CN115545503B - Power load medium-short term prediction method and system based on parallel time sequence convolutional neural network - Google Patents
Power load medium-short term prediction method and system based on parallel time sequence convolutional neural network Download PDFInfo
- Publication number
- CN115545503B CN115545503B CN202211259893.2A CN202211259893A CN115545503B CN 115545503 B CN115545503 B CN 115545503B CN 202211259893 A CN202211259893 A CN 202211259893A CN 115545503 B CN115545503 B CN 115545503B
- Authority
- CN
- China
- Prior art keywords
- characteristic data
- neural network
- convolutional neural
- time sequence
- parallel 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.)
- Active
Links
- 238000013527 convolutional neural network Methods 0.000 title claims abstract description 83
- 238000000034 method Methods 0.000 title claims abstract description 39
- 230000000737 periodic effect Effects 0.000 claims abstract description 48
- 238000012549 training Methods 0.000 claims abstract description 36
- 238000012545 processing Methods 0.000 claims abstract description 15
- 239000013598 vector Substances 0.000 claims description 30
- 238000013528 artificial neural network Methods 0.000 claims description 21
- 230000004913 activation Effects 0.000 claims description 15
- 239000011159 matrix material Substances 0.000 claims description 14
- 238000011156 evaluation Methods 0.000 claims description 9
- 238000010606 normalization Methods 0.000 claims description 8
- 230000008569 process Effects 0.000 claims description 6
- 230000037211 monthly cycles Effects 0.000 claims 1
- 230000003442 weekly effect Effects 0.000 claims 1
- 230000006870 function Effects 0.000 description 13
- 230000000694 effects Effects 0.000 description 5
- 238000004364 calculation method Methods 0.000 description 3
- 238000002474 experimental method Methods 0.000 description 3
- 230000001364 causal effect Effects 0.000 description 2
- 230000002354 daily effect Effects 0.000 description 2
- 238000010586 diagram Methods 0.000 description 2
- 230000004927 fusion Effects 0.000 description 2
- ORILYTVJVMAKLC-UHFFFAOYSA-N Adamantane Natural products C1C(C2)CC3CC1CC2C3 ORILYTVJVMAKLC-UHFFFAOYSA-N 0.000 description 1
- 238000012300 Sequence Analysis Methods 0.000 description 1
- 230000001133 acceleration Effects 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000008859 change Effects 0.000 description 1
- 238000012512 characterization method Methods 0.000 description 1
- 238000010835 comparative analysis Methods 0.000 description 1
- 238000010276 construction Methods 0.000 description 1
- 238000013135 deep learning Methods 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 230000010339 dilation Effects 0.000 description 1
- 230000007613 environmental effect Effects 0.000 description 1
- 230000003203 everyday effect Effects 0.000 description 1
- 238000009776 industrial production Methods 0.000 description 1
- 238000012417 linear regression Methods 0.000 description 1
- 230000007774 longterm Effects 0.000 description 1
- 238000010801 machine learning Methods 0.000 description 1
- 230000007246 mechanism Effects 0.000 description 1
- 238000010248 power generation Methods 0.000 description 1
- 238000002360 preparation method Methods 0.000 description 1
- 230000000306 recurrent effect Effects 0.000 description 1
- 238000005070 sampling Methods 0.000 description 1
- 230000006403 short-term memory Effects 0.000 description 1
- 230000002123 temporal effect Effects 0.000 description 1
- 238000010977 unit operation Methods 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR 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
- G06Q10/0631—Resource planning, allocation, distributing or scheduling for enterprises or organisations
- G06Q10/06312—Adjustment or analysis of established resource schedule, e.g. resource or task levelling, or dynamic rescheduling
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR 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; CALCULATING OR 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—Systems or methods specially adapted for specific business sectors, e.g. utilities or tourism
- G06Q50/06—Electricity, gas or water supply
-
- H—ELECTRICITY
- H02—GENERATION; CONVERSION OR DISTRIBUTION OF ELECTRIC POWER
- H02J—CIRCUIT ARRANGEMENTS OR SYSTEMS FOR SUPPLYING OR DISTRIBUTING ELECTRIC POWER; SYSTEMS FOR STORING ELECTRIC ENERGY
- H02J3/00—Circuit arrangements for ac mains or ac distribution networks
- H02J3/003—Load forecast, e.g. methods or systems for forecasting future load demand
-
- 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)
- Business, Economics & Management (AREA)
- Human Resources & Organizations (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Economics (AREA)
- General Physics & Mathematics (AREA)
- Health & Medical Sciences (AREA)
- Strategic Management (AREA)
- Entrepreneurship & Innovation (AREA)
- Marketing (AREA)
- General Business, Economics & Management (AREA)
- Tourism & Hospitality (AREA)
- General Health & Medical Sciences (AREA)
- Life Sciences & Earth Sciences (AREA)
- Development Economics (AREA)
- Operations Research (AREA)
- Game Theory and Decision Science (AREA)
- Educational Administration (AREA)
- Artificial Intelligence (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- Data Mining & Analysis (AREA)
- Evolutionary Computation (AREA)
- Quality & Reliability (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- General Engineering & Computer Science (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Public Health (AREA)
- Water Supply & Treatment (AREA)
- Primary Health Care (AREA)
- Power Engineering (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The invention relates to a power load medium-short term prediction method and system based on a parallel time sequence convolutional neural network, and belongs to the field of power system scheduling control. The method comprises the following steps: processing the training data and extracting periodic characteristic data; the periodic characteristic data comprise hour period characteristic data, day period characteristic data, week period characteristic data and month period characteristic data; creating a parallel time sequence convolutional neural network model according to the periodic characteristic data, wherein the parallel time sequence convolutional neural network model comprises four time convolutional neural network TCN models; and carrying out medium-short term prediction on the power load by using the parallel time sequence convolutional neural network model, and outputting a prediction result. The method can effectively reflect the multicycle characteristics of the load data, thereby improving the accuracy of short-term prediction in the power load.
