industrial process safety is also directly related to national economic development and people's life and property safety. Process safety, product quality, energy saving, emission reduction and efficiency enhancement have gradually become the core goals of modern industry. Although new methods and new theories for on-line analysis and monitoring of key variables in complex industrial processes are constantly being proposed, these technologies are overly dependent on accurate model identification and reliable measurement, especially for on-line analysis and process monitoring of key process variables. In recent years, quality prediction has been widely used in online evaluation of key product quality due to its advantages of fast response, low maintenance cost, and accurate prediction results.
基于数据分析进行质量预测是一种常见的预测方法，如偏最小二乘和支持向量机等，但是常见机器学习网络容易出现泛化能力问题，以及多层网络容易受到梯度消失和爆炸问题的影响。深度神经网络凭借其更好的性能被引入到预测建模，例如深度信念网络、堆叠自动编码器和循环神经网络。为了捕捉时间序列数据中的时间动态行为，动态递归神经网络应运而生，但仍存在梯度消失和梯度爆炸问题。长短期记忆网络诞生了，长短期记忆网络不仅能忘记过去无用的信息，还能判断当前的信息并将有用的信息存储在存储单元中，但该方法不能解决不同时间步长上的不同变量以及序列长度过长问题，而工业过程中质量变量具有较长的时间滞后特性。由此双向长短期记忆网络被发明出来解决了此问题，通过输入序列以顺、逆序形式输入网络进行训练，挖掘数据顺、逆序的依赖关系，进行双向时序的特征学习，以获取时序序列的特征相关性，充分挖掘其关键特征。Quality prediction based on data analysis is a common prediction method, such as partial least squares and support vector machines, but common machine learning networks are prone to generalization problems, and multi-layer networks are susceptible to gradient disappearance and explosion problems. . Deep neural networks have been introduced to predictive modeling due to their better performance, such as deep belief networks, stacked autoencoders, and recurrent neural networks. In order to capture the temporal dynamic behavior in time series data, dynamic recurrent neural networks came into being, but there are still problems of gradient disappearance and gradient explosion. The long short-term memory network was born. The long-term short-term memory network can not only forget the useless information in the past, but also judge the current information and store the useful information in the storage unit, but this method cannot solve the different variables on different time steps and The sequence length is too long, and the quality variable in the industrial process has a long time lag characteristic. Therefore, the two-way long-short-term memory network was invented to solve this problem. The input sequence is input into the network in the form of sequential and reverse order for training, and the dependence relationship between the sequence and reverse order of the data is mined, and the feature learning of the two-way time series is carried out to obtain the characteristics of the time series sequence. Correlation, and fully exploit its key features.
名词解释：Glossary:
MIC方法：最大互信息系数，即每个相关过程变量分别与产品质量变量进行互信息计算，通过互信息数来确定过程变量与质量变量的关联程度。MIC method: the maximum mutual information coefficient, that is, the mutual information calculation between each relevant process variable and the product quality variable, and the degree of correlation between the process variable and the quality variable is determined by the mutual information number.
BLSTM网络：双向长短期记忆网络，一种基于时间序列的预测方法。BLSTM network: bidirectional long short-term memory network, a time series based forecasting method.
发明内容Contents of the invention
本发明的目的在于提供一种基于BLSTM神经网络的工业过程产品质量预测方法，本发明针对提取的复杂工业数据，去除数据噪声、冗余性及加强模型的鲁棒性，同时深度挖掘工业过程时间序列的相关过程变量与产品质量变量的潜在联系，实现对产品质量的实时预测，及时的判断工业过程是否正常运行，及时避免运行状态故障，提高生产效率及减少资源的浪费，精准、直观的反映工业过程运行状态。The purpose of the present invention is to provide a method for predicting the quality of industrial process products based on BLSTM neural network. The present invention removes data noise, redundancy and enhances the robustness of the model for the extracted complex industrial data, and at the same time deeply excavates the industrial process time. The potential relationship between sequence-related process variables and product quality variables enables real-time prediction of product quality, timely judgment of whether the industrial process is running normally, timely avoiding operating status failures, improving production efficiency and reducing resource waste, and accurate and intuitive reflection Industrial process operating status.
