CN106709820A - Power system load prediction method and device based on deep belief network - Google Patents

Abstract

The embodiment of the invention provides a method and a device for predicting a load of a power system based on a deep belief network, relates to the field of power systems, and can improve convergence speed and reduce prediction errors. The specific scheme comprises the following steps: acquiring a training sample and a test sample; constructing an energy function of the RBM model; training the at least one hidden layer and the visible layer by utilizing the training sample to obtain the weight of the training sample between the nodes of the at least one hidden layer and the visible layer; and inputting the output data obtained by the training sample and the test sample into the trained DBN to obtain a predicted value of the power system load. The method is used for predicting the load of the power system.

Description

Power system load prediction method and device based on deep belief network

Technical Field

The embodiment of the invention relates to the field of power systems, in particular to a power system load prediction method and device based on a deep belief network.

Background

The load prediction of the power system is an important component of power system planning and also is the basis of the economic operation of the power system, and the load prediction of the power system fully considers the influence of relevant factors such as politics, economy, climate and the like from the known power demand to predict the future power demand. Accurate load prediction data is beneficial to power grid dispatching control and safe operation, reasonable power supply construction planning is formulated, and economic benefits and social benefits of a power system are improved. In load prediction, methods such as a regression model, a time series prediction technique, a gray theory prediction technique, and the like are often used. In recent years, with the development of artificial neural network research, the load of a power system is predicted by using a neural network, so that the prediction error is greatly reduced, and the attention of load prediction workers is drawn.

In the prior art, a traditional neural network is used for load prediction of a power system. However, a conventional neural network is a typical global approximation network, and one or more weights of the network have an effect on each output; on the other hand, the network has randomness when determining the weight, so that the relation between input and output is uncertain after each training, and the prediction result has difference. It can be seen that the prior art has the problems of slow convergence rate and large prediction error.

Disclosure of Invention

The embodiment of the invention provides a power system load prediction method and device based on a deep belief network, which can improve convergence speed and reduce prediction errors.

In order to achieve the above purpose, the embodiments of the present application adopt the following technical solutions:

in a first aspect, a method for predicting a load of a power system based on a DBN is provided, the DBN is composed of a multi-layer restricted boltzmann machine RBM and comprises at least one hidden layer and one visible layer, and the method for predicting the load of the power system comprises the following steps:

acquiring a training sample and a test sample; constructing an energy function of the RBM model;

training the at least one hidden layer and the visible layer by utilizing the training sample to obtain the weight of the training sample between the nodes of the at least one hidden layer and the visible layer;

and inputting the output data obtained by the training sample and the test sample into the trained DBN to obtain a predicted value of the power system load.

In a second aspect, a DBN-based power system load prediction apparatus is provided for executing the method provided in the first aspect.

According to the method and the device for predicting the load of the power system based on the deep confidence network, which are provided by the embodiment of the invention, the deep confidence network is formed by utilizing a multilayer limited Boltzmann mechanism and comprises a plurality of hidden layers and a visible layer, RBMs can be pre-trained in a layered manner by adopting a layered unsupervised greedy pre-training method, and the obtained result is used as an initial value of a probability model for supervised learning and training, so that the learning performance is greatly improved, the convergence speed is increased, and the prediction error is reduced.

Drawings

In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the embodiments will be briefly introduced below, and it is obvious that the drawings in the following description are only some embodiments of the present invention, and it is obvious for those skilled in the art that other drawings can be obtained based on these drawings without creative efforts.

Fig. 1 is a schematic flow chart of a method for predicting a load of an electrical power system based on a deep belief network according to an embodiment of the present invention;

FIG. 2 is a schematic diagram illustrating an RBM including a visible layer and an implicit layer;

FIG. 3 is a schematic diagram illustrating the optimal error recognition rate for three models with different structures according to an embodiment of the present invention;

FIG. 4 is a graph of average absolute percentage error as a function of iteration number for an embodiment of the present invention;

fig. 5 is a schematic structural diagram of a DBN-based power system load prediction apparatus according to an embodiment of the present invention.

Detailed Description

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 only a part of the embodiments of the present invention, and not all of the embodiments. 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.