Description
Technical Field
The invention belongs to the field of power system scheduling control, and particularly relates to a power load medium-short term prediction method and system based on a parallel time sequence convolutional neural network.
Background
The accurate prediction of the power load is a precondition for a power grid dispatching mechanism to make a power generation plan and manage power supply operation, and is also a key basis for realizing accurate metering and ensuring metering equipment to operate in an optimal state. The power load prediction can be classified into long-term prediction (in years), medium-term prediction (in months), short-term prediction (in days) and ultra-short-term prediction (in hours, minutes) according to the time scale, and the prediction targets include the system load and the bus load. The accurate medium-short term prediction can reasonably determine the unit operation mode and the equipment overhaul plan, and the daily scheduling plan and the Zhou Diaodu plan are arranged, so that the method has important significance in realizing power balance between source and load and guaranteeing the safe and economic operation of the system. With the acceleration of the construction pace of a novel power system, the access of multiple types of loads makes the change of power of the power load more complex, makes full use of a large amount of historical data, and builds a medium-short term prediction model capable of reflecting the multi-period characteristics of the load by combining deep learning and other methods, so that the prediction precision of the load is improved, and the method becomes a key problem of medium-short term prediction of the power load under the background of the novel power system.
The method for short-term prediction in the power load is mainly three types, one type is a traditional time sequence analysis method, such as a linear regression method and an ARI MA model, the method has the advantages of less data quantity, simpler model and suitability for modeling a stable time sequence. The second category is machine learning methods, which have strong data modeling capabilities, but require extensive feature engineering. The third class is neural network methods, such as BP neural network methods, deep neural network methods, including Long Short-Term Memory (LSTM), convolutional neural networks (Convolutional Neural Networks, CNN), time convolutional neural networks (Temporal Convolutional Network, TCN), which have strong modeling capability, better model effect than the former two classes of methods, but require massive training data for model training. Because the power load data has the characteristics of an hour period, a day period, a week period and a season period, the conventional neural network training model cannot reflect different period characteristics of the load, and the accuracy of a prediction result is also affected to a certain extent.
Disclosure of Invention
The invention aims to overcome the defects and shortcomings of the prior art, and provides a power load medium-short term prediction method and system based on a parallel time sequence convolutional neural network, which effectively reflect the multicycle characteristic of load data, thereby improving the accuracy of power load medium-short term prediction.
According to one aspect of the present invention, there is provided a method for mid-short term prediction of power load based on a parallel time series convolutional neural network, the method comprising the steps of:
s1: processing the training data and extracting periodic characteristic data; the periodic characteristic data comprise hour period characteristic data, day period characteristic data, week period characteristic data and month period characteristic data;
s2: creating a parallel time sequence convolutional neural network model according to the periodic characteristic data, wherein the parallel time sequence convolutional neural network model comprises four time convolutional neural network TCN models;
s3: and carrying out medium-short term prediction on the power load by using the parallel time sequence convolutional neural network model, and outputting a prediction result.
Preferably, the processing the training data includes:
the date data is represented by discrete vectors, the load value, the highest temperature and the lowest temperature are represented by floating point values, and the maximum normalization is carried out:
wherein x is the original characteristic value, x min As the minimum value in the training set, x max For maximum value in training set, x scale Is normalized value.