本发明的目的是通过以下技术方案实现的：The purpose of the present invention is achieved through the following technical solutions:
一种基于BLSTM神经网络的工业过程产品质量预测方法，所述方法包括建立训练部分和测试部分产品质量预测体系；A method for predicting the quality of industrial process products based on BLSTM neural network, said method comprising establishing a training part and a test part of the product quality prediction system;
训练部分：Training part:
1）对工业过程数据进行降噪处理同时划分训练集和测试集，然后对训练数据进行归一化处理；1) Perform noise reduction processing on the industrial process data and divide the training set and test set at the same time, and then normalize the training data;
2）标准化后的训练数据代入最大互信息系数算法，按照公式(1)-(4)得到相关系数较大的过程变量，构成新的重组训练数据，完成与质量相关特征筛选的同时也减低数据的冗余性；2) The standardized training data is substituted into the maximum mutual information coefficient algorithm, and the process variables with large correlation coefficients are obtained according to formulas (1)-(4) to form new reorganized training data, which completes the screening of quality-related features and reduces the data redundancy;
3）构建双向长短期记忆网络模型，设定模型参数及输入样本的时间窗长度，将重组的训练数据输入至网络对其网络参数进行训练；3) Construct a two-way long-short-term memory network model, set the model parameters and the time window length of the input samples, and input the reorganized training data to the network to train its network parameters;
4）每次训练后对预测结果进行RMSE，R2值评估，采用网格搜索方式获取模型最优参数，误差满足阈值时得到质量预测模型；4) After each training, RMSE and R2 value evaluation are performed on the prediction results, and the optimal parameters of the model are obtained by grid search, and the quality prediction model is obtained when the error meets the threshold;
测试部分：Test section:
1）提取训练数据各变量的最大值与最小值对测试数据进行归一化处理；1) Extract the maximum value and minimum value of each variable of the training data to normalize the test data;
2）处理后的测试数据同训练部分，完成质量相关特征筛选得到重组测试数据；2) The processed test data is the same as the training part, and the quality-related feature screening is completed to obtain the reorganized test data;
3）将重组测试数据代入质量预测模型进行质量变量预测，通过质量预测模型获得产品质量变量预测值，有效调整生产过程出现的异常，确保工业生产的正常运行。3) Substitute the reorganized test data into the quality prediction model to predict the quality variables, obtain the product quality variable prediction values through the quality prediction model, effectively adjust the abnormalities in the production process, and ensure the normal operation of industrial production.
所述的一种基于BLSTM神经网络的工业过程产品质量预测方法，所述方法，具体步骤包括以下：Described a kind of industrial process product quality prediction method based on BLSTM neural network, described method, concrete steps comprise the following:
S1：基于工业现场传感器采集过程数据或者工业系统仿真平台获取的正常样本X和产品样本Y；S1: Based on the process data collected by industrial field sensors or the normal sample X and product sample Y obtained by the industrial system simulation platform;
S2：深入了解工业过程流程，把握相关过程变量与产品质量变量之间的联系，对样本进行降噪和归一化处理；S2: In-depth understanding of industrial process flow, grasp the relationship between relevant process variables and product quality variables, and perform noise reduction and normalization processing on samples;
S3：将相关过程变量与产品质量变量数据进行最大互信息特征筛选，确定模型的低维重构数据，降低数据的冗余性和减低神经网络的计算量；S3: Screen the relevant process variables and product quality variable data for maximum mutual information features, determine the low-dimensional reconstruction data of the model, reduce data redundancy and reduce the amount of calculation of the neural network;
S4：将重构数据按照4：1比例划分为训练数据集、测试数据集；S4: Divide the reconstructed data into training data set and test data set according to the ratio of 4:1;
S5：将重构训练数据代入到双向长短期记忆网络中学习其潜在关系，通过均方根误差(RMSE)、R2系数和损失函数(loss)等指标来优化模型的参数，直到确定质量预测模型；S5: Substitute the reconstructed training data into the bidirectional long-term short-term memory network to learn its potential relationship, and optimize the parameters of the model through indicators such as root mean square error (RMSE), R2 coefficient and loss function (loss) , until the quality prediction is determined Model;
S6：将重构测试数据代入到质量预测模型中进行产品预测，通过质量预测模型获得产品质量变量预测值，有效调整生产过程出现的异常，确保工业生产的正常运行。S6: Substitute the reconstructed test data into the quality prediction model for product prediction, obtain the predicted value of product quality variables through the quality prediction model, effectively adjust the abnormalities in the production process, and ensure the normal operation of industrial production.