Aiming at the problems of low load prediction convergence speed and large prediction error of the traditional neural Network, the invention provides a power system load prediction scheme based on a Deep Belief Network (Deep Belief Network, DBN for short), which is shown in a combined figure 1, and the method comprises the following steps:

101. training samples and test samples are obtained.

102. And constructing an energy function of the RBM model.

103. And training at least one hidden layer and one visible layer by utilizing the training samples to obtain the weight of the training samples between the nodes of the at least one hidden layer and the visible layer.

104. And inputting the output data obtained by the training sample and the testing sample into the trained DBN to obtain a predicted value of the power system load.

In the specific embodiment of the invention, the DBN is composed of a multilayer Restricted Boltzmann Machine (RBM), comprises at least one hidden layer and one visible layer, pre-trains the RBM in a layered manner by adopting a layered unsupervised greedy pre-training method, and takes the obtained result as the initial value of a probability model for supervised learning and training, thereby greatly improving the learning performance, which is illustrated as follows:

first, data preprocessing

S1, preprocessing the training sample:

let all training samples be X ═ X₁,X₂,...,X_MIn which X is_iRepresenting a set of load data, M representing the number of load data, and for the input samples, the algorithm takes 4 stepsAnd (3) carrying out normalization:

calculating an average value:

where u represents the mean of the samples.

Calculating variance:

where the variance of the sample is represented.

③ whitening:

wherein, X_i' denotes a whitening parameter of the sample data.

Normalization:

wherein, X_i,nA normalization parameter representing sample data;

s2, preprocessing the test sample:

the preprocessing of the test sample needs to whiten the test sample according to the mean value and the variance of the training sample, then the test sample is uniformly normalized to be between 0 and 1 according to the maximum value and the minimum value of the training sample, and if a single test sample is T, the normalization step is as follows:

whitening:

wherein, T' represents the whitening parameter of the test sample data, T is a single test sample, u represents the mean value of the training sample, and represents the variance of the training sample.

Normalization:

wherein, T_nRepresenting the normalized parameters of the test sample data.

Here, the reason for the whitening is that there is a large correlation between adjacent elements of the natural data, and thus the redundancy of the data can be reduced by the whitening, similarly to the dimensionality reduction of Principal Component Analysis (PCA).

Second, basic principle and constitution of deep confidence network

The DBN is an energy model, has very strong characteristic learning capability and is a deep network which is researched very early. The learning method is proposed by Hinton et al, and the essence is that more useful features are learned by constructing a machine learning model with a plurality of hidden layers and massive training data, so that the accuracy of classification or prediction is finally improved.

The DBN can be seen as a complex neural network formed by a plurality of layers of RBMs, which comprises a visible layer and a plurality of hidden layers, and the preset values of the RBMs are trained by adopting a contrast divergence method. The RBM is pre-trained in a layered manner by a layered unsupervised greedy pre-training method for the deep neural network, the obtained result is used as an initial value of a probability model for supervised learning training, and the learning performance is greatly improved.

RBM is a generative stochastic neural network used to learn probability distributions about input data, and can be used in both supervised and unsupervised ways as requiredTraining, which is widely used in many ways. As shown in fig. 2, the RBM has two layers of networks respectively represented as: vector h ═ h₁,h₂,…h_m) Representing hidden layer units, and having m elements in total; vector v ═ v₁,v₂,…,v_n) Representing a visible layer, for a total of n elements. Because the elements in the layer are not connected, the elements in the layer are independent of each other, namely:

p(h|v)＝p(h₁|v)p(h₂|v)…p(h_m|v)；

p(v|h)＝p(v₁|h)p(v₂|h)…p(v_n|h)。

in general, the distribution of visible layer and hidden layer units satisfies the bernoulli form, and then:

it can be seen that the probability distribution of the entire model can be obtained from the probability distribution of each layer, and unsupervised training can be performed layer by layer. And taking the output of the lower layer RBM as the input of the upper layer RBM. And inputting the last hidden layer to a softmax classifier until the last hidden layer is reached, and then calculating an estimation error for reverse fine adjustment.