Preferably, the creating a parallel time series convolutional neural network model according to the periodic characteristic data includes:
the parallel time sequence convolutional neural network model consists of four TCN models, each TCN model corresponds to different periodic characteristic data, each type of periodic characteristic data is used as the input of a single TCN model, and the output results of the four TCN models are respectively obtained;
the output result of each TCN model is respectively passed through a Sigmoid gate control unit, and the output result is used for controlling the size of data quantity flowing into the next layer structure through a Sigmoid activation function, wherein,
wherein x is 1 、x 2 、x 3 、x 4 Respectively represent hour period characteristic data, day period characteristic data, week period characteristic data and month period characteristic data, W 1 、W 2 、W 3 、W 4 、b 1 、b 2 、b 3 、b 4 Weight matrix and bias vector parameters for linear layer, H 1 、H 2 、H 3 、H 4 The output result of the Sigmoid gating unit.
Preferably, the performing medium-short term prediction on the power load by using the parallel time sequence convolutional neural network model, and outputting a prediction result includes:
the outputs of the four TCN models are subjected to a Sigmoid gating unit to obtain an output H 1 、H 2 、H 3 、H 4 Will H 1 、H 2 、H 3 、H 4 And performing Concat splicing on the four vectors, and finally obtaining a final prediction result y through a multi-layer fully connected neural network MLP:
y＝MLP(Concat(H 1 ,H 2 ,H 3 ,H 4 ))
MLP(x)＝W n (...relu(W 2 (relu(W 1 x+b 1 ))+b 2 )...)+b n
wherein H is 1 、H 2 、H 3 、H 4 For the output of the previous layer, W i 、b i The MLP network has n layers in total, which are formed by stacking a plurality of fully connected neural networks, and the weight matrix and the bias vector parameters of the linear layers are determined by relu as an activation function.
Preferably, the method further comprises:
and calling an evaluation algorithm to evaluate the prediction result of the parallel time sequence convolutional neural network model.
According to another aspect of the present invention, there is also provided a system for short-term prediction of power load based on a parallel time series convolutional neural network, the system comprising:
the processing module is used for processing the training data and extracting periodic characteristic data; the periodic characteristic data comprise hour period characteristic data, day period characteristic data, week period characteristic data and month period characteristic data;
the creating module is used for creating a parallel time sequence convolutional neural network model according to the periodic characteristic data, wherein the parallel time sequence convolutional neural network model comprises four time convolutional neural network TCN models;
and the prediction module is used for performing medium-short term prediction on the power load by using the parallel time sequence convolutional neural network model and outputting a prediction result.
Preferably, the processing module processes the training data includes:
the date data is represented by discrete vectors, the load value, the highest temperature and the lowest temperature are represented by floating point values, and the maximum normalization is carried out:
wherein x is the originalEigenvalues, x min As the minimum value in the training set, x max For maximum value in training set, x scale Is normalized value.
Preferably, the creating module creates a parallel time sequence convolutional neural network model according to the periodic characteristic data includes:
the parallel time sequence convolutional neural network model consists of four TCN models, each TCN model corresponds to different periodic characteristic data, each type of periodic characteristic data is used as the input of a single TCN model, and the output results of the four TCN models are respectively obtained;
the output result of each TCN model is respectively passed through a Sigmoid gate control unit, and the output result is used for controlling the size of data quantity flowing into the next layer structure through a Sigmoid activation function, wherein,
wherein x is 1 、x 2 、x 3 、x 4 Respectively represent hour period characteristic data, day period characteristic data, week period characteristic data and month period characteristic data, W 1 、W 2 、W 3 、W 4 、b 1 、b 2 、b 3 、b 4 Weight matrix and bias vector parameters for linear layer, H 1 、H 2 、H 3 、H 4 The output result of the Sigmoid gating unit.
Preferably, the predicting module performs medium-short term prediction on the power load by using the parallel time sequence convolutional neural network model, and outputting a prediction result includes:
the outputs of the four TCN models are subjected to a Sigmoid gating unit to obtain an output H 1 、H 2 、H 3 、H 4 Will H 1 、H 2 、H 3 、H 4 And performing Concat splicing on the four vectors, and finally obtaining a final prediction result y through a multi-layer fully connected neural network MLP:
y＝MLP(Concat(H 1 ,H 2 ,H 3 ,H 4 ))
MLP(x)＝W n (...relu(W 2 (relu(W 1 x+b 1 ))+b 2 )...)+b n
wherein H is 1 、H 2 、H 3 、H 4 For the output of the previous layer, W i 、b i The MLP network has n layers in total, which are formed by stacking a plurality of fully connected neural networks, and the weight matrix and the bias vector parameters of the linear layers are determined by relu as an activation function.
Preferably, the system further comprises an evaluation module for:
and calling an evaluation algorithm to evaluate the prediction result of the parallel time sequence convolutional neural network model.
The beneficial effects are that: aiming at the periodic characteristics of the load, the invention provides a fusion scheme of the parallel time sequence convolutional neural network model with different periodic characteristics, and the multicycle characteristics of the load data are effectively reflected, thereby improving the accuracy of short-term prediction in the power load. The effectiveness and superiority of the proposed method in medium-short term prediction are verified through modeling and training of certain city load data.