本发明的优点与效果是：Advantage and effect of the present invention are:
1.本发明工业过程运行状态预测的稳定性和鲁棒性大大提高，双向长短期记忆网络具有可以处理长批次数据集和对信息有更强的记忆能力，它可以利用过去的信息和未来的信息进行很好的联系起来进行精准预测。相对于其他方法，双向长短期网络对工业过程状态预测精度，泛化能力和鲁棒性更强，满足工业过程监控预测的时效性要求。1. The stability and robustness of the industrial process operation state prediction of the present invention are greatly improved, and the two-way long-short-term memory network has the ability to process long-batch data sets and has a stronger memory capacity for information, and it can use past information and future The information is well connected to make accurate predictions. Compared with other methods, the bidirectional long-term and short-term network has stronger prediction accuracy, generalization ability and robustness for industrial process state, and meets the timeliness requirements of industrial process monitoring and prediction.
2.本发明工业过程数据预测更为直观和可靠，降低了监测人员对系统运行时的监管难度，提高了监测效率。本发明将工业过程运行中生产的产品质量进行预测，在线运用时，通过质量预测模型获得产品质量变量预测值，有效调整生产过程出现的异常，确保工业生产的正常运行。2. The prediction of industrial process data in the present invention is more intuitive and reliable, which reduces the difficulty for monitoring personnel to supervise the operation of the system and improves the monitoring efficiency. The invention predicts the quality of products produced during the operation of the industrial process. When used online, the predicted value of the product quality variable is obtained through the quality prediction model, so as to effectively adjust abnormalities in the production process and ensure the normal operation of industrial production.
附图说明Description of drawings
图1本发明不同模型的loss函数图；Fig. 1 is the loss function figure of different models of the present invention;
图2本发明测试样本对比图；Fig. 2 comparison chart of test samples of the present invention;
图3本发明不同模型质量变量的预测结果图；Fig. 3 is the prediction result figure of different model quality variables of the present invention;
图4本发明的整体流程示意图The overall flow diagram of Fig. 4 the present invention
图5本发明双向长短期记忆网络结构图。Fig. 5 is a structural diagram of the two-way long-short-term memory network of the present invention.
具体实施方式detailed description
下面结合附图所示实施例对本发明进行详细说明。The present invention will be described in detail below in conjunction with the embodiments shown in the accompanying drawings.
本发明一种基于BLSTM神经网络的工业过程产品质量预测方法，该方法了解整个工业过程的传感器位置设置以及采集的数据如何利用。该方法利用最大互信息系数进行特征变量筛选，去除数据的冗余性和减低网络计算量。同时，该方法对双向长短期记忆网络的参数进行网格搜索方式调优，建立一个完整的产品质量预测体系，其预测方法的整体步骤分为训练部分和测试部分。The invention discloses a method for predicting the quality of industrial process products based on BLSTM neural network. The method understands the sensor position setting of the entire industrial process and how to use the collected data. This method utilizes the maximum mutual information coefficient to filter the characteristic variables, removes the redundancy of data and reduces the amount of network calculation. At the same time, the method optimizes the parameters of the two-way long-term short-term memory network by means of grid search, and establishes a complete product quality prediction system. The overall steps of the prediction method are divided into a training part and a testing part.