S3, constructing an energy function of the RBM model:

the energy function of the RBM model satisfying bernoulli distribution for the hidden and visible layers is defined as:

wherein, a_iAnd b_jRepresents a bias term, w_ijAnd representing the connection weight between the visible unit and the hidden layer unit. θ ═ is (a, b, w) weight parameters of the RBM model, and the vector h ═ is (h)₁,h₂,…h_m) Represents a hidden layer unit, has m elements in total, and has a vector v ═ v₁,v₂,…,v_n) Representing a visible layer, for a total of n elements.

S4, calculating the joint probability distribution of the visible layer unit and the hidden layer unit:

given the model parameters, the joint probability distribution of visible and hidden layer units can be expressed as:

wherein,is a normalization factor, -E (v, h) is the energy function of the RBM model for hidden and visible layers, and the vector h ═ h₁,h₂,…h_m) The hidden layer unit is shown, and the total number of the hidden layer unit is m.

From equations (7) and (8) it can be deduced:

in the formula,

s5, calculating the edge distribution of the visible layer and the hidden layer:

according to the joint probability distribution P (v, h) of the visible layer and the hidden layer, an edge distribution can be obtained:

wherein,is a normalization factor, E (v, h) is the energy function of the RBM model for hidden and visible layers, and the vector h ═ h₁,h₂,…h_m) The hidden layer unit is shown, and the total number of the hidden layer unit is m.

S6, constructing a log-likelihood function as follows:

where l is the number of training samples, θ ═ is the weight parameter of the RBM model, P (v, w)_i) The edge distribution of the visible layer and the hidden layer.

S7, calculating the gradient of the log-likelihood function:

the logarithm likelihood function is derived according to a gradient descent method, and the following can be obtained:

finishing to obtain:

where lnL (θ) is the log-likelihood function, l is the number of training samples, θ ═ a, b, w are the weight parameters of the RBM model, and P (h | v |, w) is_i) Is the edge distribution of the visible layer and the hidden layer, E (v, h) is the energy function of the RBM model of the hidden layer and the visible layer, P (h, v)_i) Is the joint probability distribution of the visible layer and hidden layer units.

Thirdly, predicting the load of the power system by using the deep confidence network

For deep learning, the idea is to stack multiple layers, i.e., the output of one layer is used as the input for the next layer. For example, there is a system S having n layers (S1, … Sn) with I as input and O as output, which can be expressed as: I-S1-S2- … -Sn-O, a series of hierarchical features that result from the input I, i.e., S1, …, Sn, may be automatically obtained by adjusting parameters in the system so that the output O remains the input I. In this way, hierarchical representation of the input information can be achieved. The process of predicting the load of the power system by using the deep belief network is essentially a deep learning process.

The process of predicting the load of the power system by using the deep confidence network comprises the following steps: and training the preset value of the RBM by adopting a divergence contrast method. Through analysis, the algorithm adopts three hidden layers and adopts a hierarchical unsupervised greedy pre-training method to pre-train RBMs in a hierarchical mode, and the obtained result is used as an initial value of a probability model for supervised learning training, so that the learning performance is improved. Simulation data shows that compared with the traditional neural network algorithm, the algorithm has the advantages of fast convergence and small prediction error.

Firstly, training a preset value of the RBM by adopting a divergence contrast method. KL-divergence is an asymmetric measure, so the values of KL (Q | | P) and KL (P | | Q) are different. The larger the difference between the two distributions, the larger the value of KL-divergence. The maximum value of the log-likelihood function of the RBM finally evolves to calculate the difference between KL-divergence of two probability distributions:

KL(Q||P)-KL(P_k| | P) (formula 15)

Where Q is the prior distribution, P_kIs the distribution after k steps of Gibbs sampling. If the Gibbs chain reaches a plateau (due to the initial state v)⁽⁰⁾Already steady state, v), then there is P_kP, i.e. KL (P)_kP) | 0. The estimation error obtained by the contrast divergence algorithm is then equal to 0.

Secondly, the obtained result is used as an initial value of a probability model for supervised learning training, and the relative entropy is also called KL-divergence to measure the distance between two probability distributions. The KL-divergence of the two probability distributions Q and P in the state space is defined as follows:

KL (Q | | P) is the KL-divergence of the two probabilities Q and P distributed in the state space.

