CN111241755A - Power load prediction method - Google Patents
Power load prediction methodDownload PDFInfo
- Publication number
- CN111241755A CN111241755ACN202010113676.7ACN202010113676ACN111241755ACN 111241755 ACN111241755 ACN 111241755ACN 202010113676 ACN202010113676 ACN 202010113676ACN 111241755 ACN111241755 ACN 111241755A
- Authority
- CN
- China
- Prior art keywords
- data
- model
- output
- lstm
- time
- 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.)
- Pending
Links
Images
Classifications
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/044—Recurrent networks, e.g. Hopfield networks
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
- 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
- G06Q10/04—Forecasting or optimisation specially adapted for administrative or management purposes, e.g. linear programming or "cutting stock problem"
- 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
- G06Q50/06—Energy or water supply
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y04—INFORMATION OR COMMUNICATION TECHNOLOGIES HAVING AN IMPACT ON OTHER TECHNOLOGY AREAS
- Y04S—SYSTEMS INTEGRATING TECHNOLOGIES RELATED TO POWER NETWORK OPERATION, COMMUNICATION OR INFORMATION TECHNOLOGIES FOR IMPROVING THE ELECTRICAL POWER GENERATION, TRANSMISSION, DISTRIBUTION, MANAGEMENT OR USAGE, i.e. SMART GRIDS
- Y04S10/00—Systems supporting electrical power generation, transmission or distribution
- Y04S10/50—Systems or methods supporting the power network operation or management, involving a certain degree of interaction with the load-side end user applications
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Business, Economics & Management (AREA)
- Health & Medical Sciences (AREA)
- General Physics & Mathematics (AREA)
- General Health & Medical Sciences (AREA)
- Economics (AREA)
- Software Systems (AREA)
- Mathematical Physics (AREA)
- Data Mining & Analysis (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- General Engineering & Computer Science (AREA)
- Computational Linguistics (AREA)
- Human Resources & Organizations (AREA)
- Biophysics (AREA)
- Biomedical Technology (AREA)
- Artificial Intelligence (AREA)
- Life Sciences & Earth Sciences (AREA)
- Evolutionary Computation (AREA)
- Strategic Management (AREA)
- General Business, Economics & Management (AREA)
- Marketing (AREA)
- Tourism & Hospitality (AREA)
- Primary Health Care (AREA)
- Water Supply & Treatment (AREA)
- Public Health (AREA)
- Development Economics (AREA)
- Game Theory and Decision Science (AREA)
- Entrepreneurship & Innovation (AREA)
- Operations Research (AREA)
- Quality & Reliability (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
- Supply And Distribution Of Alternating Current (AREA)
Abstract
Description
Translated fromChinese技术领域technical field
本发明涉及电力信息技术领域，具体涉及一种基于长短期记忆神经网络与外部辅助信息的电力负荷预测方法。The invention relates to the technical field of power information, in particular to a power load prediction method based on a long short-term memory neural network and external auxiliary information.
背景技术Background technique
电力能源是现代社会的支柱性能源，准确有效的对电网负荷进行预测，对于电网的安全平稳运行、电力生产的经济高效至关重要，因此负荷预测一直以来是电力信息领域的一个研究热点。对于一个用能对象而言，其能耗受到自身特性、外部环境和时间周期等多个因素的影响，导致负荷数据的外在表现为随机性很大，难以基于物理机理进行有效分析、预测，线性回归等传统模型在复杂用能对象的实际负荷预测中无法满足要求。Electric power is the pillar energy of modern society. The accurate and effective prediction of power grid load is very important for the safe and stable operation of the power grid and the economical and efficient power production. Therefore, load forecasting has always been a research hotspot in the field of power information. For an energy-consuming object, its energy consumption is affected by many factors such as its own characteristics, external environment and time period, resulting in the external performance of load data being very random, and it is difficult to effectively analyze and predict based on the physical mechanism. Traditional models such as linear regression cannot meet the requirements in the actual load forecast of complex energy-consuming objects.
随着智能电网的建设、投入和信息技术的蓬勃发展，一方面用能对象的能耗数据和外部环境数据可以被实时或定期记录，产生了海量的电能负荷历史数据；另一方面基于数据的信息处理方法不断涌现，在刻画对象的潜在特性和随机性方面超越了传统的机理分析方法。目前，常用的电力负荷预测方法有人工神经网络(Artificial Neural Networks,ANN),支持向量机(Support Vector Machine,SVM)和自回归滑动平均模型(Autoregressive Integrated Moving Average Model,ARIMA)等方法。With the construction and investment of smart grid and the vigorous development of information technology, on the one hand, the energy consumption data of energy-consuming objects and external environmental data can be recorded in real time or on a regular basis, generating massive historical data of electric energy load; Information processing methods continue to emerge, surpassing traditional mechanistic analysis methods in characterizing the underlying properties and randomness of objects. At present, the commonly used power load forecasting methods include Artificial Neural Networks (ANN), Support Vector Machine (SVM) and Autoregressive Integrated Moving Average Model (ARIMA) and other methods.