Features and advantages of the present invention will become apparent by reference to the following drawings and detailed description of embodiments of the invention.
Drawings
FIG. 1 is a flow chart of a method for short-term prediction of power load based on a parallel time sequential convolutional neural network;
FIG. 2 is a sample of different periodic characteristic loads;
FIG. 3 is a TCN model network architecture;
FIG. 4 is a parallel time series convolutional neural network learning model;
FIG. 5 is a Sigmoid gating cell;
FIG. 6 is a model load prediction result;
FIG. 7 is a graph showing the results of comparative evaluation of the RMSE and MAE indicators of different models;
fig. 8 is a schematic diagram of a power load medium-short term prediction system based on a parallel time series convolutional neural network.
Detailed Description
The following description of the embodiments of the present invention will be made clearly and fully with reference to the accompanying drawings, in which it is evident that the embodiments described are only some, but not all embodiments of the invention. All other embodiments, which can be made by those skilled in the art based on the embodiments of the invention without making any inventive effort, are intended to be within the scope of the invention.
Example 1
FIG. 1 is a flow chart of a method for short-term prediction of power load based on a parallel time series convolutional neural network. As shown in fig. 1, the present embodiment provides a method for short-term prediction of power load based on a parallel time series convolutional neural network, which includes the following steps:
s1: processing the training data and extracting periodic characteristic data; the periodic characteristic data includes hour period characteristic data, day period characteristic data, week period characteristic data, and month period characteristic data.
Preferably, the processing the training data includes:
the date data is represented by discrete vectors, the load value, the highest temperature and the lowest temperature are represented by floating point values, and the maximum normalization is carried out:
wherein x is the original characteristic value, x min As the minimum value in the training set, x max For maximum value in training set, x scale Is normalized value.
Specifically, the training data used in this embodiment is power load data sampled every hour, and 24 sampling values are taken every day. In addition to the load data itself, various external environmental factors may have an effect on the power load values, such as weather data, economic and demographic data, and the like. In addition to using hourly collected electrical load signature data, two classes of signatures can be added. The first is to add the day information (monday to sunday), the highest air temperature, the lowest air temperature, and whether or not the holiday is a feature of the day cycle feature. The second is to add the data feature of GDP and the growth rate of industrial production to the month period feature.
In the training data preparation process, the week data is represented by discrete 0/1 vectors, for example, the Monday vector [1,0,0,0,0,0,0] and the Monday vector [0,0,0,0,0,0,1 ]. Whether or not the holiday data is also expressed by 0/1, 1 is the holiday, and 0 is the non-holiday. Other data such as load value, maximum temperature, minimum temperature, etc. all use floating point values and are normalized for the most value:
wherein x is the original characteristic value, x min As the minimum value in the training set, x max For maximum value in training set, x scale Is normalized value. Through normalization, all the characteristic values are adjusted to be in a similar range of a 0-1 interval, and the gradient value is updated more stably, so that the convergence of the model can be accelerated. And when load prediction is carried out, carrying out inverse normalization on the predicted value of the model, thereby obtaining a final predicted value.
The extracting the periodic characteristic data specifically comprises:
the influencing factors of the electrical load are manifold. The power loads in the morning, the daytime and the night are different, and the daily cycle law is shown. On weekdays, weekends and holidays, the power load presents different numerical characteristics and shows a cycle law. The air temperature also has a certain influence on the electric load, and the electric load is characterized by winter and summer double peaks and shows a month period rule.
In this embodiment, a sliding window is adopted, and for the current predicted point, the current load value is predicted using the load data of the previous α hour as the hour period characteristic data, the load data of the previous β day as the day period characteristic data, the load data of the previous γ week as the week period characteristic data, and the load data of the previous δ month as the month period characteristic data. The specific duty cycle characteristic data used is shown in fig. 2.
S2: and creating a parallel time sequence convolutional neural network model according to the periodic characteristic data, wherein the parallel time sequence convolutional neural network model comprises four time convolutional neural network TCN models.
Preferably, the creating a parallel time series convolutional neural network model according to the periodic characteristic data includes:
the parallel time sequence convolutional neural network model consists of four TCN models, each TCN model corresponds to different periodic characteristic data, each type of periodic characteristic data is used as the input of a single TCN model, and the output results of the four TCN models are respectively obtained;
the output result of each TCN model is respectively passed through a Sigmoid gate control unit, and the output result is used for controlling the size of data quantity flowing into the next layer structure through a Sigmoid activation function, wherein,
wherein x is 1 、x 2 、x 3 、x 4 Respectively represent hour period characteristic data, day period characteristic data, week period characteristic data and month period characteristic data, W 1 、W 2 、W 3 、W 4 、b 1 、b 2 、b 3 、b 4 Weight matrix and bias vector parameters for linear layer, H 1 、H 2 、H 3 、H 4 The output result of the Sigmoid gating unit.
