Load prediction method and system based on incremental echo state network

Abstract

The invention discloses a load prediction method and system based on an incremental echo state network, and relates to the technical field of power load prediction, and the method comprises the steps: defining a similarity measurement standard, and carrying out the retrieval based on a similar time sequence to obtain a prediction sample set; according to an empirical formula, the number of neurons of the incremental echo state network is calculated and determined, and network parameters are set according to tasks; and training the incremental echo state network by using the prediction sample set, and predicting by using the trained network. According to the method, historical data similar to a target load mode is quickly positioned by adopting a similar time sequence retrieval technology and quantitatively analyzing and comparing the characteristics of the time sequence, so that a more accurate sample set is provided for a load prediction model; by providing an empirical formula, the number of neurons of a hidden layer is adaptively adjusted according to the characteristics of load data and the requirements of a prediction task, dynamic optimization of a network structure is realized, and the flexibility and accuracy of load prediction are improved.

Description

Technical Field

[0001] The present invention relates to the technical field of electric load forecasting, and in particular to a load forecasting method and system based on an incremental echo state network. Background Art

[0002] Load forecasting of power systems is a key technology to ensure grid stability and optimize energy distribution. With the development of social economy and the diversification of energy consumption patterns, load data has shown increasingly strong nonlinear and dynamic characteristics. Traditional load forecasting methods, such as linear regression based on statistics and time series analysis, are difficult to accurately capture the complexity and variability of load data due to their inherent linear assumptions.

[0003] To address this challenge, deep learning techniques, especially recurrent neural networks (RNNs) and their variants, such as long short-term memory networks (LSTMs) and echo state networks (ESNs), have attracted widespread attention due to their powerful ability to process time series data. As a special type of RNN, ESNs perform well in processing large-scale time series data due to their random and fixed internal nodes and easy training.

[0004] In the existing technology, ESN faces a major problem when processing incremental data or real-time data: once the network structure is determined, the number of neurons is fixed. This limits the network's ability to adapt to new data, because when the data distribution changes, the fixed-structure network may not be able to effectively capture new patterns and trends. In addition, retraining the entire network to adapt to new data is not only time-consuming, but may also lead to excessive forgetting of historical data.

[0005] To solve this problem, the present invention proposes a load forecasting method based on an incremental echo state network. The incremental echo state network can dynamically adjust the number of neurons to adapt to changes in data. Compared with traditional ESNs, when receiving new data, the incremental echo state network can dynamically increase or decrease the number of neurons according to the characteristics of the data and the requirements of the forecasting task. This dynamic adjustment mechanism enables the incremental echo state network to adapt to changes in load data more flexibly while maintaining memory of historical data. Summary of the invention

[0006] In view of the above-mentioned problems, the present invention is proposed.

[0007] Therefore, the technical problem solved by the present invention is: the existing technical load forecasting method has the problem that once the number of neurons in the network structure is determined, it remains fixed, resulting in the network being unable to effectively capture new patterns and trends, and how to dynamically adjust the number of neurons to more flexibly adapt to changes in load data.

[0008] To solve the above technical problems, the present invention provides the following technical solutions: a load forecasting method based on an incremental echo state network, comprising defining a similarity metric, retrieving a prediction sample set based on similar time series; calculating and determining the number of neurons in the incremental echo state network according to an empirical formula, and setting network parameters according to the task; training the incremental echo state network using the prediction sample set, and using the trained network for prediction.

[0009] As a preferred solution of the load forecasting method based on incremental echo state network described in the present invention, the method of obtaining the forecast sample set includes setting the length before the point to be predicted as N, and the load sequence is expressed as:

[0010] S=(s[i],i=0,1,...,N-1)

[0011] The set of retrieval space history sequences is set to T h , the set of similar sequences is set to T s ; T h Contains the historical load sequence of the same period L days before the forecast date, T s That is, the prediction sample set, the capacity upper limit is Q, and the given threshold is ε; for the load sequence X∈T, if D(S,X)<ε, then X∈T s ; Repeat the judgment until T s The capacity limit has been reached or T h Iterate over the elements in .

[0012] As a preferred solution of the load forecasting method based on the incremental echo state network of the present invention, wherein: the determining the number of hidden layer neurons of the incremental echo state network includes determining the number of neurons according to an empirical formula; the empirical formula is expressed as:

[0013]

[0014] Among them, N h Represents the number of neurons, N i Represents the number of neurons in the input layer, N o Represents the number of neurons in the output layer, N s represents the number of samples in the training set, and α represents an arbitrary value variable, usually ranging from 2 to 10.

