CN111738512B - Short-term power load prediction method based on CNN-IPSO-GRU hybrid model - Google Patents

Short-term power load prediction method based on CNN-IPSO-GRU hybrid model Download PDF

Info

Publication number
CN111738512B
CN111738512B CN202010573272.6A CN202010573272A CN111738512B CN 111738512 B CN111738512 B CN 111738512B CN 202010573272 A CN202010573272 A CN 202010573272A CN 111738512 B CN111738512 B CN 111738512B
Authority
CN
China
Prior art keywords
gru
unit network
model
prediction
prediction model
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
Application number
CN202010573272.6A
Other languages
Chinese (zh)
Other versions
CN111738512A (en
Inventor
刘可真
苟家萁
骆钊
徐玥
李鹤健
和婧
王骞
刘通
阮俊枭
吴世浙
陈雪鸥
陈镭丹
迟焕斌
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Kunming University of Science and Technology
Original Assignee
Kunming University of Science and Technology
Priority date (The priority date 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 date listed.)
Filing date
Publication date
Application filed by Kunming University of Science and Technology filed Critical Kunming University of Science and Technology
Priority to CN202010573272.6A priority Critical patent/CN111738512B/en
Publication of CN111738512A publication Critical patent/CN111738512A/en
Application granted granted Critical
Publication of CN111738512B publication Critical patent/CN111738512B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Images

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
    • G06Q10/04—Forecasting or optimisation specially adapted for administrative or management purposes, e.g. linear programming or "cutting stock problem"
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/004—Artificial life, i.e. computing arrangements simulating life
    • G06N3/006—Artificial life, i.e. computing arrangements simulating life based on simulated virtual individual or collective life forms, e.g. social simulations or particle swarm optimisation [PSO]
    • 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
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/045—Combinations of networks
    • 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

Abstract

The invention discloses a short-term power load prediction method based on a CNN-IPSO-GRU mixed model, which comprises the steps of firstly collecting data such as historical load, meteorological factors, date information and the like of a power grid, carrying out data normalization processing, dividing a training set and a test set, extracting multi-dimensional characteristic vectors representing load changes by utilizing a convolutional neural network technology, and constructing a time sequence as the input of a model; then constructing a gated circulation unit network prediction model, optimizing the gated circulation unit network prediction model by using training set data through an improved particle swarm algorithm to obtain two optimal prediction model parameters, and reestablishing the gated circulation unit network model according to the obtained optimal prediction model parameters; and finally, short-term load prediction of the power grid is realized by using the test set data. The method provided by the invention can accurately predict the short-term load change trend of the power grid, and further plays an important role in reducing the loss of the generator set and ensuring the economic and reliable operation of the power grid.

Description

Short-term power load prediction method based on CNN-IPSO-GRU hybrid model
Technical Field
The invention relates to a power load prediction method, in particular to a short-term power load prediction method based on a CNN-IPSO-GRU mixed model.
Background
With the rapid development of the electric power market in China, efficient and accurate short-term load prediction is an important content of power grid research, and the accurate short-term load prediction plays an important role in reducing the loss of a generator set and ensuring the economic and reliable operation of a power grid. Therefore, a new method for improving the load prediction precision needs to be researched to improve the economic benefit of the power grid.
A large amount of research is carried out on short-term load prediction by numerous scholars at home and abroad for many years, and the short-term load prediction can be summarized into three types: statistical methods, model combination methods and machine learning methods. The statistical method mainly comprises a time sequence model, a fuzzy prediction model and the like, and the prediction effect of the method has higher requirements on the distribution characteristics of the data set; there are two main categories of combinatorial prediction methods: firstly, different prediction models are effectively combined, and secondly, a data analysis method is used for preprocessing an original data sequence, wherein the analysis method comprises a convolutional neural network, Empirical Mode Decomposition (EMD) and the like, and then prediction is carried out by combining the models; the Machine learning method mainly includes a Recurrent Neural Network (RNN), a Support Vector Machine (SVM), a Long-short-term memory Network (LSTM), a Gated Recurrent Unit (GRU), and the like.
Because the factors influencing the load fluctuation are many, the factors mainly include historical load, meteorological factors, date types and the like, and considering that the historical data set has the characteristics of complexity and time sequence, a Convolutional Neural Network (CNN) has the characteristic of reducing the complexity of the multidimensional data set, and is currently applied to the aspect of extracting feature vectors during load prediction. RNN has a memory function, and is widely applied to the aspect of predicting time series, LSTM and GRU are both special RNN, and have higher precision in the aspect of predicting time series. The GRU is improved on the basis of an LSTM structure, compared with the LSTM, the GRU has the advantages that the structure is simpler, the convergence rate is higher, and the prediction accuracy is higher. The PSO has the advantages of few adjusting parameters, concise iterative optimization thought, high convergence speed and the like, and is widely applied to the aspect of determining model parameters, but the PSO is easy to fall into local optimization in the optimization process, so that the defect of larger error result and the like is caused, and an Improved Particle Swarm Optimization (IPSO) is provided to enhance the optimization capability of the model parameters.
Disclosure of Invention
The invention aims to solve the technical problem of providing a short-term power load prediction method for optimizing a gate control cycle unit network (CNN-IPSO-GRU) mixed model based on a convolutional neural network and an improved particle swarm algorithm.
In order to solve the technical problem, the invention provides a short-term power load prediction method based on a CNN-IPSO-GRU mixed model, which comprises the following steps:
A. collecting historical load, meteorological factors and date information data of a power grid as characteristic parameters;
B. acquiring sample data of the characteristic parameters in the step A in the historical time dimension power grid load, carrying out normalization processing on the sample data, extracting a multi-dimensional characteristic vector representing load change by using a Convolutional Neural Network (CNN), constructing the multi-dimensional characteristic vector into a time sequence, and dividing training set data and test set data according to the time sequence;
C. constructing a gate control cycle unit network prediction model GRU comprising an input layer, a hidden layer and an output layer;
D. performing improved particle swarm optimization IPSO on the gated circulation unit network prediction model GRU constructed in the step C by using the training set data obtained in the step B, determining the optimal parameters of the gated circulation unit network prediction model GRU on the basis of meeting the optimal evaluation index of the prediction model, and reestablishing the optimized gated circulation unit network prediction model IPSO-GRU on the basis of the optimal parameters;
E. And D, using the test set data obtained in the step B as input variables of the optimized gated cyclic unit network prediction model IPSO-GRU in the step D, and further performing power grid short-term load prediction to obtain a prediction result.
The step B of normalizing the sample data means that the sample data is mapped between [0, 1], and the normalized formula is expressed as formula (1):
Figure GDA0003534107140000021
in the formula (1), x*The normalized data is obtained; x is the number ofmin、xmaxThe minimum value and the maximum value of the sample data set are respectively, and x is the original sample data.
The concrete construction process of the gate control cycle unit network prediction model GRU in the step C is as follows:
c1, collecting power grid historical load, meteorological factor and date information data as characteristic parameters, extracting multi-dimensional characteristic vectors representing load changes by using a Convolutional Neural Network (CNN), and constructing a time sequence as sample data of a gate control cycle unit network prediction model GRU input layer;
c2, carrying out normalization processing on the sample data of each characteristic parameter in the step C1, mapping the sample data to the position between [0 and 1], and dividing the training set data and the test set data according to the time sequence;
c3, the hidden layer trains the gated loop unit network prediction model GRU by adopting the training set data obtained in the step C2, and determines the optimal parameters of the gated loop unit network prediction model GRU under the condition that the evaluation standard error of the prediction result is smaller;
C4, adopting the gate control circulation unit network prediction model GRU obtained in the step C3 to predict the data of the test set, and obtaining a prediction result P1、P2、…Pn；
And C5, calculating n prediction results obtained in the step C4 by an average absolute percentage error mode through an output layer, performing inverse normalization processing to obtain a final prediction result, and finally performing evaluation analysis on the prediction result by combining with an experimental evaluation index.
The process of optimizing the gated cyclic unit network prediction model GRU by using the improved particle swarm optimization IPSO in the step D is as follows:
d1, initializing GRU parameters of the gate control cycle unit network prediction model constructed in the step C, setting respective value ranges and search ranges of the number m of neurons and the learning rate epsilon, and determining the maximum iteration timeNumber TmaxWith the population size ps and the maximum value omega of the inertial weight omegamaxWith a minimum value ωminAcceleration factor c1、c2Maximum and minimum values of;
d2, performing model prediction on the test set data by using the gated cycle unit network prediction model GRU constructed in the step C according to the initialized neuron number m and the learning rate epsilon in the step D1, and taking the average absolute percentage error of the obtained prediction result as the fitness value of the particles, wherein the fitness function f is a fitness function itiAs defined in formula (2):
Figure GDA0003534107140000031
wherein n represents the sample volume of the test set data; xact(i) And Xpred(i) Respectively a real value and a predicted value of the load at the moment i;
d3, taking two parameters of the number m of the neurons and the learning rate epsilon as particles, taking the average absolute percentage error of the prediction result obtained in the step D2 as a particle fitness value, and iteratively updating the speeds and the positions of the two particles by adopting formulas (3) to (7), namely updating the values of two key parameters of the gated loop unit network prediction model GRU; after the optimal values of the two parameters are obtained by improving particle swarm optimization IPSO, an optimized gated cyclic unit network prediction model IPSO-GRU is reestablished on the basis of the optimal values;
Figure GDA0003534107140000041
Figure GDA0003534107140000042
Figure GDA0003534107140000043
Figure GDA0003534107140000044
Figure GDA0003534107140000045
in the above formula, k represents the current iteration number; t ismaxRepresenting the maximum iteration number;
Figure GDA0003534107140000046
respectively representing the speed, the position, the individual local optimal solution and the global optimal solution of the particle; r is1And r2Is [0, 1 ]]A random number in between; omega (k), c1(k) And c2(k) Respectively representing the values of inertia weight and acceleration factor in the k iteration; omegamax＝0.9，ωmin＝0.4；TmaxIs the maximum iteration number; c. C1And c2∈[0.5，2.5]；a＝b＝1，c＝d＝1.5；
D4, iteratively updating the speed and the position of the particles according to formulas (3) - (7) of the improved particle swarm optimization, calculating the corresponding particle fitness value, comparing the local optimal solution with the global optimal solution, meeting a termination condition when the particle fitness value tends to be stable or the iteration frequency reaches the maximum, obtaining the optimal neuron number m and the learning rate epsilon parameter, obtaining the gated loop unit network prediction model IPSO-GRU optimized by the improved particle swarm optimization, and otherwise, returning to the step D3 to continuously perform iterative updating.
The prediction result obtained in the step E uses the average relative percentage error MAPE, the root mean square error RMSE, the prediction precision FA and the decision coefficient R2Model evaluation was performed for four evaluation indexes, which are shown in equations (8) to (11):
Figure GDA0003534107140000047
Figure GDA0003534107140000048
Figure GDA0003534107140000049
Figure GDA0003534107140000051
in the above formula: n represents the sample capacity of the test data set; xact(i) And Xpred(i) (i is 1,2, … n) is the real value and predicted value of the load at the ith moment respectively;
Figure GDA0003534107140000052
an average value representing the true value of the prediction sample; wherein the smaller the MAPE and RMSE values are, the larger the FA value is, and R is2The closer to 1 the value of (a) indicates the greater the goodness of fit, the more accurate the model prediction result.
The method comprises the steps of firstly, collecting data such as historical load, meteorological factors and date information of a power grid, carrying out data normalization processing, dividing training set data and test set data, extracting multi-dimensional characteristic vectors representing load changes by using a convolutional neural network, and constructing a time sequence as input data of a model; then constructing a gated circulation unit network prediction model, optimizing the gated circulation unit network prediction model by using training set data through an improved particle swarm algorithm to obtain two optimal prediction model parameters, and reestablishing the gated circulation unit network model according to the obtained optimal prediction model parameters; and finally, short-term load prediction of the power grid is realized by using the test set data. The method provided by the invention can accurately predict the short-term load change trend of the power grid, and further plays an important role in reducing the loss of the generator set and ensuring the economic and reliable operation of the power grid.
Drawings
FIG. 1 is a diagram illustrating the variation of particle fitness with the number of iterations in an embodiment of the present invention;
FIG. 2 is a schematic diagram illustrating the change of the number of neurons with the change of the number of iterations according to an embodiment of the present invention;
FIG. 3 is a diagram illustrating the variation of the learning rate with the variation of the number of iterations according to an embodiment of the present invention;
FIG. 4 is a schematic diagram showing comparison between a prediction curve and an actual curve of a grid load respectively performed by using a CNN-IPSO-GRU model, a CNN-PSO-GRU model, a CNN-GRU model, an LSTM model, an RNN model and a GRU model in the embodiment of the present invention;
FIG. 5 is a diagram illustrating relative error comparisons of predictions of grid loads by using a CNN-IPSO-GRU model, a CNN-PSO-GRU model, a CNN-GRU model, an LSTM model, an RNN model, and a GRU model, respectively, in an embodiment of the present invention.
Detailed Description
In order to make the objects, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the drawings in the embodiments of the present invention, and it is obvious that the described embodiments are some embodiments, but not all embodiments, of the present invention; all other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present invention.
The invention provides a short-term power load prediction method based on a CNN-IPSO-GRU mixed model, which comprises the following steps:
A. collecting historical load, meteorological factors and date information data of a power grid as characteristic parameters;
B. acquiring sample data of the characteristic parameters in the step A in the historical time dimension power grid load, carrying out normalization processing on the sample data, extracting a multi-dimensional characteristic vector representing load change by using a Convolutional Neural Network (CNN), constructing the multi-dimensional characteristic vector into a time sequence, and dividing training set data and test set data according to the time sequence;
C. constructing a gate control cycle unit network prediction model GRU comprising an input layer, a hidden layer and an output layer;
D. performing improved particle swarm optimization IPSO on the gated circulation unit network prediction model GRU constructed in the step C by using the training set data obtained in the step B, determining the optimal parameters of the gated circulation unit network prediction model GRU on the basis of meeting the optimal evaluation index of the prediction model, and reestablishing the optimized gated circulation unit network prediction model IPSO-GRU on the basis of the optimal parameters;
E. and D, using the test set data obtained in the step B as input variables of the optimized gated loop unit network prediction model IPSO-GRU in the step D, and further performing power grid short-term load prediction to obtain a prediction result.
The step B of normalizing the sample data means that the sample data is mapped between [0 and 1], and the normalized formula is expressed by formula (1):
Figure GDA0003534107140000061
in the formula (1), x*The normalized data is obtained; x is the number ofmin、xmaxThe minimum value and the maximum value of the sample data set are respectively, and x is the original sample data.
The concrete construction process of the gate control cycle unit network prediction model GRU in the step C is as follows:
c1, collecting power grid historical load, meteorological factor and date information data as characteristic parameters, extracting multi-dimensional characteristic vectors representing load changes by using a Convolutional Neural Network (CNN), and constructing a time sequence as sample data of a gate control cycle unit network prediction model GRU input layer;
c2, carrying out normalization processing on the sample data of each characteristic parameter in the step C1, mapping the sample data to the position between [0 and 1], and dividing the training set data and the test set data according to the time sequence;
c3, the hidden layer trains the gated loop unit network prediction model GRU by adopting the training set data obtained in the step C2, and determines the optimal parameters of the gated loop unit network prediction model GRU under the condition that the evaluation standard error of the prediction result is smaller;
c4, adopting the gate control circulation unit network prediction model GRU obtained in the step C3 to predict the data of the test set, and obtaining a prediction result P 1、P2、…Pn；
And C5, calculating n prediction results obtained in the step C4 by an average absolute percentage error mode through an output layer, performing inverse normalization processing to obtain a final prediction result, and finally performing evaluation analysis on the prediction result by combining with an experimental evaluation index.
The process of optimizing the gated cyclic unit network prediction model GRU by using the improved particle swarm optimization IPSO in the step D is as follows:
d1, initializing GRU parameters of the gate control cycle unit network prediction model constructed in the step C, setting respective value ranges and search ranges of the number m of the neurons and the learning rate epsilon, and determining the maximum iteration times TmaxWith population size ps, maximum value ω of inertial weight ωmaxWith a minimum value ωminAcceleration factor c1、c2Maximum and minimum values of;
d2, performing model prediction on the test set data by using the gated cycle unit network prediction model GRU constructed in the step C according to the initialized neuron number m and the learning rate epsilon in the step D1, and taking the average absolute percentage error of the obtained prediction result as the fitness value of the particles, wherein the fitness function f is a fitness functionitiAs defined in formula (2):
Figure GDA0003534107140000071
wherein n represents the sample volume of the test set data; xact(i) And Xpred(i) Respectively a real value and a predicted value of the load at the moment i;
D3, taking two parameters of the number m of the neurons and the learning rate epsilon as particles, taking the average absolute percentage error of the prediction result obtained in the step D2 as a particle fitness value, and iteratively updating the speeds and the positions of the two particles by adopting formulas (3) to (7), namely updating the values of two key parameters of the gated loop unit network prediction model GRU; after the optimal values of the two parameters are obtained by improving particle swarm optimization IPSO, an optimized gated cyclic unit network prediction model IPSO-GRU is reestablished on the basis of the optimal values;
Figure GDA0003534107140000072
Figure GDA0003534107140000073
Figure GDA0003534107140000074
Figure GDA0003534107140000081
Figure GDA0003534107140000082
in the above formula, k represents the current iteration number; t ismaxRepresenting the maximum iteration number;
Figure GDA0003534107140000083
respectively representing the speed, the position, the individual local optimal solution and the global optimal solution of the particle; r is1And r2Is [0, 1 ]]A random number in between; omega (k), c1(k) And c2(k) Respectively representing the values of inertia weight and acceleration factor in the k iteration; omegamax＝0.9，ωmin＝0.4；TmaxIs the maximum iteration number; c. C1And c2∈[0.5，2.5]；a＝b＝1，c＝d＝1.5；
D4, iteratively updating the speed and the position of the particles according to formulas (3) - (7) of the improved particle swarm optimization, calculating the corresponding particle fitness value, comparing the local optimal solution with the global optimal solution, meeting a termination condition when the particle fitness value tends to be stable or the iteration frequency reaches the maximum, obtaining the optimal neuron number m and the learning rate epsilon parameter, obtaining the gated loop unit network prediction model IPSO-GRU optimized by the improved particle swarm optimization, and otherwise, returning to the step D3 to continuously perform iterative updating.
Prediction obtained in said step EThe result uses the average relative percentage error MAPE, the root mean square error RMSE, the prediction accuracy FA and the decision coefficient R2Model evaluation was performed for four evaluation indexes, which are shown in equations (8) to (11):
Figure GDA0003534107140000084
Figure GDA0003534107140000085
Figure GDA0003534107140000086
Figure GDA0003534107140000087
in the above formula: n represents the sample capacity of the test data set; xact(i) And Xpred(i) i is 1,2 and … n are respectively the real value and the predicted value of the load at the ith moment;
Figure GDA0003534107140000088
an average value representing the true value of the prediction sample; wherein the smaller the MAPE and RMSE values are, the larger the FA value is, and R is2The closer to 1 the value of (a) indicates the greater the goodness of fit, the more accurate the model prediction result.
Examples
Aiming at the defects that factors influencing the load change of a power grid have the characteristics of complexity and time sequence and key parameters are selected according to experience in the conventional machine learning prediction method, the embodiment provides a short-term power load prediction method for optimizing a gated cyclic unit network (CNN-IPSO-GRU) mixed model based on a convolutional neural network and an improved particle swarm optimization algorithm, and the method comprises the steps of firstly extracting multi-dimensional feature vectors representing the load change by using the convolutional neural network, and constructing a time sequence to be input into a gated cyclic unit network model; and then, carrying out iterative optimization on the hyper-parameters (the number of neurons in a hidden layer and the learning rate) in the gating cycle unit model by using an improved particle swarm algorithm, obtaining optimal parameters on the premise of highest prediction precision, and finally completing the short-term load prediction of the power grid.
The sample data set uses power grid load and meteorological data of 2 years in total from 1 month and 1 day of 2013 to 12 months and 31 days of 2014 in a certain area in China, the data acquisition period is 15min, 96 points are acquired in one day, the power grid load and the meteorological data of 1 month to 6 months of 2013 to 2014 are used as a training data set, and the sample data is used for prediction under different models in 12 months of 2014 to verify the short-term power load prediction method based on the convolutional neural network and the improved particle swarm optimization gated cycle unit network model.
In the process of optimizing the gated loop unit network model by using the improved particle swarm optimization, the settings of model parameters are shown in table 1.
TABLE 1 optimization of parameters of gated cyclic unit network model by improved particle swarm optimization
Parameter(s) Value taking Parameter(s) Value taking
Number of iterations T max 100 Maximum value of inertial weight ωmax 0.9
Population size ps 50 Minimum value of inertial weight ωmin 0.4
Value range of m [1,32] Acceleration factor c1Initial value 0.5
m search ranges [-4,4] Acceleration factor c1Final value of 2.5
Range of values of epsilon [0.0001,0.01] Acceleration factor c2Initial value 0.5
Epsilon search range [-0.001,0.001] Acceleration factor c2Final value of 2.5
Fig. 1, fig. 2, and fig. 3 show the rule that the fitness of the particle, the number of neurons in the hidden layer, and the learning rate change with the change of the number of iterations, respectively, and it can be seen that the fitness of the PSO-GRU and the IPSO-GRU particle finally stabilizes at 2.236 and 1.513 with the change of the number of iterations; the number of neurons in the 2-layer hidden layer, m1 and m2, is finally stabilized at 8 and 16 respectively; the learning rate is finally stabilized at 0.0022 along with the change of the iteration times.
The CNN-IPSO-GRU model, CNN-PSO-GRU, CNN-GRU, LSTM and RNN model provided by the invention are subjected to load prediction by combining evaluation indexes, and the evaluation indexes of load prediction in continuous half years are shown in table 2. The analysis shows that: compared with CNN-GRU, LSTM and RNN prediction models, the proposed CNN-IPSO-GRU method reduces MAPE indexes by 3.80%, 12.79%, 31.84% and 37.94% respectively, reduces RMSE indexes by 8.47%, 16.61%, 40.65% and 43.24% respectively, and improves FA indexes by 0.09%, 0.33%, 1.06% and 1.39% respectively, thereby showing that the GRU model has better prediction effect in processing time sequence problems and reflecting the necessity of optimizing hyper-parameters by using a particle swarm algorithm; the CNN-IPSO-GRU model provided by the invention is further compared with the CNN-PSO-GRU prediction effect, wherein MAPE and RMSE are respectively reduced by 40.84 percent and 24.06 percent, and FA is improved by 0.92 percent, which shows that the prediction precision can be further improved by optimizing the hyper-parameters by using the improved particle swarm optimization, and the reliability is higher; finally, the effects of the 5 prediction models are quantitatively analyzed, and the calculation result of the decision coefficient R2 shows that: the R2 value of the CNN-IPSO-GRU prediction model provided by the invention is 1.007 which is closest to 1, namely the model prediction effect is best.
TABLE 2 comparison of different prediction models
Model (model) MAPE/％ RMSE/％ FA/％ R2
CNN-IPSO-GRU 1.318 106.021 98.682 1.007
CNN-PSO-GRU 2.228 139.610 97.772 0.989
CNN-GRU 2.316 152.532 97.684 1.012
GRU 2.555 167.413 97.445 1.014
LSTM 3.269 235.222 96.731 0.973
RNN 3.590 245.977 96.410 1.034
The size change curve of the load actual value and other model predicted values in a certain day 12 months 2014 is shown in fig. 4, the absolute percentage error change line graph of the corresponding predicted point is shown in fig. 5, and it can be seen from the graph that 5 prediction models have good prediction performance, but the CNN-IPSO-GRU model provided by the invention has the advantages of minimum prediction error, highest accuracy, better fitting capability for actual load data and basically consistent with the actual load change trend.
Table 3 is a summary table of the actual and predicted load values and absolute percentage errors monitored during the first 15min of each hour of the day for different algorithm model validation. It can be seen that the average absolute percentage error of the prediction result of the method provided by the invention at 24 monitoring points is 1.205%, the maximum absolute percentage error is 3.038%, and compared with CNN-PSO-GRU, CNN-GRU, LSTM and RNN models, the MAPE index is respectively reduced by 45.00%, 49.45%, 52.60%, 60.54% and 62.43%, which proves that the average error and the maximum error of the provided CNN-IPSO-GRU prediction method are superior to those of other models, and the prediction precision is higher.
TABLE 3 actual and predicted values and absolute percent error
Figure GDA0003534107140000111
The short-term power load prediction method for optimizing the gated cyclic unit network (CNN-IPSO-GRU) hybrid model based on the convolutional neural network and the improved particle swarm algorithm can quickly search and determine the optimal parameters of the gated cyclic unit network model, is high in training efficiency, solves the problems of insufficient model fitting capability and low prediction precision caused by parameter selection according to experience, further improves the load prediction precision, and plays an important role in reducing the loss of a generator set and ensuring the economic and reliable operation of a power grid.
The above description is only of the preferred embodiments of the present invention, and it should be noted that: it will be apparent to those skilled in the art that various modifications and adaptations can be made without departing from the principles of the invention and these are intended to be within the scope of the invention.

Claims (2)

1. A short-term power load prediction method based on a CNN-IPSO-GRU hybrid model is characterized by comprising the following steps:
A. collecting historical load, meteorological factors and date information data of a power grid as characteristic parameters;
B. acquiring sample data of the characteristic parameters in the step A in the historical time dimension power grid load, carrying out normalization processing on the sample data, extracting a multi-dimensional characteristic vector representing load change by using a Convolutional Neural Network (CNN), constructing the multi-dimensional characteristic vector into a time sequence, and dividing training set data and test set data according to the time sequence;
C. Constructing a gate control cycle unit network prediction model GRU comprising an input layer, a hidden layer and an output layer;
D. performing improved particle swarm optimization IPSO on the gated circulation unit network prediction model GRU constructed in the step C by using the training set data obtained in the step B, determining the optimal parameters of the gated circulation unit network prediction model GRU on the basis of meeting the optimal evaluation index of the prediction model, and reestablishing the optimized gated circulation unit network prediction model IPSO-GRU on the basis of the optimal parameters;
E. using the test set data obtained in the step B as an input variable of the optimized gated loop unit network prediction model IPSO-GRU in the step D, and further performing power grid short-term load prediction to obtain a prediction result;
the step B of normalizing the sample data means that the sample data is mapped between [0 and 1], and the normalized formula is expressed by formula (1):
Figure FDA0003534107130000011
in the formula (1), x*The normalized data is obtained; x is the number ofmin、xmaxRespectively the minimum value and the maximum value of the sample data set, wherein x is original sample data;
the concrete construction process of the gate control cycle unit network prediction model GRU in the step C is as follows:
c1, collecting power grid historical load, meteorological factor and date information data as characteristic parameters, extracting multi-dimensional characteristic vectors representing load changes by using a Convolutional Neural Network (CNN), and constructing a time sequence as sample data of a gate control cycle unit network prediction model GRU input layer;
C2, carrying out normalization processing on the sample data of each characteristic parameter in the step C1, mapping the sample data to the position between [0 and 1], and dividing the training set data and the test set data according to the time sequence;
c3, the hidden layer trains the gate control cycle unit network prediction model GRU by adopting the training set data obtained in the step C2, and the optimal parameters of the gate control cycle unit network prediction model GRU are determined under the condition that the evaluation standard error of the prediction result is smaller;
c4, adopting the gate control cycle unit network prediction model GRU obtained in the step C3 to predict the data of the test set, and obtaining a prediction result P1、P2、…Pn；
C5, calculating n prediction results obtained in the step C4 by an average absolute percentage error mode through an output layer, performing inverse normalization processing to obtain a final prediction result, and finally performing evaluation analysis on the prediction result by combining with an experimental evaluation index;
the process of optimizing the gated cyclic unit network prediction model GRU by using the improved particle swarm optimization IPSO in the step D is as follows:
d1, initializing GRU parameters of the gate control cycle unit network prediction model constructed in the step C, setting respective value ranges and search ranges of the number m of the neurons and the learning rate epsilon, and determining the maximum iteration times TmaxWith the population size ps and the maximum value omega of the inertial weight omega maxWith a minimum value ωminAcceleration factor c1、c2Maximum and minimum values of;
d2, performing model prediction on the test set data by using the gated cycle unit network prediction model GRU constructed in the step C according to the initialized neuron number m and the learning rate epsilon in the step D1, and taking the average absolute percentage error of the obtained prediction result as the fitness value of the particles, wherein the fitness function f is a fitness functionitiAs defined in formula (2):
Figure FDA0003534107130000021
wherein n represents the sample volume of the test set data; xact(i) And Xpred(i) Respectively a real value and a predicted value of the load at the moment i;
d3, taking two parameters of the number m of the neurons and the learning rate epsilon as particles, taking the average absolute percentage error of the prediction result obtained in the step D2 as a particle fitness value, and iteratively updating the speeds and the positions of the two particles by adopting formulas (3) to (7), namely updating the values of two key parameters of the gated loop unit network prediction model GRU; after the optimal values of the two parameters are obtained by improving particle swarm optimization IPSO, an optimized gated cyclic unit network prediction model IPSO-GRU is reestablished on the basis of the optimal values;
Figure FDA0003534107130000022
Figure FDA0003534107130000023
Figure FDA0003534107130000024
Figure FDA0003534107130000025
Figure FDA0003534107130000031
in the above formula, k represents the current iteration number; t ismaxRepresenting the maximum iteration number;
Figure FDA0003534107130000032
respectively representing the speed, the position, the individual local optimal solution and the global optimal solution of the particle; r is 1And r2Is [0, 1 ]]A random number in between; omega (k), c1(k) And c2(k) Respectively representing the values of inertia weight and acceleration factor in the k iteration; omegamax＝0.9，ωmin＝0.4；TmaxIs the maximum iteration number; c. C1And c2∈[0.5，2.5]；a＝b＝1，c＝d＝1.5；
D4, iteratively updating the speed and the position of the particles according to formulas (3) - (7) of the improved particle swarm optimization, calculating the corresponding particle fitness value, comparing the local optimal solution with the global optimal solution, meeting a termination condition when the particle fitness value tends to be stable or the iteration frequency reaches the maximum, obtaining the optimal neuron number m and the learning rate epsilon parameter, obtaining the gated loop unit network prediction model IPSO-GRU optimized by the improved particle swarm optimization, and otherwise, returning to the step D3 to continuously perform iterative updating.
2. The CNN-IPSO-GRU hybrid model-based short-term power load prediction method of claim 1, wherein: the prediction result obtained in the step E uses the average relative percentage error MAPE, the root mean square error RMSE, the prediction precision FA and the decision coefficient R2Model evaluation was performed for four evaluation indexes, which are shown in equations (8) to (11):
Figure FDA0003534107130000033
Figure FDA0003534107130000034
Figure FDA0003534107130000035
Figure FDA0003534107130000036
in the above formula: n represents the sample capacity of the test data set; xact(i) And Xpred(i) Respectively an actual value and a predicted value of the load at the ith moment, wherein i is 1,2, … n;
Figure FDA0003534107130000037
An average value representing the true value of the prediction sample; wherein the smaller the MAPE and RMSE values are, the larger the FA value is, and R is2The closer to 1 the value of (a) indicates the greater the goodness of fit, the more accurate the model prediction result.
CN202010573272.6A 2020-06-22 2020-06-22 Short-term power load prediction method based on CNN-IPSO-GRU hybrid model Active CN111738512B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202010573272.6A CN111738512B (en) 2020-06-22 2020-06-22 Short-term power load prediction method based on CNN-IPSO-GRU hybrid model

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202010573272.6A CN111738512B (en) 2020-06-22 2020-06-22 Short-term power load prediction method based on CNN-IPSO-GRU hybrid model

Publications (2)

Publication Number Publication Date
CN111738512A CN111738512A (en) 2020-10-02
CN111738512B true CN111738512B (en) 2022-05-24

Family

ID=72650305

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202010573272.6A Active CN111738512B (en) 2020-06-22 2020-06-22 Short-term power load prediction method based on CNN-IPSO-GRU hybrid model

Country Status (1)

Country Link
CN (1) CN111738512B (en)

Families Citing this family (31)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN112241811B (en) * 2020-10-20 2024-04-16 浙江大学 Layered hybrid performance prediction method for customized product in Internet +' environment
CN112232577B (en) * 2020-10-23 2024-04-16 浙江八达电子仪表有限公司 Power load probability prediction system and method for multi-core smart meter
CN112348271B (en) * 2020-11-12 2024-01-30 华北电力大学 Short-term photovoltaic power prediction method based on VMD-IPSO-GRU
CN112396152A (en) * 2020-11-17 2021-02-23 郑州大学 Flood forecasting method based on CS-LSTM
CN112737109B (en) * 2020-12-18 2022-05-31 北京国电通网络技术有限公司 Real-time safety early warning system of smart power grids
CN112819221A (en) * 2021-01-29 2021-05-18 江苏大学 Grain condition prediction method based on IPSO-GRU network
CN112819224B (en) * 2021-01-29 2021-10-08 南京工程学院 Unit output prediction and confidence evaluation method based on deep learning fusion model
CN112836431A (en) * 2021-02-05 2021-05-25 河北工业大学 Penicillin fermentation process fault prediction method based on PSO-LSTM
CN112990554A (en) * 2021-02-23 2021-06-18 中国大唐集团科学技术研究院有限公司华中电力试验研究院 Method for predicting short-term voltage trend of generator outlet PT fuse in slow melting
CN113191544B (en) * 2021-04-28 2022-08-12 河海大学 Short-term power load prediction method based on multilayer neural network
CN113222281A (en) * 2021-05-31 2021-08-06 国网山东省电力公司潍坊供电公司 Power distribution network short-term load prediction method and device based on improved AlexNet-GRU model
CN113469412B (en) * 2021-06-02 2024-04-09 国核电力规划设计研究院有限公司 Real-time operation strategy optimization method and system for comprehensive energy system
CN113554466B (en) * 2021-07-26 2023-04-28 国网四川省电力公司电力科学研究院 Short-term electricity consumption prediction model construction method, prediction method and device
CN113516316B (en) * 2021-07-29 2023-03-24 昆明理工大学 Attention-GRU short-term load prediction method based on sparrow search optimization
CN113837894B (en) * 2021-08-06 2023-12-19 国网江苏省电力有限公司南京供电分公司 Non-invasive resident user load decomposition method based on residual convolution module
CN113610328A (en) * 2021-08-25 2021-11-05 浙江浙能台州第二发电有限责任公司 Power generation load prediction method
CN113705094B (en) * 2021-08-27 2024-01-23 江苏杰瑞信息科技有限公司 PSO-GRU-based ship fuel pipeline fault prediction method
CN113688770B (en) * 2021-09-02 2024-02-06 重庆大学 Method and device for supplementing long-term wind pressure missing data of high-rise building
CN113837480B (en) * 2021-09-29 2023-11-07 河北工业大学 Impact load prediction method based on improved GRU and differential error compensation
CN114037145B (en) * 2021-11-05 2022-10-28 河北师范大学 Network security situation prediction method and system
CN113762486B (en) * 2021-11-11 2022-04-08 中国南方电网有限责任公司超高压输电公司广州局 Method and device for constructing fault diagnosis model of converter valve and computer equipment
CN114282614B (en) * 2021-12-27 2023-09-26 淮阴工学院 Medium-long runoff prediction method for optimizing CNN-GRU based on random forest and IFDA
CN114707772B (en) * 2022-06-06 2022-08-23 山东大学 Power load prediction method and system based on multi-feature decomposition and fusion
CN115270921B (en) * 2022-06-22 2023-05-02 天纳能源科技（上海）有限公司 Power load prediction method, system and storage medium based on combined prediction model
CN115049038A (en) * 2022-07-01 2022-09-13 科大智能电气技术有限公司 Optimization method, system, equipment and readable storage medium for park load prediction model
CN115440317B (en) * 2022-07-20 2023-05-26 中南大学 Method and device for predicting uranium leaching concentration, electronic equipment and storage medium
CN115271463B (en) * 2022-07-29 2023-06-02 昂顿科技（上海）有限公司 Intelligent energy management system
CN115967077A (en) * 2022-08-30 2023-04-14 湖南农业大学 Short-term prediction power load prediction method and system
CN115511191A (en) * 2022-09-30 2022-12-23 固远晨通科技发展有限公司 High and steep slope displacement point prediction method
CN117540265A (en) * 2023-10-09 2024-02-09 国网浙江省电力有限公司紧水滩水力发电厂 Concrete dam deformation prediction method based on IPSO-transducer model
CN117391260A (en) * 2023-12-11 2024-01-12 山东科技大学 Dissolved oxygen prediction method for optimizing convolutional neural network parameters by Kelvin optimization algorithm