传统技术存在以下技术问题：The traditional technology has the following technical problems:
但是这些方法在实际应用中，通常无法对能耗数据本身的周期性和外部随机性进行兼顾，因而最终的预测准确率有限。However, in practical applications of these methods, the periodicity and external randomness of the energy consumption data themselves cannot be taken into account, so the final prediction accuracy is limited.
发明内容SUMMARY OF THE INVENTION
本发明要解决的技术问题是提供一种基于长短期记忆神经网络与外部辅助信息的电力负荷预测方法，该方法能够基于历史能耗数据和外部环境数据，有效地分析用能对象的自身能耗特性以及同外部因素的耦合关系，充分地挖掘电力能耗的周期性特点，在模型训练完成后，能够基于实际历史数据对未来时刻的电力负荷进行准确预测。该发明可以使用于电力系统分析、电网预测调度等电力信息领域的问题。The technical problem to be solved by the present invention is to provide a power load forecasting method based on long short-term memory neural network and external auxiliary information, which can effectively analyze the energy consumption of energy-consuming objects based on historical energy consumption data and external environment data The characteristics and coupling relationship with external factors can fully exploit the periodic characteristics of power consumption. After the model training is completed, the power load in the future can be accurately predicted based on actual historical data. The invention can be used for problems in the field of power information such as power system analysis, power grid forecasting and dispatching.
为了解决上述技术问题，本发明提供一种基于长短期记忆神经网络与外部辅助信息的电力负荷预测方法，具体步骤如下：In order to solve the above technical problems, the present invention provides a power load prediction method based on a long short-term memory neural network and external auxiliary information. The specific steps are as follows:
第一步，历史数据汇总The first step is to summarize historical data
根据所选定的用能对象，对其电力能耗数据进行采集汇总，并记录下对应时刻的外部环境数据；According to the selected energy-consuming object, collect and summarize its power consumption data, and record the external environmental data at the corresponding time;
第二步，数据预处理，获取时序数据池The second step, data preprocessing, to obtain the time series data pool
对汇总的历史数据进行预处理，剔除异常值，之后进行归一化处理，将得到的数据作为模型训练所需的时序数据池；The aggregated historical data is preprocessed, outliers are removed, and then normalized, and the obtained data is used as the time series data pool required for model training;
第三步，LSTM网络建模与模型训练The third step, LSTM network modeling and model training
在时序数据池中，以长度为n的历史时序数据(xt-n+1,...,xt)做为模型输入，来对未来连续m个时刻的能耗数据(yt+1,...,yt+m)作为模型输出，构建LSTM网络模型，并采用时序数据池中的数据对模型进行训练。In the time series data pool, the historical time series data (xt-n+1 ,...,xt ) of length n is used as the model input to analyze the energy consumption data (yt+1 ,...,yt+m ) as the model output, build the LSTM network model, and use the data in the time series data pool to train the model.
第四步，根据实际历史数据和训练完成的LSTM网络模型，对未来时刻的电力负荷进行预测。The fourth step is to predict the power load in the future according to the actual historical data and the trained LSTM network model.
在其中一个实施例中，在第一步中，所需记录的外部环境数据包括温度、湿度、风力、降水等可获取到的气象数据和该时刻所对应的节假日信息。In one embodiment, in the first step, the external environmental data to be recorded includes available meteorological data such as temperature, humidity, wind power, precipitation, etc., and holiday information corresponding to the moment.
在其中一个实施例中，在第二步中，对异常的能耗数据值进行剔除，具体地：In one of the embodiments, in the second step, the abnormal energy consumption data values are eliminated, specifically:
计算所有获取到的历史数据的平均值与标准差：Calculate the mean and standard deviation of all acquired historical data:
对(μ-3σ，μ+3σ)区间外的能耗数据进行剔除。The energy consumption data outside the interval of (μ-3σ, μ+3σ) is eliminated.
在其中一个实施例中，在第二步中，对小范围的缺失数据进行线性填补，具体地：In one of the embodiments, in the second step, a small range of missing data is linearly filled, specifically:
其中m-n应当小于3，不属于该情况的对相应的缺失时序数据条进行整体剔除。Among them, m-n should be less than 3, and if it does not belong to this situation, the corresponding missing time series data bars will be eliminated as a whole.
在其中一个实施例中，在第二步中，为消除不同量纲的影响，对清洗、填补后的能耗、气象数据进行归一化，具体地：In one embodiment, in the second step, in order to eliminate the influence of different dimensions, the energy consumption and weather data after cleaning and filling are normalized, specifically:
在其中一个实施例中，在第二步中，对节假日信息进行二值化处理，具体地，如果当日为工作日，则x＝1，否则x＝0。In one of the embodiments, in the second step, binarization processing is performed on the holiday information, specifically, if the current day is a working day, x=1, otherwise x=0.