[0015] As a preferred solution of the load forecasting method based on incremental echo state network described in the present invention, the network parameters include the spectral radius of the internal connection weight matrix of the reserve pool, the scale of the input unit and the sparsity of the reserve pool.

[0016] As a preferred solution of the load forecasting method based on the incremental echo state network described in the present invention, wherein: the spectral radius of the internal connection weight matrix of the reserve pool includes the initialization process of the internal connection matrix W of the reserve pool is to randomly generate a connection matrix W0 of a scale of N×N, and the connection weights are uniformly distributed and selected from [-1,1]; calculate the spectral radius λ of W0 max , expressed as:

[0017]

[0018] Normalize and multiply W1 by the coefficient α', 0<α'<1, expressed as:

[0019] W=α'W1.

[0020] As a preferred solution of the load forecasting method based on the incremental echo state network described in the present invention, the training of the incremental echo state network includes initializing the input weight matrix W according to the parameters set according to the network scale. in , W and feedback weight W back ; At initialization, the reserve pool state vector X(0) at time 0 = 0; Network state collection, assuming that the training sample set has a total of M pairs (u(i), y(i)) (i = 1, 2, 3, ..., M), then the reserve pool state is updated, expressed as:

[0021] x(n+1)=f(W in u(n+1)+Wx(n)+W back y(n))

[0022] Among them, u(n+1) represents the input of the n+1th step, x(n) represents the state vector of the reserve pool, and y(n) represents the output vector; assuming that the discarded state length is m, the network state matrix S formed by the training sample set is (Mm)×(K+N+L) dimensional; among them, the (i+1)th row of the network state matrix consists of the input, the reserve pool state, and the training sample output, expressed as:

[0023] S(i+1,:)=[u(n+1)',X(n+1)',y(n)']

[0024] Collect signal status and assume the output function f out is the tanh function, then the target output of the training sample is the inverse function, which is tanh -1 (y(i)), thus forming the output state matrix D.

[0025] As a preferred solution of the load forecasting method based on the incremental echo state network described in the present invention, the training of the incremental echo state network also includes that the input weights, internal connection weights and feedback weights of the echo state network are all randomly determined, and the output weights are expressed as:

[0026] W out =(S T S) -1 S T D

[0027] For (S T S) -1 It is usually replaced by the pseudo-inverse, that is, (S T S) +1 .

[0028] Another object of the present invention is to provide a load forecasting system based on an incremental echo state network, which can enable the incremental echo state network to adapt to changes in load data more flexibly through a dynamic adjustment mechanism, thereby solving the problem that current technology is difficult to dynamically adjust the number of neurons.

[0029] As a preferred solution of the load forecasting system based on the incremental echo state network described in the present invention, it includes: a sample retrieval module, a network construction module, and a network training prediction module; the sample retrieval module is used to define a similarity metric and retrieve and obtain a prediction sample set based on similar time series; the network construction module is used to calculate and determine the number of neurons in the incremental echo state network according to an empirical formula, and set network parameters according to the task; the network training prediction module is used to train the incremental echo state network using the prediction sample set, and use the trained network for prediction.

[0030] A computer device comprises a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the steps of a load prediction method based on an incremental echo state network.

[0031] A computer-readable storage medium stores a computer program, which, when executed by a processor, implements the steps of a load prediction method based on an incremental echo state network.

[0032] Beneficial effects of the present invention: The load forecasting method based on incremental echo state network provided by the present invention adopts similar time series retrieval technology, quantitatively analyzes and compares the characteristics of time series, and quickly locates historical data similar to the target load pattern, thereby providing a more accurate sample set for the load forecasting model; by proposing an empirical formula, the number of hidden layer neurons is adaptively adjusted according to the characteristics of load data and the requirements of the forecasting task, the dynamic optimization of the network structure is achieved, and the flexibility and accuracy of load forecasting are improved; the load forecasting method adopted integrates multiple technical advantages of ESN, including random and fixed internal node connections, dynamic time series processing capabilities, and innovative similarity retrieval and neuron number adjustment mechanisms, which significantly improves the performance of load forecasting; by controlling the sparsity of the reserve pool, the amount of calculation in the network training process is reduced, and the calculation complexity is reduced. BRIEF DESCRIPTION OF THE DRAWINGS

[0033] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings required for use in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other accompanying drawings can be obtained based on these accompanying drawings without paying creative work.