Citations (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN107578124A (en) * 2017-08-28 2018-01-12 国网山东省电力公司电力科学研究院 The Short-Term Load Forecasting Method of GRU neutral nets is improved based on multilayer
CN108288115A (en) * 2018-03-15 2018-07-17 安徽大学 A kind of daily short-term express delivery amount prediction technique of loglstics enterprise
CN109345019A (en) * 2018-10-10 2019-02-15 南京邮电大学 A kind of micro-capacitance sensor economic load dispatching optimisation strategy based on improvement particle swarm algorithm
CN110596492A (en) * 2019-09-17 2019-12-20 昆明理工大学 Transformer fault diagnosis method based on particle swarm optimization random forest model
CN110751318A (en) * 2019-09-26 2020-02-04 上海电力大学 IPSO-LSTM-based ultra-short-term power load prediction method
CN111191841A (en) * 2019-12-30 2020-05-22 润联软件系统(深圳)有限公司 Power load prediction method and device, computer equipment and storage medium

Family Cites Families (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20200160176A1 (en) * 2018-11-16 2020-05-21 Royal Bank Of Canada System and method for generative model for stochastic point processes
US11159408B2 (en) * 2019-06-25 2021-10-26 Intel Corporation Link performance prediction technologies

Patent Citations (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN107578124A (en) * 2017-08-28 2018-01-12 国网山东省电力公司电力科学研究院 The Short-Term Load Forecasting Method of GRU neutral nets is improved based on multilayer
CN108288115A (en) * 2018-03-15 2018-07-17 安徽大学 A kind of daily short-term express delivery amount prediction technique of loglstics enterprise
CN109345019A (en) * 2018-10-10 2019-02-15 南京邮电大学 A kind of micro-capacitance sensor economic load dispatching optimisation strategy based on improvement particle swarm algorithm
CN110596492A (en) * 2019-09-17 2019-12-20 昆明理工大学 Transformer fault diagnosis method based on particle swarm optimization random forest model
CN110751318A (en) * 2019-09-26 2020-02-04 上海电力大学 IPSO-LSTM-based ultra-short-term power load prediction method
CN111191841A (en) * 2019-12-30 2020-05-22 润联软件系统(深圳)有限公司 Power load prediction method and device, computer equipment and storage medium

Non-Patent Citations (6)

* Cited by examiner, † Cited by third party
Title
Air Quality Prediction in Smart Cities Using Machine Learning Technologies Based on Sensor Data: A Review;Ditsuhi Iskandaryan 等;《Applied Sciences》;20200401;第10卷(第7期);第1-32页 *
基于PSO算法优化GRU神经网络的短期负荷预测;王康 等;《广东电力》;20200430;第33卷(第4期);第90-96页 *
基于注意力机制的CNN-GRU短期电力负荷预测方法;赵兵 等;《电网技术》;20190123;第43卷(第12期);第4370-4376页 *
基于粒子群优化-长短期记忆网络模型的变压器油中溶解气体浓度预测方法;刘可真 等;《电网技术》;20200410;第44卷(第7期);第2778-2784页 *
短期电力负荷预测模型及其应用研究;秦光宇 等;《价格理论与实践》;20200225(第2期);第75-78,174页 *
融合快速信息交流和局部搜索的粒子群算法;姜建国 等;《哈尔滨工程大学学报》;20150531;第36卷(第5期);第687-691页 *

Also Published As

Publication number Publication date
CN111738512A (en) 2020-10-02

Similar Documents

Publication Publication Date Title
CN111738512B (en) Short-term power load prediction method based on CNN-IPSO-GRU hybrid model
CN106446540B (en) A kind of Wind turbines health status real time evaluating method
CN111027772B (en) Multi-factor short-term load prediction method based on PCA-DBILSTM
CN112990556A (en) User power consumption prediction method based on Prophet-LSTM model
CN111860982A (en) Wind power plant short-term wind power prediction method based on VMD-FCM-GRU
CN112488395A (en) Power distribution network line loss prediction method and system
CN110751318A (en) IPSO-LSTM-based ultra-short-term power load prediction method
CN112232561B (en) Power load probability prediction method based on constrained parallel LSTM fractional regression
CN110163429A (en) A kind of short-term load forecasting method based on similar day optimal screening
CN111832839B (en) Energy consumption prediction method based on sufficient incremental learning
CN112434848A (en) Nonlinear weighted combination wind power prediction method based on deep belief network
CN116596044B (en) Power generation load prediction model training method and device based on multi-source data
CN111525587A (en) Reactive load situation-based power grid reactive voltage control method and system
CN113139596A (en) Optimization algorithm of low-voltage transformer area line loss neural network
CN110987436A (en) Bearing fault diagnosis method based on excitation mechanism
CN116187835A (en) Data-driven-based method and system for estimating theoretical line loss interval of transformer area
CN112288157A (en) Wind power plant power prediction method based on fuzzy clustering and deep reinforcement learning
CN115600729A (en) Grid load prediction method considering multiple attributes
CN116757057A (en) Air quality prediction method based on PSO-GA-LSTM model
CN115358437A (en) Power supply load prediction method based on convolutional neural network
CN113033898A (en) Electrical load prediction method and system based on K-means clustering and BI-LSTM neural network
Chen et al. Short-term Wind Speed Forecasting Based on Singular Spectrum Analysis, Fuzzy C-Means Clustering, and Improved POABP
CN109697531A (en) A kind of logistics park-hinterland Forecast of Logistics Demand method
CN113449919A (en) Power consumption prediction method and system based on feature and trend perception
CN113762591B (en) Short-term electric quantity prediction method and system based on GRU and multi-core SVM countermeasure learning

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