包括以下步骤：Include the following steps:
S1：基于工业现场传感器采集过程数据或者工业系统仿真平台获取的正常样本X和产品样本Y；S1: Based on the process data collected by industrial field sensors or the normal sample X and product sample Y obtained by the industrial system simulation platform;
S2：深入了解工业过程流程，把握相关过程变量与产品质量变量之间的联系，对样本进行降噪和归一化处理；S2: In-depth understanding of industrial process flow, grasp the relationship between relevant process variables and product quality variables, and perform noise reduction and normalization processing on samples;
S3：将相关过程变量与产品质量变量数据进行最大互信息特征筛选，确定模型的低维重构数据，降低数据的冗余性和减低神经网络的计算量；S3: Screen the relevant process variables and product quality variable data for maximum mutual information features, determine the low-dimensional reconstruction data of the model, reduce data redundancy and reduce the amount of calculation of the neural network;
S4：将重构数据按照4：1比例划分为训练数据集、测试数据集；S4: Divide the reconstructed data into training data set and test data set according to the ratio of 4:1;
S5：将重构训练数据代入到双向长短期记忆网络中学习其潜在关系，通过均方根误差(RMSE)、R2系数和损失函数(loss)等指标来优化模型的参数，直到确定质量预测模型；S5: Substitute the reconstructed training data into the bidirectional long-term short-term memory network to learn its potential relationship, and optimize the parameters of the model through indicators such as root mean square error (RMSE), R2 coefficient and loss function (loss) , until the quality prediction is determined Model;
S6：将重构测试数据代入到质量预测模型中进行产品预测，通过质量预测模型获得产品质量变量预测值，有效调整生产过程出现的异常，确保工业生产的正常运行。S6: Substitute the reconstructed test data into the quality prediction model for product prediction, obtain the predicted value of product quality variables through the quality prediction model, effectively adjust the abnormalities in the production process, and ensure the normal operation of industrial production.
进一步的改进，步骤S3所述的对相关过程变量与产品质量变量进行最大互信息系数特征筛选，具体处理如下：As a further improvement, the feature screening of the maximum mutual information coefficient for the relevant process variables and product quality variables described in step S3 is specifically processed as follows:
S31：MIC概率公式如下：S31: The MIC probability formula is as follows:
S32：MIC筛选公式如下：S32: The MIC screening formula is as follows:
上式中
进一步的改进，步骤S4所述的将工业过程数据划分为训练数据和测试数据，具体步骤如下：As a further improvement, the industrial process data described in step S4 is divided into training data and test data, and the specific steps are as follows:
S4：本发明运用Tennessee Eastman工业过程数据案例进行方法验证，采集的过程数据由三大部分构成:过程控制变量、过程测量变量、成分测量变量，首先，采集47个过程变量来预测5个难于测量的产品变量，实验数据由5次TE正常工业过程数据组合而成，根据其同时间节点进行对应重构，每个变量都重构了4800个样本数据，将其划分为训练集部分和测试集部分，训练集部分为前4000个样本，测试集部分为后800个样本；其次，通过了解整套流程，期待变量之间的外在联系关系，进一步挖掘其深层次联系。S4: The present invention uses the Tennessee Eastman industrial process data case to carry out method verification, and the collected process data consists of three major parts: process control variables, process measurement variables, and component measurement variables. First, collect 47 process variables to predict 5 difficult-to-measure The product variable, the experimental data is composed of 5 TE normal industrial process data, and the corresponding reconstruction is carried out according to the same time node. Each variable has reconstructed 4800 sample data, which is divided into training set and test set. part, the training set part is the first 4000 samples, and the test set part is the last 800 samples; secondly, by understanding the whole process, expecting the external relationship between variables, and further excavating its deep connection.
进一步的改进，本发明使用的双向长短期记忆网络解决了神经网络难于对长序列建模问题，还可以保持模型性能的鲁棒性、稳定性；网络内部利用正反向两个长短期记忆网络很好的将上下文信息完美的联合了起来，实现了深层次关键变量的特征特征挖掘从而达到了更准确的预测。所述的步骤S5所述使用双向长短期记忆网络参数调优过程，获得预测值Y的具体步骤如下：As a further improvement, the two-way long-short-term memory network used in the present invention solves the problem that the neural network is difficult to model long sequences, and can also maintain the robustness and stability of the model performance; the network uses two forward and reverse long-term short-term memory networks It perfectly combines the context information and realizes the feature mining of deep-level key variables to achieve more accurate predictions. The specific steps for obtaining the predicted value Y using the two-way long-short-term memory network parameter tuning process described in step S5 are as follows:
S51:按照前面划分好的训练数据和测试数据,先将训练数据代入至双向长短期记忆网络训练；S51: According to the previously divided training data and test data, first substitute the training data into the bidirectional long-term short-term memory network training;
S52：双向长短期记忆网络结构如图5：正向层：自左向右循环神经网络层更新公式为
反向层：自右向左循环神经网络层更新公式为Reverse layer: The update formula of the right-to-left cyclic neural network layer is
输出层：前后两层循环神经网络层叠加后输出为Output layer: after superposition of the front and back two layers of cyclic neural network layers, the output is
式中：
S53：训练数据代入模型来网格搜索方式优化内部的各个参数，直至得出的预测结果达到最佳、loss函数达到最小和模型的评价指标达到合格状态的质量预测模型；S53: Substituting the training data into the model to optimize various internal parameters by means of grid search until the obtained prediction result reaches the best, the loss function reaches the minimum and the evaluation index of the model reaches the quality prediction model in a qualified state;
S54：将测试数据代入到质量预测模型进行预测结果和计算评价指标，从而实现对工业过程运行状态是否正常进行判断。S54: Substituting the test data into the quality prediction model to predict the result and calculate the evaluation index, so as to realize whether the operation state of the industrial process is normal or not.