[0034] Figure 1 An overall flow chart of a load forecasting method based on an incremental echo state network provided for the first embodiment of the present invention.

[0035] Figure 2 A topological diagram of an ESN of a load prediction method based on an incremental echo state network provided in the first embodiment of the present invention.

[0036] Figure 3 An overall flow chart of a load forecasting system based on an incremental echo state network is provided for the third embodiment of the present invention. DETAILED DESCRIPTION

[0037] In order to make the above-mentioned purposes, features and advantages of the present invention more obvious and easy to understand, the specific implementation methods of the present invention are described in detail below in conjunction with the drawings of the specification. Obviously, the described embodiments are part of the embodiments of the present invention, but not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary persons in the art without creative work should fall within the scope of protection of the present invention.

[0038] Example 1, reference Figure 1-Figure 2 , as an embodiment of the present invention, provides a load forecasting method based on an incremental echo state network, comprising:

[0039] S1: Define similarity metrics and retrieve prediction sample sets based on similar time series.

[0040] Furthermore, obtaining the prediction sample set includes setting the length before the point to be predicted to be N, and the load sequence is expressed as:

[0041] S=(s[i],i=0,1,...,N-1)

[0042] The set of retrieval space history sequences is set to T h , the set of similar sequences is set to T s ; T h Contains the historical load sequence of the same period L days before the forecast date, T s That is, the prediction sample set, the capacity upper limit is Q, and the given threshold is ε; for the load sequence X∈T, if D(S,X)<ε, then X∈T s ; Repeat the judgment until T s The capacity limit has been reached or T h Iterate over the elements in .

[0043] It should be noted that determining the number of hidden layer neurons of the incremental echo state network includes determining the number of neurons according to an empirical formula; the empirical formula is expressed as:

[0044]

[0045] Among them, N h Represents the number of neurons, N i Represents the number of neurons in the input layer, N o Represents the number of neurons in the output layer, N s represents the number of samples in the training set, and α represents an arbitrary value variable, usually ranging from 2 to 10.

[0046] It should also be noted that if Figure 2 The figure shows the topological structure of the incremental echo state network ESN, which includes the input layer, hidden layer and output layer from left to right. During the training process, the input connection weight matrix is ​​randomly initialized, and the output connection weight matrix changes.

[0047] It should also be noted that the number of neurons is determined. Specifically, assuming that the network input node is K, the number of neurons contained in the reserve pool is N, and the number of output nodes is L, the connections between the various layers of the network are as follows:

[0048] The input layer of K nodes and the hidden layer of N nodes are connected by input weights. The dimension of the matrix is ​​N×K, denoted as W in; The N nodes of the reserve pool are connected through the internal connection weight matrix, the dimension of which is N×N, denoted as W; the connection from the output layer to the reserve pool becomes the feedback weight, the dimension of which is N×L, denoted as W back The connection between the input layer, the reservoir layer, and the output layer itself becomes the output weight. The dimension of this matrix is ​​(K+N+L)×L, denoted as W out , the input weight, reserve pool weight and feedback weight are randomly generated, and the output weight is obtained by linear regression.

[0049] Assume that the input of the network at step n is u(n), the state vector of the reservoir is x(n), and the output vector is y(n), expressed as:

[0050] u(n)=[u1(n),u2(n),...,u k (n)] T

[0051] x(n)=[x1(n),x2(n),...,x N (n)] T

[0052] y(n)=[y1(n),y2(n),...,y L (n)] T

[0053] Update the status of each unit in the reserve pool, expressed as:

[0054] x(n+1)=f(W in u(n+1)+Wx(n)+W back y(n))

[0055] The output of the network is expressed as:

[0056] y(n+1)=f out (W out (u(n+1),x(n+1),y(n)))

[0057] Where, f=(f1,f2,...,f N ) represents the activation function of each node in the reserve pool, which is generally taken as the sigmoid function or the hyperbolic tangent function (tanh). represents the output function of the network output node, and (u(n+1),x(n+1),y(n)) represents the network state vector composed of input, internal unit function and previous output.

[0058] S2: Calculate and determine the number of neurons in the incremental echo state network according to the empirical formula, and set the network parameters according to the task.

[0059] Furthermore, the network parameters include the spectral radius of the internal connection weight matrix of the reservoir, the input unit scale, and the sparsity of the reservoir.