And thirdly, the obtained preset value and the initial value are used as input data and transmitted to the next layer, and training is carried out in the next layer. Training one layer of network each time, and training layer by layer. Specifically, a first layer is trained by using calibration-free data, parameters of the first layer are learned during training (the layer can be regarded as a hidden layer of a three-layer neural network which enables the difference between output and input to be minimum), after an n-1 th layer is obtained through learning, the output of the n-1 th layer is used as the input of the n-th layer, and the n-th layer is trained, so that the parameters of each layer are obtained respectively. Thus, the training process is a repetitive iterative process. And when all layers are trained, tuning is carried out by using a wake-sleep algorithm.

A wake stage: the cognitive process, which produces an abstract representation (node state) of each layer by the features of the outside world and the upward weights (cognitive weights), and modifies the downward weights (generating weights) between the layers using gradient descent.

sleep stage: and a generation process, namely generating the state of the bottom layer through the representation of the top layer and the downward weight, and simultaneously modifying the upward weight between the layers.

After pre-training, the DBN can adjust the discrimination performance by using the BP algorithm through utilizing the labeled data. Here, a tag set is added to the top layer (promote associative memory), and a classification surface of the network is obtained through a bottom-up, learned recognition weight. This performance may be better than a network trained solely by the BP algorithm. This can be interpreted intuitively, and the BP algorithm of DBNs only needs to search a local weight parameter space, which is faster in training and shorter in convergence time than the forward neural network.

After the training and tuning steps are finished, the weights are connected together at the highest two layers (namely the last hidden layer and the last visible layer), so that the output of the lower layer provides a reference clue or association to the top layer, and the top layer can contact the top layer with the memory content of the top layer. Thus, the load value of the power system is predicted.

In a specific embodiment, S8: calculating KL-divergence of the distribution of two probabilities Q and P of the visible layer and the hidden layer in a state space according to the formula 16; s9: and finally evolving the maximum value of the log-likelihood function of the RBM model into calculating the KL-divergence difference of the two probability distributions Q and P according to the formula 15. S10: and S1-S9 are used for carrying out layered training on training samples through an RBM model to obtain weights of the training samples between nodes of a visible layer and nodes of a hidden layer, the weights are input into the visible layer from the hidden layer, the visible layer can memorize the content of the weights and obtain a load predicted value of the training samples, and the load predicted value is used for the load prediction process of the power system. And S11, adding the average load value, the maximum load value, the average air temperature and the predicted daily average air temperature of the training sample obtained in the step S10 to the test sample as input data of the deep neural network algorithm, and outputting the data, namely the predicted load value of the power system, through calculation of S1-S9.

Fourth, the experimental setup

The load test uses load data of east silovack electric power company, 2007, from 1 month to 10 months, as training data, and data of 11 months to 12 months, as prediction comparison data, provided by the EUNITE competition. Wherein the input data of the neural network comprises: the previous day 48 sampling points (average load data every half hour), the average temperature of the previous day, the load peak of the previous day, the load valley of the previous day, the average value of the previous day, the predicted average temperature of the day. The prediction algorithm predicts 48 points of data the next day in one go.

In the application of load prediction, load prediction data within 0-24 hours every half hour is generally needed, 48 historical data of 0-24 points in a day before the prediction day are taken as algorithm, and the average load value, the maximum load value, the average air temperature and the average air temperature of the prediction day of the numerical weather forecast in the day before the prediction day are added to be used as input data of the deep neural network algorithm. The output data is load data every half hour from 0 to 24 points on the prediction day.

In order to compare the performance of the algorithm, a BP Neural Network, a Self-organizing Fuzzy Neural Network (hereinafter, referred to as Self-organizing Fuzzy Neural Network) and a deep confidence Network are selected for comparison. The BP neural network has the characteristic of capturing nonlinear law well, and when the number of neurons in the hidden layer is enough, the 3-layer perceptron model can realize the approximation of any nonlinear function. The main advantage of the SOFNN algorithm is that it can automatically determine the network structure and give model parameters with good prediction accuracy.

