CN-115392571-B - Short-term load prediction method for optimizing deep extreme learning machine by improving whale algorithm

CN115392571BCN 115392571 BCN115392571 BCN 115392571BCN-115392571-B

Abstract

A short-term load prediction method for optimizing a deep extreme learning machine by an improved whale algorithm is characterized in that a Tent chaotic map is introduced to initialize an initial whale population, the deep extreme learning machine is used as a basic load prediction model to mine deep hidden information of data, the improved whale algorithm is used for carrying out parameter optimization, and finally, factors such as temperature, humidity and the like are considered to have great influence on load change, so that a multidimensional IWOA-DELM load prediction model is established. The method can solve the problem that initial population distribution of a whale algorithm is not wide enough, enriches model input quantity by considering influence of climate factors on load values, and improves short-term load prediction accuracy.

Inventors

LIU JILONG
WANG JI
CHENG XIAOYAN
LI ZHENDONG
LIU XIN
Lv Chunhui
ZHANG YAN
Hou kunming
JIA XUAN
CHEN YAXIAO
JIANG YIKANG
DONG LILI
WU MENG
Yan tengfei

Assignees

国网山东省电力公司聊城供电公司
国家电网有限公司

Dates

Publication Date: 20260505
Application Date: 20220825

Claims (6)

1. A short-term load prediction method for improving whale algorithm optimization deep extreme learning machine is characterized by comprising the following steps: Dividing load data into a training set and a testing set in a time scale, wherein the training set is used as a training sample of a load prediction model, the testing set is used as a testing sample for obtaining the prediction precision of the load prediction model according to training of the training sample, and the data in the training set and the testing set are normalized; Setting parameters of a whale algorithm, namely, whale population scale, iteration times, variable dimension, variable upper limit and variable lower limit, randomly initializing position values of all whale individuals by using a method for initializing parameters through Tent chaotic mapping shown in a formula (2), wherein the position value of each whale individual represents initial input weight of a deep extreme learning machine model; Wherein X n is the initial position of the nth whale, X n+1 is the initial position of the (n+1) th whale, a is a constant of [0,1], the return value of X n+1 is 1 if X n+1 >1, and the return value of X n+1 is 0 if X n+1 < 0; constructing a deep extreme learning machine model, pre-training the deep extreme learning machine by using a training set, and selecting the root mean square error of the model training set as an objective function to calculate the fitness value of each whale individual; Sequencing the fitness value of each whale individual obtained in the third step, taking the position of the whale individual with the smallest fitness value as the optimal position of the iteration, driving other whale individuals to change positions to move to the optimal position so as to surround the prey, attacking the prey according to the working principle of a subsequent whale algorithm, and randomly searching the next prey; And fifthly, repeating the third step and the fourth step until the maximum iteration times set in the second step are reached, taking the last iteration result as the optimal initial input weight, and introducing the optimal initial input weight into the deep extreme learning machine model established in the third step, so that a load prediction result can be obtained.
2. The method for short-term load prediction for improved whale algorithm optimized deep extreme learning machine of claim 1 wherein said data in said training and test sets of step one comprises load data, humidity, maximum daily temperature, minimum daily temperature, average daily temperature.
3. The method for predicting short-term load in a deep learning machine optimized by a whale algorithm according to claim 1, wherein the samples in the training set in the first step are the first 90% of the load data divided by time scale, and the samples in the test set are the last 10% of the load data.
4. The method for improving short-term load prediction of a whale algorithm optimized deep extreme learning machine according to claim 1, wherein the normalization method in the first step is as shown in formula (1): Where Y i is any value in the dataset that needs to be normalized, Y is the value after normalization by Y i , Y min is the minimum value in the selected sample range, and Y max is the maximum value in the selected sample range.
5. The method for predicting short-term load of improved whale algorithm optimized deep extreme learning machine according to claim 1, wherein the deep extreme learning machine model in the third step is a multi-level neural network structure composed of self-encoders of the extreme learning machine, and each layer is an ELM structure of the extreme learning machine: Assuming that there are N different sets of input-output samples (x i ,t i ), where input sample x i ＝[x i1 ,…,x in ] T ∈R n , output sample t i ＝[t i1 ,…,t im ] T ∈R m , then ELM with L hidden nodes, excitation function G (x), is represented by equation (3): Wherein β i ＝[β i1 ,β i2 ,…,β im ] T is the output weight of the i-th hidden layer node, a i ＝[a i1 ,a i2 ,…,a in ] T is the input weight of the i-th hidden layer node, b i is the bias of the i-th hidden layer node, and G (a i ,b i ,x j ) is the output of the i-th hidden layer; the matrix form of equation (3) is shown in equation (4): Where a i and b i are randomly set, β i , which enables the ELM algorithm to output the optimal solution, can be calculated from H and T, as shown in equation (5): β=H + T (5) Wherein H + is the generalized inverse of H; the solution of the orthogonal mapping method is adopted to obtain: β=(H T H) -1 HT (6) And introducing a regularization coefficient C, wherein the solution of the regularization coefficient C is shown in the formula (7): Wherein I is an identity matrix; Since the extreme learning machine self-encoder features an input equal to an output and both weights and offsets are orthogonalized, the weight β can be converted from equation (7) to equation (8): wherein X is the input matrix of each layer of self-encoder; The first layer of the model adopts original data to solve to obtain an output weight matrix beta 1 , in the extraction process, the input weight matrix W i of each layer is a transposed matrix of beta i , and then for each hidden layer of DELM, H i-1 of the previous layer is used as an input matrix of the next layer until the last layer.
6. The method for predicting short-term load by optimizing deep extreme learning machine by using improved whale algorithm according to claim 1, wherein in said step three, the root mean square error RMSE of the model training set is selected as the objective function to calculate the fitness value of each whale individual as shown in formula (9): Where n is the number of samples, y i is the load predictor, Is the actual value of the load.

Description

Short-term load prediction method for optimizing deep extreme learning machine by improving whale algorithm [ Field of technology ] The invention relates to application of a deep extreme learning machine in short-term load prediction, in particular to a short-term load prediction method for optimizing the deep extreme learning machine by improving whale algorithm. [ Background Art ] The electric power system is a skeleton supporting industrial development across various industries, and each development innovation of the electric power system will bring about the progress of the industrial industry. Along with the proposal of the 'double carbon' target, large-scale clean energy is connected into a power grid, and the power industry is transformed into the 'low carbon' and 'green' directions. Meanwhile, under the pressure of resource shortage and environmental problems, the power system needs to find the balance between a supply end and a consumption end, namely the balance between the generated energy and the used energy, for the stability, the safety and the energy conservation of the power system. Therefore, when a power enterprise makes a power generation plan, load prediction is required to predict the load amount of a future user side as an important reference. Load prediction in a broad sense can be classified into ultra-short-term load prediction, medium-term load prediction, and long-term load prediction according to the length of time. In principle, no matter what load prediction is, future load quantity is predicted according to historical load data and by referring to factors such as day type, climate and the like to a certain extent. The accurate load prediction is a foundation for load management of the power enterprises and a corresponding foundation for demand sides of the power enterprises, and has great significance for improving the electric energy quality, reducing the power generation cost and ensuring the economic and safe operation of the power grid. Short-term load prediction is to predict the load amount at a fixed time point (most of the whole point) for several hours or days in the future, and is an important link in power system load prediction. The traditional short-term load prediction method mainly adopts the traditional mathematical method, the prediction precision is generally not ideal, and some artificial intelligence-based prediction methods cannot be widely used in short-term load prediction because of large prediction precision fluctuation. Based on this, it is necessary to propose a load prediction method with high prediction accuracy. [ Invention ] The invention aims to provide a short-term load prediction method for improving a whale algorithm optimized deep extreme learning machine, which aims to solve the problem of low load prediction precision of the existing method and is a simple and easy-to-implement short-term load prediction method. The technical scheme of the invention is that the short-term load prediction method for improving whale algorithm optimization depth extreme learning machine is characterized by comprising the following steps: Dividing load data into a training set and a testing set in a time scale, wherein the training set is used as a training sample of a load prediction model, the testing set is used as a testing sample for obtaining the prediction precision of the load prediction model according to training of the training sample, and the data in the training set and the testing set are normalized; the data in the training set and the testing set in the first step comprise load data, humidity, daily maximum temperature, daily minimum temperature and daily average temperature. The samples in the training set in the first step are the first 90% of the data of the load data after time scale division, and the samples in the test set are the last 10% of the data. The normalization method in the first step is shown in the formula (1): Where Y i is any value in the dataset that needs to be normalized, Y is the value after normalization by Y i, Y min is the minimum value in the selected sample range, and Y max is the maximum value in the selected sample range. Setting parameters of a whale algorithm, namely, whale population scale, iteration times, variable dimension, variable upper limit and variable lower limit, randomly initializing position values of all whale individuals by using a method for initializing parameters through Tent chaotic mapping shown in a formula (2), wherein the position value of each whale individual represents initial input weight of a deep extreme learning machine model; Wherein X n is the initial position of the nth whale, X n+1 is the initial position of the (n+1) th whale, a is a constant of [0,1], the return value of X n+1 is 1 if X n+1 >1, and the return value of X n+1 is 0 if X n+1 < 0; constructing a deep extreme learning machine model, pre-training the deep extreme learning machine by using a training set, and selectin