[0060] It should be noted that the spectral radius of the internal connection weight matrix of the reserve pool includes the initialization process of the internal connection matrix W of the reserve pool, which is to randomly generate a connection matrix W0 of size N×N, and the connection weights are uniformly distributed and selected from [-1,1]; calculate the spectral radius λ of W0 max , expressed as:

[0061]

[0062] Normalize and multiply W1 by the coefficient α', 0<α'<1, expressed as:

[0063] W=α'W1.

[0064] It should also be noted that the radius of the internal connection weight spectrum of the reserve pool refers to the maximum absolute value of the eigenvalue of the internal connection weight matrix W of the reserve pool, denoted by λ max .

[0065] The spectral radius is an important parameter of the reserve pool and has an important influence on the echo state properties of the echo state network. max When <1, the echo state network has the echo state property, which can ensure that the influence of the historical reserve pool state and input on the network will disappear after a sufficiently long time. The spectral radius less than 1 is a sufficient but not necessary condition for the echo state network to have the echo state property.

[0066] It should also be noted that the input unit scale includes in Uniform distribution, by assigning the input scale a to a variation interval [-a, a], W in The value of is selected from the variation range. For the linear task W in When W is small, the tanh() activation function is close to the linear state. in When the scale is larger, neurons tend to reach saturation, approaching 1 and -1, and changing in a more nonlinear form.

[0067] S3: Use the prediction sample set to train the incremental echo state network, and use the trained network to make predictions.

[0068] Furthermore, training the incremental echo state network includes initializing the input weight matrix W according to the parameters set according to the network size. in , W and feedback weight W back ; At initialization, the reserve pool state vector X(0) at time 0 = 0; Network state collection, assuming that the training sample set has a total of M pairs (u(i), y(i)) (i = 1, 2, 3, ..., M), then the reserve pool state is updated, expressed as:

[0069] x(n+1)=f(W in u(n+1)+Wx(n)+W back y(n))

[0070] Among them, u(n+1) represents the input of the n+1th step, x(n) represents the state vector of the reserve pool, and y(n) represents the output vector; assuming that the discarded state length is m, the network state matrix S formed by the training sample set is (Mm)×(K+N+L) dimensional; among them, the (i+1)th row of the network state matrix consists of the input, the reserve pool state, and the training sample output, expressed as:

[0071] S(i+1,:)=[u(n+1)',X(n+1)',y(n)']

[0072] Collect signal status and assume the output function f out is the tanh function, then the target output of the training sample is the inverse function, which is tanh -1 (y(i)), thus forming the output state matrix D.

[0073] It should be noted that the training of the incremental echo state network also includes that the input weights, internal connection weights and feedback weights of the echo state network are all randomly determined, and the output weights are expressed as:

[0074] W out =(S T S) -1 S T D

[0075] For (S T S) -1 It is usually replaced by the pseudo-inverse, that is, (S T S) +1 .

[0076] Embodiment 2 is an embodiment of the present invention, which provides a load forecasting method based on an incremental echo state network. In order to verify the beneficial effects of the present invention, scientific demonstration is carried out through economic benefit calculation and simulation experiments.

[0077] First, this embodiment aims to verify a power load forecasting method based on incremental echo state network, which improves the accuracy and adaptability of load forecasting by combining similar time series retrieval and dynamic neuron adjustment mechanism.

[0078] The historical load data of a regional power system from January 2022 to June 2023 are collected, including hourly load values ​​and related meteorological data. The meteorological data include temperature, humidity, wind speed, etc. The data are divided into two categories according to date type: weekdays and weekends, and further subdivided into three time periods: morning, afternoon, and evening. The Euclidean distance is used as the similarity metric to calculate the difference between the two load sequences. Considering the influence of meteorological factors, meteorological data such as temperature, humidity, and wind speed are also included in the similarity calculation. Taking the load forecast at 8 am on June 1, 2023 as an example, the load sequences at 8 am on weekdays and weekends similar to 8 am on June 1 are retrieved from the historical data, as well as the corresponding meteorological data. The similarity threshold ε=0.1 is set, and the top 10 sequences similar to the current load pattern are selected as the prediction sample set.

[0079] The number of hidden layer neurons is calculated according to the empirical formula, the number of input layer neurons M = 6, the number of output layer neurons L = 1, the number of training set samples K = 10, the number of hidden layer neurons N = 10, the spectral radius of the internal connection weight matrix of the reserve pool is set to 0.9, the input unit scale is 1, and the sparsity of the reserve pool is 0.5. The incremental echo state network is trained with the predicted sample set to learn the patterns and trends in the load data. The trained network is used to predict the load at 8:00 am on June 1, 2023, and compared with the actual load value.

[0080] Table 1 Incremental echo state network load prediction data

[0081] Test subjects date Time Forecast load MW Actual load MW Error MW Prediction 1 2023-06-01 8 a.m. 1300 1297 3 Prediction 2 2023-06-02 8 a.m. 1285 1280 5 Prediction 3 2023-06-03 8 a.m. 1270 1271 1 Prediction 4 2023-06-04 8 a.m. 1254 1255 1 Prediction 5 2023-06-05 8 a.m. 1240 1238 2 Prediction 6 2023-06-06 8 a.m. 1225 1228 3

[0082] It can be seen from the data in Table 1 that the prediction errors of the present invention are all less than 5MW, indicating that the prediction accuracy of the present method is relatively high. By comparing the prediction results with the actual load values, it can be seen that the present method has the following beneficial effects: in terms of high prediction accuracy, the present method can effectively capture the complex patterns and trends in the load data, and the prediction results are highly consistent with the actual load values ​​with small errors; in terms of strong adaptability, the incremental echo state network can dynamically adjust the number of neurons according to the changes in the load data, so that it can more flexibly adapt to the load prediction tasks in different scenarios; in terms of high efficiency, the present method quickly obtains the prediction sample set through similar time series retrieval, and dynamically adjusts the network structure using empirical formulas, thereby improving the prediction efficiency.

[0083] Example 3, reference Figure 3 , as an embodiment of the present invention, provides a load prediction system based on incremental echo state network, including a sample retrieval module, a network construction module, and a network training prediction module.

[0084] The sample retrieval module is used to define the similarity metric and retrieve the prediction sample set based on similar time series; the network construction module is used to calculate and determine the number of neurons in the incremental echo state network according to the empirical formula, and set the network parameters according to the task; the network training prediction module is used to train the incremental echo state network using the prediction sample set and use the trained network for prediction.

[0085] If the function is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, or the part of the technical solution, can be embodied in the form of a software product. The computer software product is stored in a storage medium, including several instructions for a computer device (which can be a personal computer, server, or network device, etc.) to perform all or part of the steps of the methods of each embodiment of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), disk or optical disk, etc. Various media that can store program codes.

[0086] The logic and / or steps represented in the flowchart or otherwise described herein, for example, can be considered as an ordered list of executable instructions for implementing logical functions, and can be embodied in any computer-readable medium for use by an instruction execution system, device or apparatus (such as a computer-based system, a system including a processor, or other system that can fetch instructions from an instruction execution system, device or apparatus and execute instructions), or in conjunction with such instruction execution systems, devices or apparatuses. For the purposes of this specification, "computer-readable medium" can be any device that can contain, store, communicate, propagate or transmit a program for use by an instruction execution system, device or apparatus, or in conjunction with such instruction execution systems, devices or apparatuses.

[0087] More specific examples of computer-readable media (a non-exhaustive list) include the following: an electrical connection with one or more wires (electronic device), a portable computer disk case (magnetic device), a random access memory (RAM), a read-only memory (ROM), an erasable and programmable read-only memory (EPROM or flash memory), an optical fiber device, and a portable compact disk read-only memory (CDROM). In addition, the computer-readable medium may even be a paper or other suitable medium on which the program is printed, since the program may be obtained electronically, for example, by optically scanning the paper or other medium, followed by editing, deciphering, or processing in another suitable manner as necessary, and then stored in a computer memory.

[0088] It should be understood that the various parts of the present invention can be implemented by hardware, software, firmware or a combination thereof. In the above-mentioned embodiments, multiple steps or methods can be implemented by software or firmware stored in a memory and executed by a suitable instruction execution system. For example, if implemented by hardware, as in another embodiment, it can be implemented by any one of the following technologies known in the art or their combination: a discrete logic circuit having a logic gate circuit for implementing a logic function for a data signal, a dedicated integrated circuit having a suitable combination of logic gate circuits, a programmable gate array (PGA), a field programmable gate array (FPGA), etc. It should be noted that the above embodiments are only used to illustrate the technical solution of the present invention and are not limited. Although the present invention is described in detail with reference to the preferred embodiments, it should be understood by those skilled in the art that the technical solution of the present invention can be modified or replaced by equivalents without departing from the spirit and scope of the technical solution of the present invention, which should be included in the scope of the claims of the present invention.