The criterion is Mean Absolute Percent Error (MAPE), defined as follows:

wherein D is_iAndthe real value and the predicted value of the maximum load on the i th day of a certain month in 1997, and n is the number of days of a certain month in 1997.

Fifth, evaluation of Performance

(1) Effect of network architecture on algorithm Performance

Firstly, the influence of a network structure (hidden layer, namely the number of neurons) on a prediction effect is experimentally researched, load data of 1-10 months in 2007 are taken as training data, and load data of 11 months are taken as test data. The network structure uses 1, 2, 3 layers of hidden layers, respectively. The numbers of neurons were 20, 50, 100, 200, and 400, respectively. Here, the number of neurons in each layer is the same for convenience of comparison. For the initial parameter settings, the following parameters are selected manually from these ranges to obtain the optimal recognition rate: BP learning rate (0.1, 0.05, 0.02, 0.01, 0.005), pre-training learning rate (0.01, 0.005, 0.002, 0.001). FIG. 3 shows the optimal error recognition rate for different structures of the three models. As can be seen from fig. 3, when the number of hidden layers is small or the number of neurons is small, the performance of the BP network without pre-training is better, when only 1 hidden layer is provided, the error rate of the SOFNN is equivalent to that of the BP when the number of neurons reaches 200, and when the number of hidden layers is 2 layers and 3 layers, the performance of the SOFNN approaches and exceeds that of the BP when the number of neurons reaches 100 and 50. Similarly, it was found that the performance of SOFNN was superior to DBN when the number of neurons was small, and the opposite was true when the number of neurons was large. And when the number of implicit layers is the same, because of the over-fitting problem. The performance of the BP network does not increase along with the number of nodes, but decreases when the number of nodes is excessive, and the depth model is increased and tends to be stable all the time, which shows that the performance of the pre-trained depth network is gradually optimized along with the enlargement of the network scale (including the increase of the number of implicit layers or the increase of the number of nodes). It can be seen that too few number of hidden layers and number of hidden nodes may degrade the performance of the depth model. The reason can be explained in this way, the pre-training model is used for extracting core features in input features, and due to the limitation of sparsity conditions, the number of neurons is assumed to be too small, and for the input of some input samples, only a small number of neurons are activated, and the features cannot represent original input, so that some information amount is lost, and the performance is reduced. Although the performance of the depth model is better when the network size is larger, the training time is also lengthened, and therefore, the performance and the training time need to be balanced.

(2) Algorithm convergence comparison

Figure 4 is a plot of mean absolute percentage error versus number of iterations. The algorithm uses a model of three hidden layers, 100 neurons per layer, 100 per layer. The learning rate during pre-training is 0.01. The learning rate at fine tuning is 0.1, and the momentum term coefficient is 0.5. Load data of 1 month to 11 months in 2007 are taken as training data, and load data of 12 months is taken as test data. And comparing the average absolute percentage error changes of BP, SOFNN and DBN under 1000 iterations.

Analyzing fig. 4, it can be seen that the error rates of the three models gradually decrease as the number of iterations increases, because the distribution of the network parameters becomes closer to the minimum point as the number of training times increases. However, after a certain training frequency is reached, the error rate of the BP network oscillates and tends to increase gradually, and the error rate of the DBN network after pre-training is stably reduced, which indirectly proves that the pre-training can make the distribution area of the initial parameters of the network closer to the minimum value point, and effectively avoid local oscillation.

The load prediction provides important basis for power distribution network management decision and operation mode. The invention provides a load prediction algorithm adopting a deep belief network, and solves the problems of slow learning rate and low prediction efficiency of the existing neural network algorithm. Simulation results show that compared with a traditional neural network algorithm, the power load prediction effect of the algorithm based on the deep belief network is obvious.

An embodiment of the present invention further provides a DBN-based power system load prediction apparatus for performing the power system load prediction method described in the foregoing embodiment, and as shown in fig. 5, the power system load prediction apparatus includes:

a data processing unit 501, configured to obtain a training sample and a test sample; constructing an energy function of the RBM model;

a training unit 502, configured to train at least one hidden layer and one visible layer by layer using a training sample, to obtain a weight of the training sample between nodes of the at least one hidden layer and the visible layer;

and a prediction unit 503, configured to input the output data obtained by the training sample and the test sample into the trained DBN, so as to obtain a predicted value of the power system load.