在其中一个实施例中，在第三步中，所构建的LSTM网络的输入信息包括：前24小时的电能负荷，预测时刻的温度、湿度、气压、风向和节假日信息，共计29组数据。In one embodiment, in the third step, the input information of the constructed LSTM network includes: the electric energy load of the previous 24 hours, the temperature, humidity, air pressure, wind direction and holiday information at the predicted time, a total of 29 sets of data.
在其中一个实施例中，在第三步中，所构建的LSTM网络共包含两个隐层，第一个隐层包括20个LSTM单元，第二个隐层包括10个LSTM单元。In one of the embodiments, in the third step, the constructed LSTM network includes a total of two hidden layers, the first hidden layer includes 20 LSTM units, and the second hidden layer includes 10 LSTM units.
在其中一个实施例中，在第三步中，所构建模型中的每个LSTM单元中，共包含单元状态和隐状态两个状态和遗忘门、输入门和输出门三个门限，具体地：In one of the embodiments, in the third step, each LSTM unit in the constructed model includes two states, the unit state and the hidden state, and three thresholds: the forget gate, the input gate and the output gate, specifically:
单元状态表示当前单元自身的属性，在t时刻记为Ct；The unit state represents the property of the current unit itself, which is denoted as C t at timet ;
隐状态表示当前神经元对外输出的属性，在t时刻记为ht；The hidden state represents the attribute of the current neuron's external output, which is recorded as h t at timet ;
遗忘门根据当前时刻的输出和前一时刻的隐状态，来对前一时刻的单元状态进行选择，The forget gate selects the unit state of the previous moment according to the output of the current moment and the hidden state of the previous moment,
ft＝σ(Wf·[ht-1,xt]+bf)ft =σ(Wf ·[ht-1 ,xt ]+bf )
其中σ(·)为sigmoid函数，Wf和bf分别为遗忘门的权重和偏执项；where σ( ) is the sigmoid function, and Wf and bf are the weights and paranoid terms of the forget gate, respectively;
输入门根据输入和前一时刻的隐状态来产生候选单元状态，并根据门限情况进行筛选，The input gate generates candidate unit states according to the input and the hidden state of the previous moment, and filters according to the threshold conditions,
it＝σ(Wi·[ht-1,xt]+bi)it =σ(Wi ·[ht-1 ,xt ]+bi )
类似的，其中tanh(·)为双曲正切函数，WC和bC为产生候选单元状态的权重和偏执项，Wi和bi分别为输入门限的权重和偏执项；Similarly, where tanh( ) is the hyperbolic tangent function,WC and bC are the weights and bias terms for generating candidate cell states, and Wi andbi are the weights and bias terms for the input threshold, respectively;
之后对单元状态进行更新；Then update the unit status;
输出门根据当前时刻的单元状态确定隐状态的值，并根据当前时刻输入和前一时刻隐状态进行筛选，The output gate determines the value of the hidden state according to the unit state at the current moment, and filters it according to the input at the current moment and the hidden state at the previous moment,
ot＝σ(Wo·[ht-1,xt]+bo)ot =σ(Wo ·[ht-1 ,xt ]+bo )
ht＝ot·tanh(Ct)ht =ot ·tanh(Ct )
其中，Wo和bo分别为输出门限的权重和偏执项，最终产生隐状态ht作为LSTM单元的整体输出传递到下一层和下一时刻。Among them, Wo and bo are the weight and paranoid term of the output threshold, respectively, and finally the hidden state ht is generated as the overall output of the LSTM unit and passed to the next layer and the next moment.
在其中一个实施例中，在第三步中，所构建的LSTM网络输出层为一个线性层，具体地，In one of the embodiments, in the third step, the constructed LSTM network output layer is a linear layer, specifically,
ypred＝Wh·ht+bhypred =Wh ·ht +bh
其中，Wh和bh分别为输出层的权重和偏执项。Among them, Wh and bh are the weights and bias terms of the output layer, respectively.
在其中一个实施例中，在第三步中，所构建的LSTM网络模型的评价指标为均方根误差，具体地，In one of the embodiments, in the third step, the evaluation index of the constructed LSTM network model is the root mean square error, specifically,
在其中一个实施例中，在第三步中，对于所构建的LSTM网络模型，采用ADAM算法进行训练，学习率设为0.001，训练时长为1500epochs。In one embodiment, in the third step, ADAM algorithm is used for training the constructed LSTM network model, the learning rate is set to 0.001, and the training duration is 1500 epochs.
基于同样的发明构思，本申请还提供一种计算机设备，包括存储器、处理器及存储在存储器上并可在处理器上运行的计算机程序，所述处理器执行所述程序时实现任一项所述方法的步骤。Based on the same inventive concept, the present application also provides a computer device, including a memory, a processor, and a computer program stored in the memory and running on the processor, the processor implements any one of the above when executing the program. steps of the method described.
基于同样的发明构思，本申请还提供一种计算机可读存储介质，其上存储有计算机程序，该程序被处理器执行时实现任一项所述方法的步骤。Based on the same inventive concept, the present application also provides a computer-readable storage medium on which a computer program is stored, and when the program is executed by a processor, implements the steps of any one of the methods.
基于同样的发明构思，本申请还提供一种处理器，所述处理器用于运行程序，其中，所述程序运行时执行任一项所述的方法。Based on the same inventive concept, the present application also provides a processor for running a program, wherein the program executes any one of the methods when the program runs.
本发明的有益效果：Beneficial effects of the present invention:
本发明基于长短期记忆神经网络与外部辅助信息的电力负荷预测方法，将电能负荷预测抽象为一个时间序列分析问题，采用人工智能方法进行求解。首先对历史数据进行有效地预处理，通过清洗、线性填补和归一化，保证了数据的完整性和可利用性；采用LSTM网络作为能耗建模的基准模型，充分考虑了能耗信息的时间序列特性，避免了普通循环神经网络存在的梯度消失问题；同时分析了电能消耗相关的外部因素，进行联合建模；训练方法对于不同数据下的模型具有较好的鲁棒性。从实验验证情况来看，本发明在应用中能够有效提高用能对象负荷预测的准确性。The invention abstracts the power load forecasting as a time series analysis problem based on the long-short-term memory neural network and external auxiliary information, and adopts artificial intelligence method to solve it. First, the historical data is effectively preprocessed, and the integrity and availability of the data are ensured through cleaning, linear filling and normalization; the LSTM network is used as the benchmark model for energy consumption modeling, and the energy consumption information is fully considered. The time series feature avoids the vanishing gradient problem of ordinary recurrent neural networks; at the same time, the external factors related to power consumption are analyzed, and joint modeling is carried out; the training method has better robustness for models under different data. From the point of experimental verification, the present invention can effectively improve the accuracy of load prediction of energy-consuming objects in application.
附图说明Description of drawings
图1是本发明基于长短期记忆神经网络与外部辅助信息的电力负荷预测方法的流程图。FIG. 1 is a flow chart of the power load prediction method based on the long short-term memory neural network and external auxiliary information of the present invention.
图2是本发明基于长短期记忆神经网络与外部辅助信息的电力负荷预测方法中LSTM单元的结构图。FIG. 2 is a structural diagram of the LSTM unit in the power load prediction method based on the long short-term memory neural network and external auxiliary information of the present invention.
图3是本发明基于长短期记忆神经网络与外部辅助信息的电力负荷预测方法中1500epochs训练效果图。FIG. 3 is a diagram showing the effect of 1500 epochs training in the power load prediction method based on the long short-term memory neural network and external auxiliary information of the present invention.
图4是本发明基于长短期记忆神经网络与外部辅助信息的电力负荷预测方法中模型预测效果图。FIG. 4 is a model prediction effect diagram of the power load prediction method based on the long short-term memory neural network and external auxiliary information of the present invention.
具体实施方式Detailed ways
下面结合附图和具体实施例对本发明作进一步说明，以使本领域的技术人员可以更好地理解本发明并能予以实施，但所举实施例不作为对本发明的限定。The present invention will be further described below with reference to the accompanying drawings and specific embodiments, so that those skilled in the art can better understand the present invention and implement it, but the embodiments are not intended to limit the present invention.
本发明基于长短期记忆神经网络与外部辅助信息的电力负荷预测方法，如图1所示，按照如下详细步骤进行处理：The power load prediction method based on the long short-term memory neural network and external auxiliary information of the present invention, as shown in FIG. 1, is processed according to the following detailed steps:
历史数据采集汇总Summary of historical data collection
根据所选定的预测对象，对其每小时的历史能耗数据进行采集汇总，从图4中的实际负荷曲线可以看出，电能负荷除了存在一定的周期性外，还同时存在波动性和偶然性，这是由于受到外部因素影响。根据考察分析，外部气象因素与能耗关系较为密切，因此结合可获取的数据情况，选取温度、湿度、气压和风力共4中气象数据加入到能耗模型中；除此之外，节假日也对能耗影响至关重要，因此也应当考虑到能耗模型当中。采集汇总后的数据样表如表1所示。According to the selected forecast object, the hourly historical energy consumption data is collected and summarized. From the actual load curve in Figure 4, it can be seen that in addition to a certain periodicity, the electric energy load also has fluctuation and contingency at the same time. , which is due to external factors. According to the investigation and analysis, external meteorological factors are closely related to energy consumption. Therefore, combined with the available data, four meteorological data including temperature, humidity, air pressure and wind force are selected and added to the energy consumption model. The impact of energy consumption is critical and should therefore also be factored into the energy consumption model. A sample table of the collected data is shown in Table 1.
表1Table 1
数据预处理data preprocessing
对采集汇总后的数据进行预处理操作，便于后续模型处理。首先剔除异常负荷及其对应的数据条，具体地，计算所有获取到的历史能耗数据的平均值与方差：The collected and summarized data are preprocessed to facilitate subsequent model processing. First, the abnormal load and its corresponding data bar are removed. Specifically, the average value and variance of all the obtained historical energy consumption data are calculated:
对(μ-3σ，μ+3σ)区间外的能耗数据进行剔除。The energy consumption data outside the interval of (μ-3σ, μ+3σ) is eliminated.
其次，对小范围的缺失数据进行线性填补，具体地，Second, linear imputation is performed on a small range of missing data, specifically,
其中m-n应当小于3，不属于该情况的对相应的缺失时序数据条进行整体剔除。Among them, m-n should be less than 3, and if it does not belong to this situation, the corresponding missing time series data bars will be eliminated as a whole.
然后，为消除不同量纲带来的影响，对清洗、填补后的能耗、气象数据进行归一化，具体地：Then, in order to eliminate the influence of different dimensions, the energy consumption and meteorological data after cleaning and filling are normalized, specifically:
最后，对节假日信息进行二值化处理，具体地，如果当日为工作日，则x＝1，否则x＝0。Finally, binarize the holiday information, specifically, if the current day is a working day, then x=1, otherwise, x=0.
模型建立与训练Model building and training
对于负荷预测任务而言，需要根据历史数据来对目标时刻的负荷进行预测。在此，本发明不失一般性的采用前24小时的能耗数据和预测时刻的外部辅助信息来对负荷进行建模预测。具体地，模型输入包括前24小时的电能负荷，预测时刻的温度、湿度、气压、风向和节假日信息，共计29组数据；模型输出为预测时刻的负荷。For the load prediction task, it is necessary to predict the load at the target time based on historical data. Here, without loss of generality, the present invention uses the energy consumption data of the previous 24 hours and the external auxiliary information at the predicted time to model and predict the load. Specifically, the model input includes the electric energy load in the previous 24 hours, the temperature, humidity, air pressure, wind direction and holiday information at the predicted time, a total of 29 sets of data; the model output is the load at the predicted time.
对此，本发明基于PyTorch深度学习框架来建立LSTM网络模型，具体地，模型包含两个隐层和一个输出层。第一个隐层包含20个LSTM单元，第二个隐层包含10个LSTM单元。每个LSTM单元包含单元状态和隐状态两个状态和遗忘门、输入门和输出门三个门限，具体地：In this regard, the present invention establishes an LSTM network model based on the PyTorch deep learning framework. Specifically, the model includes two hidden layers and one output layer. The first hidden layer contains 20 LSTM units and the second hidden layer contains 10 LSTM units. Each LSTM unit contains two states: unit state and hidden state, and three thresholds: forget gate, input gate and output gate, specifically:
单元状态表示当前单元自身的属性，在t时刻记为Ct；The unit state represents the property of the current unit itself, which is denoted as C t at timet ;
隐状态表示当前神经元对外输出的属性，在t时刻记为ht；The hidden state represents the attribute of the current neuron's external output, which is recorded as h t at timet ;
遗忘门根据当前时刻的输出和前一时刻的隐状态，来对前一时刻的单元状态进行选择，The forget gate selects the unit state of the previous moment according to the output of the current moment and the hidden state of the previous moment,
ft＝σ(Wf·[ht-1,xt]+bf)ft =σ(Wf ·[ht-1 ,xt ]+bf )
其中σ(·)为sigmoid函数，Wf和bf分别为遗忘门的权重和偏执项；where σ( ) is the sigmoid function, and Wf and bf are the weights and paranoid terms of the forget gate, respectively;
输入门根据输入和前一时刻的隐状态来产生候选单元状态，并根据门限情况进行筛选，The input gate generates candidate unit states according to the input and the hidden state of the previous moment, and filters according to the threshold conditions,
it＝σ(Wi·[ht-1,xt]+bi)it =σ(Wi ·[ht-1 ,xt ]+bi )
类似的，其中tanh(·)为双曲正切函数，WC和bC为产生候选单元状态的权重和偏执项，Wi和bi分别为输入门限的权重和偏执项；Similarly, where tanh( ) is the hyperbolic tangent function,WC and bC are the weights and bias terms for generating candidate cell states, and Wi andbi are the weights and bias terms for the input threshold, respectively;
之后对单元状态进行跟新；Then update the unit state;
输出门根据当前时刻的单元状态确定隐状态的值，并根据当前时刻输入和前一时刻隐状态进行筛选，The output gate determines the value of the hidden state according to the unit state at the current moment, and filters it according to the input at the current moment and the hidden state at the previous moment,
ot＝σ(Wo·[ht-1,xt]+bo)ot =σ(Wo ·[ht-1 ,xt ]+bo )
ht＝ot·tanh(Ct)ht =ot ·tanh(Ct )
其中，Wo和bo分别为输出门限的权重和偏执项，最终产生隐状态ht作为LSTM单元的整体输出传递到下一层和下一时刻。Among them, Wo and bo are the weight and paranoid term of the output threshold, respectively, and finally the hidden state ht is generated as the overall output of the LSTM unit and passed to the next layer and the next moment.
模型的输出层为一个线性层，具体地，The output layer of the model is a linear layer, specifically,
ypred＝Wh·ht+bhypred =Wh ·ht +bh
其中，Wh和bh分别为输出层的权重和偏执项。Among them, Wh and bh are the weights and bias terms of the output layer, respectively.
模型构建完成后，设定模型损失函数为均方根误差，具体地，After the model is constructed, the model loss function is set as the root mean square error. Specifically,
基于所获取、处理的数据，采用ADAM算法对模型进行训练，学习速率为0.001，训练时长为1500epochs。模型训练效果如图3所示。Based on the acquired and processed data, the ADAM algorithm is used to train the model with a learning rate of 0.001 and a training duration of 1500 epochs. The model training effect is shown in Figure 3.
实际负荷预测Actual load forecast
模型训练完成后，基于实际的历史数据和预测时刻的外部辅助信息，对未来时间的负荷进行预测，用来辅助参考电网的分析、运维、调度等工作。由于训练数据是归一化后的，模型得到预测值后，需要再进行反变换得到最终的预测负荷，具体地。After the model training is completed, based on the actual historical data and external auxiliary information at the predicted time, the load in the future time is predicted, which is used to assist the analysis, operation and maintenance, and scheduling of the reference power grid. Since the training data is normalized, after the model obtains the predicted value, it needs to be inversely transformed to obtain the final predicted load, specifically.
为了验证本发明的实际效果，本发明采用2011年卢布尔雅那某处全年的真实负荷数据和气象数据来进行实验分析，其中节假日只考虑平日与周末两种情况。实验结果表明所构建的模型能够准确预测用能对象的负荷变化，对于电网的分析、预警可以起到较大的积极作用。In order to verify the actual effect of the present invention, the present invention adopts the real load data and meteorological data of a certain place in Ljubljana in 2011 for experimental analysis, and only considers two situations of weekdays and weekends for holidays. The experimental results show that the constructed model can accurately predict the load changes of energy-consuming objects, and can play a positive role in the analysis and early warning of the power grid.
具体地，采用前90％的数据作为训练集，用来指导模型训练，后10％数据作为验证集，用来验证模型预测效果。模型经过1500epochs的训练后，损失函数下降到较低的水平，此时结束训练，采用模型来预测验证集中的负荷，预测结果如图4所示，其中纵坐标为负荷，横坐标为时间，蓝色曲线为真实负荷，橙色曲线为预测负荷。可以看出，真实负荷具有很强的周期性和一定的随机性。而由基于历史数据和外部辅助数据的LSTM网络模型所得到的最终预测值，可以准确的拟合真实负荷的趋势，仅在每日的波峰和波谷存在少量误差。Specifically, the first 90% of the data is used as a training set to guide model training, and the last 10% of the data is used as a validation set to verify the model prediction effect. After the model has been trained for 1500 epochs, the loss function drops to a lower level. At this time, the training ends, and the model is used to predict the load in the validation set. The prediction result is shown in Figure 4, where the ordinate is the load, the abscissa is the time, and the blue The colored curve is the actual load, and the orange curve is the predicted load. It can be seen that the real load has strong periodicity and certain randomness. The final predicted value obtained by the LSTM network model based on historical data and external auxiliary data can accurately fit the trend of the real load, with only a small amount of error in the daily peaks and troughs.
本发明仅利用用能对象的历史能耗数据和较为轻易获取的外部辅助信息来对未来时刻的负荷进行有效预测，不涉及具体的用能对象机理研究，无需相应的先验知识，而是通过LSTM网络构建数据模型的方式来完成负荷预测任务。基于PyTorch框架开发，整体方法流程简便可靠。同时，该模型在应用时，可以根据实际情况，灵活的调整数据的时间粒度和预测不长，相对于传统的方法更加灵活，有利于实际负荷预测任务的准确性和实时性。The invention only uses the historical energy consumption data of the energy-consuming objects and the external auxiliary information that can be easily obtained to effectively predict the load at the future time, does not involve the research of the specific energy-consuming object mechanism, and does not require corresponding prior knowledge. The way the LSTM network builds the data model to complete the load forecasting task. Developed based on the PyTorch framework, the overall method process is simple and reliable. At the same time, when the model is applied, the time granularity of the data can be flexibly adjusted according to the actual situation and the prediction is not long, which is more flexible than the traditional method, which is beneficial to the accuracy and real-time performance of the actual load prediction task.