Specifically, time series prediction is a problem of series modeling, where the model inputs values from time 0 to time T-1, predicts the value of time T, and the model aims at making the true value and the predicted value as close as possible.
The time convolutional neural network TCN is a variant of the convolutional neural network, combines the characteristics of the convolutional neural network RNN (Recurrent Neural Network) and the convolutional neural network CNN, is a causal convolutional of dilation, and can be used for sequence modeling tasks. The structure of the time sequence convolution neural network is shown in fig. 3, in the figure, taking the convolution kernel size as 3 and the network depth as 3 as an example, a calculation method of the last output node is drawn, d is the expansion factor size and represents the step length between calculation nodes, and the result of the later layer in the model is that the input of the former layer is obtained through expansion convolution operation.
The parallel time series convolutional neural network structure is shown in fig. 4. The parallel TCN network is composed of four TCN structures, and each TCN structure independently models different characteristic data, namely an hour period characteristic, a day period characteristic, a week period characteristic and a month period characteristic. And taking each type of periodic characteristics as the input of a single TCN model to respectively obtain the output results of the four TCN models.
The output result of each TCN model structure passes through a Sigmoid gating unit, and the gating unit structure is shown in fig. 5. The Sigmoid activation function is used to control the size of the amount of data flowing into the next layer structure.
Wherein x is 1 、x 2 、x 3 、x 4 Input data representing hour cycle characteristics, day cycle characteristics, week cycle characteristics, month cycle characteristics, respectively, W 1 、W 2 、W 3 、W 4 、b 1 、b 2 、b 3 、b 4 Is a weight matrix and a bias vector parameter of the linear layer.
The gating unit is used for controlling the contribution quantity of input data in different periods, the value range of the output of the Sigmoid is (0, 1), the larger the Sigmoid is, the larger the contribution quantity of the period characteristic is, the closer the Sigmoid is to 0, and the smaller the contribution quantity of the period characteristic is.
S3: and carrying out medium-short term prediction on the power load by using the parallel time sequence convolutional neural network model, and outputting a prediction result.
Preferably, the performing medium-short term prediction on the power load by using the parallel time sequence convolutional neural network model, and outputting a prediction result includes:
the outputs of the four TCN models are subjected to a Sigmoid gating unit to obtain an output H 1 、H 2 、H 3 、H 4 Will H 1 、H 2 、H 3 、H 4 And performing Concat splicing on the four vectors, and finally obtaining a final prediction result y through a multi-layer fully connected neural network MLP:
y＝MLP(Concat(H 1 ,H 2 ,H 3 ,H 4 ))
MLP(x)＝W n (...relu(W 2 (relu(W 1 x[b 1 ))+b 2 )...)+b n
wherein H is 1 、H 2 、H 3 、H 4 For the output of the previous layer, W i 、b i The MLP network has n layers in total, which are formed by stacking a plurality of fully connected neural networks, and the weight matrix and the bias vector parameters of the linear layers are determined by relu as an activation function.
Specifically, parallel TCN networks model data characteristics of different periods at the same time, and outputs of four TCN network models are subjected to Sigmoid gating units to obtain an output H 1 、H 2 、H 3 、H 4 Thereafter, H is 1 、H 2 、H 3 、H 4 And performing Concat splicing on the four vectors, and finally obtaining a final prediction result y through a multi-layer fully connected neural network MLP:
y＝MLP(Concat(H 1 ,H 2 ,H 3 ,H 4 ))
wherein H is 1 、H 2 、H 3 、H 4 Is the output of the previous layer. The multi-layer neural network MLP is formed by stacking a plurality of fully connected neural networks (Full Connection):
MLP(x)＝W n (...relu(W 2 (relu(W 1 x+b 1 ))+b 2 )...)+b n
in which W is i 、b i For the weight matrix and the bias vector parameters of the linear layer, relu is an activation function, and the MLP network shares n layers.
A single TCN convolutional neural network may not be able to capture all of the different periodic characteristics simultaneously. By utilizing a plurality of TCN networks, each TCN network is only responsible for learning one periodic characteristic, so that the model has stronger learning and characterization capabilities.
Preferably, the method further comprises:
and calling an evaluation algorithm to evaluate the prediction result of the parallel time sequence convolutional neural network model.
Specifically, model training and prediction are performed by using a PyTorch training framework, an Adam optimizer is selected as an optimizer, and the initial learning rate is set to be 0.001. Model overfitting is prevented by dropout, the value of which is set to 0.2.
In order to debug out the optimal parameters, a grid search method is used, comparison experiments are conducted on all possible parameter combinations of the number of convolution filters, the size of convolution kernels and the number of expansion causal convolution layers, and an optimal super-parameter combination value of an experiment result is selected. The number of convolution filters is from [16,32,64,128 ]]Is selected from [3,5,7,9 ] convolution kernel sizes]The number of layers of the swell-causal convolution is selected from [3,4,5,6,7 ]]Is selected from the group consisting of the expansion factor of the ith layer of 2 i-1 . The last MLP layer in the network has 3 layers, each layer has 64 nodes, and the activation function selects the ReLU function.