[0089] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical solutions of the present invention, which should all be included in the scope of the claims of the present invention.

Claims

1. A load forecasting method based on incremental echo state network, characterized in that: include: Define similarity metrics and retrieve prediction sample sets based on similar time series; According to the empirical formula, the number of neurons in the incremental echo state network is calculated and determined, and the network parameters are set according to the task; The incremental echo state network is trained using the prediction sample set, and the trained network is used for prediction.

2. The load forecasting method based on incremental echo state network according to claim 1, characterized in that: The method of obtaining the prediction sample set includes assuming that the length before the point to be predicted is N, and the load sequence is expressed as: S=(s[i],i=0,1,...,N-1) The set of retrieval space history sequences is set to T h , the set of similar sequences is set to T s ; T h Contains the historical load sequence of the same period L days before the forecast date, T s That is, the prediction sample set, the upper capacity limit is Q, and the given threshold is ε; For a load sequence X∈T, if D(S,X)<ε, then X∈T s ; Repeat the judgment until T s The capacity limit has been reached or T h Iterate over the elements in .

3. The load forecasting method based on incremental echo state network according to claim 2, characterized in that: Determining the number of hidden layer neurons of the incremental echo state network includes determining the number of neurons according to an empirical formula; The empirical formula is expressed as: Among them, N h Represents the number of neurons, N i Represents the number of neurons in the input layer, N o Represents the number of neurons in the output layer, N s represents the number of samples in the training set, and α represents an arbitrary value variable, usually ranging from 2 to 10.

4. The load forecasting method based on incremental echo state network according to claim 3, characterized in that: The network parameters include the spectral radius of the internal connection weight matrix of the reserve pool, the scale of the input unit and the sparsity of the reserve pool.

5. The load forecasting method based on incremental echo state network according to claim 4, characterized in that: The spectral radius of the internal connection weight matrix of the reserve pool includes the initialization process of the internal connection matrix W of the reserve pool, which is to randomly generate a connection matrix W0 of size N×N, and the connection weights are uniformly distributed and selected from [-1,1]; Calculate the spectral radius λ of W0 max , expressed as: Normalize and multiply W1 by the coefficient α′, 0<α′<1, expressed as: W = α′W1.

6. The load forecasting method based on incremental echo state network according to claim 5, characterized in that: The training of the incremental echo state network includes initializing the input weight matrix W according to the parameters set according to the network scale. in , W and feedback weight W back ; During initialization, the state vector of the reserve pool at time 0 is X(0) = 0; Network status collection, assuming that the training sample set has a total of M pairs (u(i), y(i)) (i = 1, 2, 3, ..., M), then the reserve pool status is updated, expressed as: x(n+1)=f(W in u(n+1)+Wx(n)+W back y(n)) Among them, u(n+1) represents the input of the n+1th step, x(n) represents the state vector of the reservoir, and y(n) represents the output vector; Assuming that the discarded state length is m, the network state matrix S formed by the training sample set is (Mm)×(K+N+L) dimensional; Among them, the (i+1)th row of the network state matrix consists of input, reserve pool state and training sample output, expressed as: S(i+1,:)=[u(n+1)′, X(n+1)′, y(n)′] Collect signal status and assume the output function f out is the tanh function, then the target output of the training sample is the inverse function, which is tanh -1 (y(i)), thus forming the output state matrix D.

7. The load forecasting method based on incremental echo state network according to claim 6, characterized in that: The training of the incremental echo state network also includes that the input weights, internal connection weights and feedback weights of the echo state network are all randomly determined, and the output weights are expressed as: W out =(S T S) -1 S T D For (S T S) -1 It is usually replaced by the pseudo-inverse, that is, (S T S) +1 .

8. A system using the load forecasting method based on incremental echo state network as claimed in any one of claims 1 to 7, characterized in that: Including sample retrieval module, network construction module, network training prediction module; The sample retrieval module is used to define a similarity metric and retrieve a prediction sample set based on similar time series; The network construction module is used to calculate and determine the number of neurons in the incremental echo state network according to an empirical formula, and to set network parameters according to the task; The network training prediction module is used to train the incremental echo state network using the prediction sample set and to perform prediction using the trained network.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the load forecasting method based on incremental echo state network according to any one of claims 1 to 7 are implemented.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the load forecasting method based on incremental echo state network according to any one of claims 1 to 7 are implemented.

Images