进一步的改进，步骤S5所用的评价指标较少，可以通过多几个评价指标来对模型进行评价，这样可以更加全面对模型评价，使预测结果可以更加的精准，对工业过程也可以有更好的实质性监控。For further improvement, the evaluation indicators used in step S5 are less, and the model can be evaluated by several more evaluation indicators, so that the model can be evaluated more comprehensively, the prediction results can be more accurate, and the industrial process can also be better. substantive monitoring.
模型预测质量变量过程如下：The process of model prediction quality variable is as follows:
训练部分：Training part:
1）对工业过程数据进行降噪处理同时划分训练集和测试集，然后对训练数据进行归一化处理；1) Perform noise reduction processing on the industrial process data and divide the training set and test set at the same time, and then normalize the training data;
2）标准化后的训练数据代入最大互信息系数算法，按照公式(1)-(4)得到相关系数较大的过程变量，构成新的重组训练数据，完成与质量相关特征筛选的同时也减低数据的冗余性；2) The standardized training data is substituted into the maximum mutual information coefficient algorithm, and the process variables with large correlation coefficients are obtained according to formulas (1)-(4) to form new reorganized training data, which completes the screening of quality-related features and reduces the data redundancy;
3）构建双向长短期记忆网络模型，设定模型参数及输入样本的时间窗长度，将重组的训练数据输入至网络对其网络参数进行训练；3) Construct a two-way long-short-term memory network model, set the model parameters and the time window length of the input samples, and input the reorganized training data to the network to train its network parameters;
4）每次训练后对预测结果进行RMSE，R2值评估，采用网格搜索方式获取模型最优参数，误差满足阈值时得到质量预测模型；4) After each training, RMSE and R2 value evaluation are performed on the prediction results, and the optimal parameters of the model are obtained by grid search, and the quality prediction model is obtained when the error meets the threshold;
测试部分：Test section:
1）提取训练数据各变量的最大值与最小值对测试数据进行归一化处理；1) Extract the maximum value and minimum value of each variable of the training data to normalize the test data;
2）处理后的测试数据同训练部分2）完成质量相关特征筛选得到重组测试数据；2) The processed test data is the same as the training part 2) Complete the screening of quality-related features to obtain the reorganized test data;
3）将重组测试数据代入质量预测模型进行质量变量预测，通过质量预测模型获得产品质量变量预测值，有效调整生产过程出现的异常，确保工业生产的正常运行。3) Substitute the reorganized test data into the quality prediction model to predict the quality variables, obtain the product quality variable prediction values through the quality prediction model, effectively adjust the abnormalities in the production process, and ensure the normal operation of industrial production.
实施例：Example:
Tennessee Eastman过程是Eastman化学公司研发的一个按照实际化工反应过程模拟的一个仿真平台。此过程采集的数据由三大部分构成:过程控制变量、过程测量变量、成分测量变量，本文采用47个过程变量(变量编号为XMEAS(1)-XMEAS(36),XMV(1)-XMV(11))来预测5个难于测量的产品变量(变量编号为XMEAS(37)-XMEAS(41))，本文实验数据由5次TE正常工业过程数据组合而成，根据其同时间节点进行对应重构，每个变量都重构了4800个样本数据，将其分为训练集部分和测试集部分，训练集部分为前4000个样本，测试集部分为后800个样本。The Tennessee Eastman process is a simulation platform developed by Eastman Chemical Company to simulate the actual chemical reaction process. The data collected in this process consists of three parts: process control variables, process measurement variables, and component measurement variables. This paper uses 47 process variables (variable numbers are XMEAS(1)-XMEAS(36), XMV(1)-XMV( 11)) to predict 5 difficult-to-measure product variables (the variable numbers are XMEAS(37)-XMEAS(41)). The experimental data in this paper is composed of 5 TE normal industrial process data, and the corresponding re- For each variable, 4800 sample data are reconstructed and divided into training set and test set. The training set is the first 4000 samples, and the test set is the last 800 samples.