The Root Mean Square Error (RMSE) and the Mean Absolute Error (MAE) are used herein to measure the predictive effect of the model, and the calculation formulas for RMSE and MAE are as follows:
where N is the number of samples, y i Is the actual load value that is to be applied,is a predictive value of the model. The smaller the values of RMSE and MAE, the higher the prediction accuracy of the model.
The experimental result of the model is shown in fig. 7, the curve comparison graph of load prediction is shown in fig. 6, the performance effects of parallel TCN and LSTM, CNN, TCN models are compared in the experiment, the indexes of RMSE and MAE of the parallel TCN models are best, the prediction error is minimum, and the effect is optimal.
Aiming at the periodic characteristics of the load, the embodiment provides a fusion scheme of the parallel time sequence convolutional neural network model with different periodic characteristics, and the multicycle characteristics of the load data are effectively reflected, so that the accuracy of short-term prediction in the power load is improved. The effectiveness and superiority of the proposed method in medium-short term prediction are verified through modeling and training of certain city load data.
Example 2
Fig. 8 is a schematic diagram of a power load medium-short term prediction system based on a parallel time series convolutional neural network. As shown in fig. 8, the present embodiment further provides a system for short-term prediction in power load based on a parallel time series convolutional neural network, the system comprising:
a processing module 801, configured to process training data and extract periodic feature data; the periodic characteristic data comprise hour period characteristic data, day period characteristic data, week period characteristic data and month period characteristic data;
a creating module 802, configured to create a parallel time-series convolutional neural network model according to the periodic feature data, where the parallel time-series convolutional neural network model includes four time-convolutional neural network TCN models;
and the prediction module 803 is used for performing medium-short term prediction on the power load by using the parallel time sequence convolutional neural network model and outputting a prediction result.
Preferably, the processing module 801 processes training data includes:
the date data is represented by discrete vectors, the load value, the highest temperature and the lowest temperature are represented by floating point values, and the maximum normalization is carried out:
wherein x is the original characteristic value, x min As the minimum value in the training set, x max For maximum value in training set, x scale Is normalized value.
Preferably, the creating module 802 creates a parallel time series convolutional neural network model according to the periodic characteristic data includes:
the parallel time sequence convolutional neural network model consists of four TCN models, each TCN model corresponds to different periodic characteristic data, each type of periodic characteristic data is used as the input of a single TCN model, and the output results of the four TCN models are respectively obtained;
the output result of each TCN model is respectively passed through a Sigmoid gate control unit, and the output result is used for controlling the size of data quantity flowing into the next layer structure through a Sigmoid activation function, wherein,
wherein x is 1 、x 2 、x 3 、x 4 Respectively represent hour period characteristic data, day period characteristic data, week period characteristic data and month period characteristic data, W 1 、W 2 、W 3 、W 4 、b 1 、b 2 、b 3 、b 4 Weight matrix and bias vector parameters for linear layer, H 1 、H 2 、H 3 、H 4 The output result of the Sigmoid gating unit.
Preferably, the predicting module 803 uses the parallel time series convolutional neural network model to perform medium-short term prediction on the power load, and outputting a prediction result includes:
the outputs of the four TCN models are subjected to a Sigmoid gating unit to obtain an output H 1 、H 2 、H 3 、H 4 Will H 1 、H 2 、H 3 、H 4 And performing Concat splicing on the four vectors, and finally obtaining a final prediction result y through a multi-layer fully connected neural network MLP:
y＝MLP(Concat(H 1 ,H 2 ,H 3 ,H 4 ))
MLP(x)＝W n (...relu(W 2 (relu(W 1 x+b 1 ))+b 2 )...)+b n
wherein H is 1 、H 2 、H 3 、H 4 For the output of the previous layer, W i 、b i The MLP network has n layers in total, which are formed by stacking a plurality of fully connected neural networks, and the weight matrix and the bias vector parameters of the linear layers are determined by relu as an activation function.
Preferably, the system further comprises an evaluation module 804 for:
and calling an evaluation algorithm to evaluate the prediction result of the parallel time sequence convolutional neural network model.
The implementation process of the functions implemented by each module in this embodiment 2 is the same as the implementation process of each step in embodiment 1, and will not be described here again.
The foregoing description is only of the preferred embodiments of the present invention and is not intended to limit the scope of the invention, and all equivalent structural changes made by the specification and drawings of the present invention or direct/indirect application in other related technical fields are included in the scope of the present invention.