Optionally, the data processing unit 501 is further configured to process the load data X_iFormed training sample X ═ X₁,X₂,...,X_MNormalizing; wherein X_iRepresenting a group of load data, and M representing the number of the load data; and the device is also used for whitening the test sample according to the mean value and the variance of the training sample, and uniformly normalizing the test sample to be between 0 and 1 according to the maximum value and the minimum value of the training sample.

Optionally, the visible layer and the hidden layer units are distributed to satisfy the bernoulli form;

the energy function of the RBM model satisfying bernoulli distribution for the visible layer and the hidden layer defined by the data processing unit 501 is:

wherein, a_iAnd b_jRepresents a bias term, w_ijAnd representing the connection weight between the visible unit and the hidden layer unit. θ ═ is (a, b, w) weight parameters of the RBM model, and the vector h ═ is (h)₁,h₂,…h_m) Represents a hidden layer unit, has m elements in total, and has a vector v ═ v₁,v₂,…,v_n) Representing a visible layer, for a total of n elements.

Optionally, the training unit 502 is specifically configured to train a preweight of the RBM by using a divergence contrast device, and use an obtained result as an initial value of the supervised learning training probability model; the obtained preset value and the initial value are used as input data and transmitted to a next layer of network, and training is carried out in the next layer of network; training a network layer by layer each time, and training at least one hidden layer and one visible layer by layer.

Optionally, the output data obtained from the training samples at least include an average load value, a maximum load value, an average air temperature, and a predicted daily average air temperature of a numerical weather forecast of the training samples.

According to the method and the device for predicting the load of the power system based on the deep belief network, which are provided by the embodiment of the invention, the deep belief network is formed by utilizing a multilayer limited Boltzmann mechanism and comprises a plurality of hidden layers and a visible layer, RBMs can be pre-trained in a layered mode by adopting a layered unsupervised greedy pre-training method, and the obtained result is used as an initial value of a probability model for supervised learning and training, so that the learning performance is greatly improved, the convergence speed is improved, the prediction error is reduced, and the problems of low learning rate and low prediction efficiency of the existing neural network algorithm are solved. Simulation results show that compared with a traditional neural network algorithm, the power load prediction effect of the algorithm based on the deep belief network is obvious.

The above description is only for the specific embodiments of the present invention, but the scope of the present invention is not limited thereto, and any person skilled in the art can easily conceive of the changes or substitutions within the technical scope of the present invention, and the changes or substitutions should be covered within the scope of the present invention. Therefore, the protection scope of the present invention shall be subject to the protection scope of the claims.

Claims (10)

1. A power system load prediction method based on a Deep Belief Network (DBN), wherein the DBN is composed of a plurality of layers of Restricted Boltzmann Machines (RBMs) and comprises at least one hidden layer and one visible layer, and the power system load prediction method comprises the following steps:

acquiring a training sample and a test sample; constructing an energy function of the RBM model;

training the at least one hidden layer and the visible layer by utilizing the training sample to obtain the weight of the training sample between the nodes of the at least one hidden layer and the visible layer;

and inputting the output data obtained by the training sample and the test sample into the trained DBN to obtain a predicted value of the power system load.

2. The method of claim 1, wherein the constructing the RBM model energy function is preceded by:

for the load data X_iFormed training sample X ═ X₁,X₂,...,X_MNormalizing; wherein X_iRepresenting a group of load data, and M representing the number of the load data;

and whitening the test sample according to the mean value and the variance of the training sample, and uniformly normalizing the test sample to be between 0 and 1 according to the maximum value and the minimum value of the training sample.

3. The power system load prediction method of claim 1, wherein the distribution of visible layer and hidden layer units satisfies bernoulli form; the energy function for constructing the RBM model comprises the following steps:

the energy function of the RBM model satisfying bernoulli distribution for the visible and hidden layers is defined as:

$E(v,h)=-\underset{i=1}{\overset{n}{\Σ}}\underset{j=1}{\overset{m}{\Σ}}{v}_i{w}_{ij}{h}_j-\underset{i=1}{\overset{n}{\Σ}}{a}_i{v}_i-\underset{j=1}{\overset{m}{\Σ}}{b}_j{h}_j;$

wherein, a_iAnd b_jRepresents a bias term, w_ijRepresenting the connection weight between the visible unit and the hidden layer unit, theta ═ a, b, w are weight parameters of the RBM model, and the vector h ═ h₁,h₂,…h_m) Represents a hidden layer unit, has m elements in total, and has a vector v ═ v₁,v₂,…,v_n) Representing a visible layer, for a total of n elements.

4. The method according to claim 3, wherein the training the at least one hidden layer and the visible layer by using the training samples layer by layer to obtain the weight of the training samples between the at least one hidden layer and the visible layer node comprises:

training the preweight of the RBM by adopting a divergence contrast method, and taking the obtained result as an initial value of a probability model for supervised learning training;

the obtained preset value and the initial value are used as input data and transmitted to a next layer of network, and training is carried out in the next layer of network;

and training the network one layer at a time, and training the at least one hidden layer and the visible layer by layer.

5. The power system load prediction method according to claim 3,

the output data obtained from the training samples at least comprise the average load value, the maximum load value, the average air temperature and the predicted daily average air temperature of the numerical weather forecast of the training samples.

6. An electric power system load prediction device based on a Deep Belief Network (DBN), wherein the DBN is composed of a plurality of layers of Restricted Boltzmann Machines (RBMs) and comprises at least one hidden layer and one visible layer, and the electric power system load prediction device comprises:

the data processing unit is used for acquiring a training sample and a test sample; constructing an energy function of the RBM model;

the training unit is used for training the at least one hidden layer and the visible layer by utilizing the training samples to obtain the weight of the training samples between the nodes of the at least one hidden layer and the visible layer;

and the prediction unit is used for inputting the output data obtained by the training sample and the test sample into the trained DBN to obtain a predicted value of the power system load.

7. The power system load prediction device of claim 6,

the data processing unit is also used for processing the load data X_iFormed training sample X ═ X₁,X₂,...,X_MNormalizing; wherein X_iRepresenting a group of load data, and M representing the number of the load data; and the device is also used for whitening the test sample according to the mean value and the variance of the training sample and uniformly normalizing the test sample to be between 0 and 1 according to the maximum value and the minimum value of the training sample.

8. The power system load prediction device of claim 6,

the distribution of the visible layer and the hidden layer units meets the Bernoulli form;

the energy function of the RBM model satisfying Bernoulli distribution of the visible layer and the hidden layer defined by the data processing unit is as follows:

$E(v,h)=-\underset{i=1}{\overset{n}{\Σ}}\underset{j=1}{\overset{m}{\Σ}}{v}_i{w}_{ij}{h}_j-\underset{i=1}{\overset{n}{\Σ}}{a}_i{v}_i-\underset{j=1}{\overset{m}{\Σ}}{b}_j{h}_j;$

wherein, a_iAnd b_jRepresents a bias term, w_ijRepresenting the connection weight between the visible unit and the hidden layer unit, theta ═ a, b, w are weight parameters of the RBM model, and the vector h ═ h₁,h₂,…h_m) Represents a hidden layer unit, has m elements in total, and has a vector v ═ v₁,v₂,…,v_n) Representing a visible layer, for a total of n elements.

9. The power system load prediction device according to claim 8,

the training unit is specifically used for training the preset value of the RBM by adopting a device for comparing divergence, and taking the obtained result as an initial value of the probability model for supervised learning training; the obtained preset value and the initial value are used as input data and transmitted to a next layer of network, and training is carried out in the next layer of network; and training the network one layer at a time, and training the at least one hidden layer and the visible layer by layer.

10. The power system load prediction device according to claim 8,

the output data obtained from the training samples at least comprise the average load value, the maximum load value, the average air temperature and the predicted daily average air temperature of the numerical weather forecast of the training samples.