1. 本发明所采用的评价指标如下：1. the evaluation index that the present invention adopts is as follows:
均方根误差(RMSE)：Root Mean Square Error (RMSE):
式中：
R2系数：
式中：
表1.不同模型对产品质量的预测结果Table 1. Prediction results of different models on product quality
通过表1及说明书附图可以看出，传统的长短期记忆网络和双向长短期记忆网络对工业过程中质量变量等数据的预测结果并没有很精准，均方根误差、R2系数等评价指标来看预测误差还是较大的，长短期记忆网络只能简简单单的进行一个短批次的预测，具有很大的局限性；双向长短期记忆网络可以把上下文信息联系起来挖掘深层次的潜在信息，同时对大批次数据进行做预测的时候也可以很好的进行，但是工业过程数据太多无关紧要的变量对模型预测造成严重干扰，导致其预测效果不佳；本发明利用最大互信息系数来筛选其特征变量做降维重构输入数据，使模型的鲁棒性和泛化能力得到了加强。It can be seen from Table 1 and the accompanying drawings that traditional long-short-term memory networks and bidirectional long-short-term memory networks are not very accurate in predicting data such as quality variables in industrial processes, and evaluation indicators such as root mean square error andR2 coefficient are not very accurate. From the point of view, the prediction error is still relatively large. The long-term short-term memory network can only simply make a short-term prediction, which has great limitations; the two-way long-term short-term memory network can link contextual information to mine deep potential. information, and it can also be performed well when predicting large batches of data at the same time, but too many irrelevant variables in industrial process data cause serious interference to model prediction, resulting in poor prediction results; the present invention utilizes the maximum mutual information coefficient To filter its characteristic variables to reduce the dimensionality and reconstruct the input data, the robustness and generalization ability of the model have been strengthened.
尽管本发明的实施方案已公开如上，但并不仅仅限于说明书和实施方案中所列运用，它完全可以被适用于各种适合本发明的领域，对于熟悉本领域的人员而言，可容易地实现另外的修改，因此在不背离权利要求及等同范围所限定的一般概念下，本发明并不限于特定的细节和这里所示出与描述的图例。Although the embodiment of the present invention has been disclosed as above, it is not limited to the use listed in the specification and embodiment, it can be applied to various fields suitable for the present invention, and it can be easily understood by those skilled in the art Further modifications can be effected, so the invention is not limited to the specific details and examples shown and described herein without departing from the general concept defined by the claims and their equivalents.
Claims (2)
Translated fromChinesePriority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210537012.2ACN115438897A (en) | 2022-05-18 | 2022-05-18 | Industrial process product quality prediction method based on BLSTM neural network |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210537012.2ACN115438897A (en) | 2022-05-18 | 2022-05-18 | Industrial process product quality prediction method based on BLSTM neural network |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN115438897Atrue CN115438897A (en) | 2022-12-06 |
Family
ID=84240899
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202210537012.2AWithdrawnCN115438897A (en) | 2022-05-18 | 2022-05-18 | Industrial process product quality prediction method based on BLSTM neural network |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN115438897A (en) |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116052789A (en)* | 2023-03-29 | 2023-05-02 | 河北大景大搪化工设备有限公司 | Toluene chlorination parameter automatic optimization system based on deep learning |
| CN116312861A (en)* | 2023-05-09 | 2023-06-23 | 济南作为科技有限公司 | Denitration system gas concentration prediction method, device, equipment and storage medium |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111047012A (en)* | 2019-12-06 | 2020-04-21 | 重庆大学 | Air quality prediction method based on deep bidirectional long-short term memory network |
| CN111103420A (en)* | 2019-11-20 | 2020-05-05 | 华东理工大学 | A method for predicting the quality of phenolic resin products under uncertainty of raw materials |
| CN111985719A (en)* | 2020-08-27 | 2020-11-24 | 华中科技大学 | Power load prediction method based on improved long-term and short-term memory network |
| CN113569993A (en)* | 2021-08-27 | 2021-10-29 | 浙江工业大学 | Method for constructing quality prediction model in polymerization reaction process |
- 2022
- 2022-05-18CNCN202210537012.2Apatent/CN115438897A/ennot_activeWithdrawn
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111103420A (en)* | 2019-11-20 | 2020-05-05 | 华东理工大学 | A method for predicting the quality of phenolic resin products under uncertainty of raw materials |
| CN111047012A (en)* | 2019-12-06 | 2020-04-21 | 重庆大学 | Air quality prediction method based on deep bidirectional long-short term memory network |
| CN111985719A (en)* | 2020-08-27 | 2020-11-24 | 华中科技大学 | Power load prediction method based on improved long-term and short-term memory network |
| CN113569993A (en)* | 2021-08-27 | 2021-10-29 | 浙江工业大学 | Method for constructing quality prediction model in polymerization reaction process |
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116052789A (en)* | 2023-03-29 | 2023-05-02 | 河北大景大搪化工设备有限公司 | Toluene chlorination parameter automatic optimization system based on deep learning |
| CN116052789B (en)* | 2023-03-29 | 2023-09-15 | 河北大景大搪化工设备有限公司 | Toluene chlorination parameter automatic optimization system based on deep learning |
| CN116312861A (en)* | 2023-05-09 | 2023-06-23 | 济南作为科技有限公司 | Denitration system gas concentration prediction method, device, equipment and storage medium |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN114048600B (en) | Digital twin-driven multi-model fusion industrial system anomaly detection method | |
| CN117290800B (en) | Timing sequence anomaly detection method and system based on hypergraph attention network | |
| CN111813084B (en) | Mechanical equipment fault diagnosis method based on deep learning | |
| CN106779069A (en) | A kind of abnormal electricity consumption detection method based on neutral net | |
| CN113792754B (en) | Converter transformer DGA online monitoring data processing method for firstly removing abnormal state and then repairing | |
| CN114519923B (en) | Intelligent diagnosis and early warning method and system for power plant | |
| CN114282443A (en) | Remaining service life prediction method based on MLP-LSTM supervised joint model | |
| CN115809405A (en) | Fan main shaft gear box temperature anomaly detection method based on multi-feature fusion | |
| CN119149923A (en) | Digital twin method, system, computing device and storage medium based on industrial automation | |
| CN112116198A (en) | Data-driven process industrial state perception network key node screening method | |
| CN114943328A (en) | SARIMA-GRU time sequence prediction model based on BP neural network nonlinear combination | |
| CN115185937A (en) | SA-GAN architecture-based time sequence anomaly detection method | |
| CN115438897A (en) | Industrial process product quality prediction method based on BLSTM neural network | |
| CN107295537A (en) | A kind of method and system for wireless sensor network reliability of testing and assessing | |
| CN114548701B (en) | Process early warning method and system for coupling structure analysis and estimation of all measurement points | |
| Bai et al. | Towards trustworthy remaining useful life prediction through multi-source information fusion and a novel LSTM-DAU model | |
| CN117034169A (en) | Power grid main transformer equipment abnormal state prediction method based on time sequence causality network | |
| CN119089342A (en) | A time series anomaly detection method based on reconstruction and prediction | |
| CN115081741A (en) | A method for intelligent prediction of natural gas metrology verification based on neural network | |
| CN113988210A (en) | Distorted data restoration method, device and storage medium for structural monitoring sensor network | |
| CN117407675A (en) | Lightning arrester leakage current prediction method based on multi-variable reconstruction combined dynamic weight | |
| CN115688865A (en) | Industrial soft measurement method for long and short term memory network for flue gas of desulfurization process | |
| CN114818493A (en) | A method for quantitative evaluation of the integrity of tunnel rock mass | |
| CN118569081A (en) | Dynamic evolution and credibility assessment method for equipment digital twin data drive | |
| CN112801426A (en) | Industrial process fault fusion prediction method based on correlation parameter mining |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| WW01 | Invention patent application withdrawn after publication | ||
| WW01 | Invention patent application withdrawn after publication | Application publication date:20221206 |