Claims (7)
1. A method for short-term prediction of power load based on a parallel time series convolutional neural network, the method comprising the steps of:
s1: processing the training data and extracting periodic characteristic data; the periodic characteristic data comprise hour period characteristic data, day period characteristic data, week period characteristic data and month period characteristic data;
s2: creating a parallel time sequence convolutional neural network model according to the periodic characteristic data, wherein the parallel time sequence convolutional neural network model comprises four time convolutional neural network TCN models;
s3: performing medium-short term prediction on the power load by using the parallel time sequence convolutional neural network model, and outputting a prediction result;
the processing of the training data comprises:
the date data is represented by discrete vectors, the load value, the highest temperature and the lowest temperature are represented by floating point values, and the maximum normalization is carried out:
wherein x is the original characteristic value, x min As the minimum value in the training set, x max For maximum value in training set, x scale Is normalized value;
the creating a parallel time sequence convolutional neural network model according to the periodic characteristic data comprises the following steps:
the parallel time sequence convolutional neural network model consists of four TCN models, each TCN model corresponds to different periodic characteristic data, each type of periodic characteristic data is used as the input of a single TCN model, and the output results of the four TCN models are respectively obtained;
the output result of each TCN model is respectively passed through a Sigmoid gate control unit, and the output result is used for controlling the size of data quantity flowing into the next layer structure through a Sigmoid activation function, wherein,
wherein x is 1 、x 2 、x 3 、x 4 Respectively represent hour period characteristic data, day period characteristic data, week period characteristic data and month period characteristic data, W 1 、W 2 、W 3 、W 4 、b 1 、b 2 、b 3 、b 4 Weight matrix and bias vector parameters for linear layer, H 1 、H 2 、H 3 、H 4 The output result of the Sigmoid gating unit;
the method for performing medium-short term prediction on the power load by using the parallel time sequence convolutional neural network model, and outputting a prediction result comprises the following steps:
the outputs of the four TCN models are subjected to a Sigmoid gating unit to obtain an output H 1 、H 2 、H 3 、H 4 Will H 1 、H 2 、H 3 、H 4 And performing Concat splicing on the four vectors, and finally obtaining a final prediction result y through a multi-layer fully connected neural network MLP:
y＝MLP(Concat(H 1 ,H 2 ,H 3 ,H 4 ))
MLP(x)＝W n (...relu(W 2 (relu(W 1 x+b 1 ))+b 2 )...)+b n
wherein H is 1 、H 2 、H 3 、H 4 For the output of the previous layer, W i 、b i The MLP network has n layers in total, which are formed by stacking a plurality of fully connected neural networks, and the weight matrix and the bias vector parameters of the linear layers are determined by relu as an activation function.
2. The method according to claim 1, wherein the method further comprises:
and calling an evaluation algorithm to evaluate the prediction result of the parallel time sequence convolutional neural network model.
3. A power load medium-short term prediction system based on a parallel time sequence convolutional neural network, which is applied to a power load medium-short term prediction method based on a parallel time sequence convolutional neural network as recited in claim 2, and is characterized in that the system comprises:
the processing module is used for processing the training data and extracting periodic characteristic data; the periodic characteristic data comprise hour period characteristic data, day period characteristic data, week period characteristic data and month period characteristic data;
the creating module is used for creating a parallel time sequence convolutional neural network model according to the periodic characteristic data, wherein the parallel time sequence convolutional neural network model comprises four time convolutional neural network TCN models;
and the prediction module is used for performing medium-short term prediction on the power load by using the parallel time sequence convolutional neural network model and outputting a prediction result.
4. The system of claim 3, wherein the processing module to process training data comprises:
the date data is represented by discrete vectors, the load value, the highest temperature and the lowest temperature are represented by floating point values, and the maximum normalization is carried out:
wherein x is the original characteristic value, x min As the minimum value in the training set, x max For maximum value in training set, x scale Is normalized value.
5. The system of claim 4, wherein the creation module creating a parallel sequential convolutional neural network model from the periodic feature data comprises:
the parallel time sequence convolutional neural network model consists of four TCN models, each TCN model corresponds to different periodic characteristic data, each type of periodic characteristic data is used as the input of a single TCN model, and the output results of the four TCN models are respectively obtained;
the output result of each TCN model is respectively passed through a Sigmoid gate control unit, and the output result is used for controlling the size of data quantity flowing into the next layer structure through a Sigmoid activation function, wherein,
wherein x is 1 、x 2 、x 3 、x 4 Respectively represent the characteristic data of the hour period,Daily cycle characteristic data, weekly cycle characteristic data, monthly cycle characteristic data, W 1 、W 2 、W 3 、W 4 、b 1 、b 2 、b 3 、b 4 Weight matrix and bias vector parameters for linear layer, H 1 、H 2 、H 3 、H 4 The output result of the Sigmoid gating unit.