以上所述实施例仅是为充分说明本发明而所举的较佳的实施例，本发明的保护范围不限于此。本技术领域的技术人员在本发明基础上所作的等同替代或变换，均在本发明的保护范围之内。本发明的保护范围以权利要求书为准。The above-mentioned embodiments are only preferred embodiments for fully illustrating the present invention, and the protection scope of the present invention is not limited thereto. Equivalent substitutions or transformations made by those skilled in the art on the basis of the present invention are all within the protection scope of the present invention. The protection scope of the present invention is subject to the claims.
Claims (10)
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010113676.7ACN111241755A (en) | 2020-02-24 | 2020-02-24 | Power load prediction method |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010113676.7ACN111241755A (en) | 2020-02-24 | 2020-02-24 | Power load prediction method |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN111241755Atrue CN111241755A (en) | 2020-06-05 |
Family
ID=70873344
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202010113676.7APendingCN111241755A (en) | 2020-02-24 | 2020-02-24 | Power load prediction method |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN111241755A (en) |
Cited By (12)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112733457A (en)* | 2021-01-18 | 2021-04-30 | 武汉大学 | Load electricity utilization characteristic modeling method for improving double-layer long-short term memory network |
| CN112990556A (en)* | 2021-02-24 | 2021-06-18 | 江苏大学 | User power consumption prediction method based on Prophet-LSTM model |
| CN113033766A (en)* | 2020-10-23 | 2021-06-25 | 广州博纳信息技术有限公司 | Energy efficiency trend prediction method based on LSTM neural network algorithm model |
| CN113077105A (en)* | 2021-04-16 | 2021-07-06 | 国网安徽省电力有限公司 | Long-holiday load prediction method and device |
| CN113239614A (en)* | 2021-04-22 | 2021-08-10 | 西北工业大学 | Atmospheric turbulence phase space-time prediction algorithm |
| CN113792490A (en)* | 2021-09-16 | 2021-12-14 | 国网江苏省电力有限公司营销服务中心 | A modeling method for energy consumption of cement mixer based on support vector regression machine |
| CN113962456A (en)* | 2021-10-19 | 2022-01-21 | 江苏方天电力技术有限公司 | Medium-and-long-term load prediction method considering industry relevance |
| CN114326391A (en)* | 2021-12-13 | 2022-04-12 | 哈尔滨工程大学 | Building energy consumption prediction method |
| CN114611826A (en)* | 2022-03-25 | 2022-06-10 | 华润电力(广东)销售有限公司 | Power supply and demand prediction method and related equipment |
| CN114623569A (en)* | 2021-11-04 | 2022-06-14 | 国网浙江省电力有限公司湖州供电公司 | Cluster air conditioner load differentiation regulation and control method based on deep reinforcement learning |
| CN114912682A (en)* | 2022-05-13 | 2022-08-16 | 广东电网有限责任公司 | Load prediction method based on information fusion and related device |
| CN115471362A (en)* | 2022-09-26 | 2022-12-13 | 东南大学 | Comprehensive energy source-load prediction method for depth feature-guided two-stage transfer learning |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109034500A (en)* | 2018-09-04 | 2018-12-18 | 湘潭大学 | A kind of mid-term electric load forecasting method of multiple timings collaboration |
| CN109659933A (en)* | 2018-12-20 | 2019-04-19 | 浙江工业大学 | A kind of prediction technique of power quality containing distributed power distribution network based on deep learning model |
| CN110059844A (en)* | 2019-02-01 | 2019-07-26 | 东华大学 | Energy storage device control method based on set empirical mode decomposition and LSTM |
| CN110738344A (en)* | 2018-07-20 | 2020-01-31 | 中国农业大学 | Distributed reactive power optimization method and device for power system load forecasting |
- 2020
- 2020-02-24CNCN202010113676.7Apatent/CN111241755A/enactivePending
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110738344A (en)* | 2018-07-20 | 2020-01-31 | 中国农业大学 | Distributed reactive power optimization method and device for power system load forecasting |
| CN109034500A (en)* | 2018-09-04 | 2018-12-18 | 湘潭大学 | A kind of mid-term electric load forecasting method of multiple timings collaboration |
| CN109659933A (en)* | 2018-12-20 | 2019-04-19 | 浙江工业大学 | A kind of prediction technique of power quality containing distributed power distribution network based on deep learning model |
| CN110059844A (en)* | 2019-02-01 | 2019-07-26 | 东华大学 | Energy storage device control method based on set empirical mode decomposition and LSTM |
Cited By (16)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113033766A (en)* | 2020-10-23 | 2021-06-25 | 广州博纳信息技术有限公司 | Energy efficiency trend prediction method based on LSTM neural network algorithm model |
| CN112733457B (en)* | 2021-01-18 | 2022-03-15 | 武汉大学 | An Improved Modeling Method for Load Electricity Characteristics of Two-layer Long Short-Term Memory Networks |
| CN112733457A (en)* | 2021-01-18 | 2021-04-30 | 武汉大学 | Load electricity utilization characteristic modeling method for improving double-layer long-short term memory network |
| CN112990556A (en)* | 2021-02-24 | 2021-06-18 | 江苏大学 | User power consumption prediction method based on Prophet-LSTM model |
| CN113077105A (en)* | 2021-04-16 | 2021-07-06 | 国网安徽省电力有限公司 | Long-holiday load prediction method and device |
| CN113077105B (en)* | 2021-04-16 | 2023-11-24 | 国网安徽省电力有限公司 | A long holiday load forecasting method and device |
| CN113239614A (en)* | 2021-04-22 | 2021-08-10 | 西北工业大学 | Atmospheric turbulence phase space-time prediction algorithm |
| CN113792490A (en)* | 2021-09-16 | 2021-12-14 | 国网江苏省电力有限公司营销服务中心 | A modeling method for energy consumption of cement mixer based on support vector regression machine |
| CN113962456A (en)* | 2021-10-19 | 2022-01-21 | 江苏方天电力技术有限公司 | Medium-and-long-term load prediction method considering industry relevance |
| CN113962456B (en)* | 2021-10-19 | 2024-11-22 | 江苏方天电力技术有限公司 | A medium- and long-term load forecasting method taking into account industry correlation |
| CN114623569B (en)* | 2021-11-04 | 2023-09-29 | 国网浙江省电力有限公司湖州供电公司 | Cluster air conditioner load differential regulation and control method based on deep reinforcement learning |
| CN114623569A (en)* | 2021-11-04 | 2022-06-14 | 国网浙江省电力有限公司湖州供电公司 | Cluster air conditioner load differentiation regulation and control method based on deep reinforcement learning |
| CN114326391A (en)* | 2021-12-13 | 2022-04-12 | 哈尔滨工程大学 | Building energy consumption prediction method |
| CN114611826A (en)* | 2022-03-25 | 2022-06-10 | 华润电力(广东)销售有限公司 | Power supply and demand prediction method and related equipment |
| CN114912682A (en)* | 2022-05-13 | 2022-08-16 | 广东电网有限责任公司 | Load prediction method based on information fusion and related device |
| CN115471362A (en)* | 2022-09-26 | 2022-12-13 | 东南大学 | Comprehensive energy source-load prediction method for depth feature-guided two-stage transfer learning |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN111241755A (en) | Power load prediction method | |
| Massaoudi et al. | A novel stacked generalization ensemble-based hybrid LGBM-XGB-MLP model for Short-Term Load Forecasting | |
| CN110414788B (en) | Electric energy quality prediction method based on similar days and improved LSTM | |
| CN110648014B (en) | A regional wind power forecasting method and system based on spatiotemporal quantile regression | |
| CN112990556A (en) | User power consumption prediction method based on Prophet-LSTM model | |
| CN113449919B (en) | Power consumption prediction method and system based on feature and trend perception | |
| CN114282443B (en) | Remaining service life prediction method based on MLP-LSTM supervised joint model | |
| CN111915092B (en) | Ultra-short-term wind power forecasting method based on long-short-term memory neural network | |
| CN110571792A (en) | A method and system for analyzing and evaluating the operating state of a power grid control system | |
| CN107590567A (en) | Recurrent neural network short-term load prediction method based on information entropy clustering and attention mechanism | |
| CN105654207A (en) | Wind power prediction method based on wind speed information and wind direction information | |
| CN114218872A (en) | Method for predicting remaining service life based on DBN-LSTM semi-supervised joint model | |
| CN112396234A (en) | User side load probability prediction method based on time domain convolutional neural network | |
| CN112803398A (en) | Load prediction method and system based on empirical mode decomposition and deep neural network | |
| CN111461463A (en) | A short-term load forecasting method, system and equipment based on TCN-BP | |
| CN106971238A (en) | The Short-Term Load Forecasting Method of Elman neutral nets is obscured based on T S | |
| CN116739118A (en) | A power load forecasting method based on LSTM-XGBoost to implement error correction mechanism | |
| CN112766535B (en) | Building load prediction method and system considering load curve characteristics | |
| CN118693887B (en) | An optimization method for controlling wind power prediction error during black start of overcapacity energy storage | |
| CN112988538B (en) | Artificial intelligence development platform monitoring alarm data prediction method, device and medium | |
| CN117851802A (en) | Water quality prediction method and device and computer readable storage medium | |
| CN117175584A (en) | Controllable load prediction method, device and equipment based on power data | |
| Manoj et al. | FWS-DL: forecasting wind speed based on deep learning algorithms | |
| Boaisha et al. | Forecasting model based on fuzzy time series approach | |
| Zhang et al. | Short-term electrical load forecasting based on Attention-GRU networks |
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 | ||
| RJ01 | Rejection of invention patent application after publication | ||
| RJ01 | Rejection of invention patent application after publication | Application publication date:20200605 |