6. The system of claim 5, wherein the prediction module performs a short-term prediction of the electrical load using the parallel time series convolutional neural network model and outputting a prediction result comprises:
the outputs of the four TCN models are subjected to a Sigmoid gating unit to obtain an output H 1 、H 2 、H 3 、H 4 Will H 1 、H 2 、H 3 、H 4 And performing Concat splicing on the four vectors, and finally obtaining a final prediction result y through a multi-layer fully connected neural network MLP:
y＝MLP(Concat(H 1 ,H 2 ,H 3 ,H 4 ))
MLP(x)＝W n (...relu(W 2 (relu(W 1 x+b 1 ))+b 2 )...)+b n
wherein H is 1 、H 2 、H 3 、H 4 For the output of the previous layer, W i 、b i The MLP network has n layers in total, which are formed by stacking a plurality of fully connected neural networks, and the weight matrix and the bias vector parameters of the linear layers are determined by relu as an activation function.
7. The system of claim 6, further comprising an evaluation module for:
and calling an evaluation algorithm to evaluate the prediction result of the parallel time sequence convolutional neural network model.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202211259893.2A CN115545503B (en) | 2022-10-14 | 2022-10-14 | Power load medium-short term prediction method and system based on parallel time sequence convolutional neural network |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202211259893.2A CN115545503B (en) | 2022-10-14 | 2022-10-14 | Power load medium-short term prediction method and system based on parallel time sequence convolutional neural network |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN115545503A CN115545503A (en) | 2022-12-30 |
| CN115545503B true CN115545503B (en) | 2024-02-02 |
Family
ID=84735068
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202211259893.2A Active CN115545503B (en) | 2022-10-14 | 2022-10-14 | Power load medium-short term prediction method and system based on parallel time sequence convolutional neural network |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN115545503B (en) |
Families Citing this family (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116436002B (en) * | 2023-06-13 | 2023-09-05 | 成都航空职业技术学院 | Building electricity utilization prediction method |
Citations (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN114358389A (en) * | 2021-12-13 | 2022-04-15 | 重庆邮电大学 | Short-term power load prediction method combining VMD decomposition and time convolution network |
-
2022
- 2022-10-14 CN CN202211259893.2A patent/CN115545503B/en active Active
Patent Citations (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN114358389A (en) * | 2021-12-13 | 2022-04-15 | 重庆邮电大学 | Short-term power load prediction method combining VMD decomposition and time convolution network |
Also Published As
| Publication number | Publication date |
|---|---|
| CN115545503A (en) | 2022-12-30 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Liu et al. | Short-term load forecasting using a long short-term memory network | |
| CN108280551B (en) | Photovoltaic power generation power prediction method utilizing long-term and short-term memory network | |
| CN109685252B (en) | Building energy consumption prediction method based on cyclic neural network and multi-task learning model | |
| Khotanzad et al. | A neuro-fuzzy approach to short-term load forecasting in a price-sensitive environment | |
| CN108764539B (en) | Upstream and downstream water level prediction method for cascade power station | |
| CN111260136A (en) | Building short-term load prediction method based on ARIMA-LSTM combined model | |
| CN111027772A (en) | Multi-factor short-term load prediction method based on PCA-DBILSTM | |
| CN112330215A (en) | Urban vehicle demand prediction method, equipment and storage medium | |
| CN113554466B (en) | Short-term electricity consumption prediction model construction method, prediction method and device | |
| Lu et al. | A hybrid model based on convolutional neural network and long short-term memory for short-term load forecasting | |
| CN114742278A (en) | Building energy consumption prediction method and system based on improved LSTM | |
| CN115545503B (en) | Power load medium-short term prediction method and system based on parallel time sequence convolutional neural network | |
| Wang et al. | Short-term load forecasting of power system based on time convolutional network | |
| CN114119273A (en) | Park comprehensive energy system non-invasive load decomposition method and system | |
| CN113591368A (en) | Comprehensive energy system multi-energy load prediction method and system | |
| CN112288140A (en) | Keras-based short-term power load prediction method, storage medium and equipment | |
| Sun et al. | The short-term load forecasting using an artificial neural network approach with periodic and nonperiodic factors: a case study of Tai'an, Shandong Province, China | |
| CN111178612A (en) | LSTM load prediction method of grid user based on big data ODPS engine | |
| CN114611757A (en) | Electric power system short-term load prediction method based on genetic algorithm and improved depth residual error network | |
| Ibrahim et al. | LSTM neural network model for ultra-short-term distribution zone substation peak demand prediction | |
| CN115860277B (en) | Data center energy consumption prediction method and system | |
| CN115481788B (en) | Phase change energy storage system load prediction method and system | |
| Prema et al. | LSTM based Deep Learning model for accurate wind speed prediction | |
| CN110659775A (en) | LSTM-based improved electric power short-time load prediction algorithm | |
| Zuo | Integrated forecasting models based on LSTM and TCN for short-term electricity load forecasting |
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 | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |