Short-term load prediction method and device, storage medium and terminal

Abstract

The invention discloses a short-term load prediction method and device, a storage medium and a terminal, relates to the technical field of artificial intelligence, and mainly aims to solve the problem of low precision of an existing short-term load prediction result. The method includes: acquiring historical load data of a power system; and performing load prediction operation according to the historical load data based on a pre-constructed load prediction model to obtain a short-term load prediction result of the power system, the load prediction model being formed by mixing a plurality of load prediction sub-models, the load prediction model is obtained through a polar bear algorithm training stage, a global optimization stage and a linear integration stage in sequence, the polar bear algorithm training stage is used for generating an initial parameter set of the load prediction model, and the global optimization stage is used for updating the initial parameter set; the linear integration stage is used for determining the weight of each load prediction sub-model.

Description

Technical Field

[0001] The present application relates to the field of artificial intelligence technology, and in particular to a short-term load forecasting method and device, storage medium, and terminal. Background Art

[0002] In the power system, short-term load forecasting (STLF) plays a vital role. Accurate forecasting results can not only ensure the reliable operation of the power system, but also save operating costs. Traditional statistical methods include exponential smoothing, regression analysis, Kalman filter, autoregressive moving average, autoregressive integrated moving average, etc. These statistical methods have great advantages in potential physical explanations, mature technology and easy-to-understand algorithms. However, affected by weather conditions, holidays and other factors, the power load characteristics have high nonlinearity and randomness, resulting in low accuracy of forecast results based on statistical methods.

[0003] In recent years, with the development of artificial intelligence technology, artificial neural networks, convolutional neural networks, support vector machines, extreme learning machines, fuzzy logic frameworks, expert systems, deep Boltzmann machines, radial basis function neural networks and other artificial intelligence technologies have been widely used in STLF. However, due to the volatility, sensitivity and time correlation of power load data, the prediction results based on a single artificial intelligence method are of low accuracy. Summary of the invention

[0004] In view of this, the present application provides a short-term load forecasting method and device, a storage medium, and a terminal, the main purpose of which is to improve the problem of low accuracy of existing short-term load forecasting results.

[0005] According to one aspect of the present application, a short-term load forecasting method is provided, comprising:

[0006] Obtain historical load data of the power system;

[0007] Based on the pre-constructed load forecasting model, load forecasting operations are performed according to the historical load data to obtain a short-term load forecasting result of the power system. The load forecasting model is composed of a mixture of multiple load forecasting sub-models, and is obtained through the polar bear algorithm training stage, the global optimization stage, and the linear integration stage in sequence. The polar bear algorithm training stage is used to generate an initial parameter group for the load forecasting model, the global optimization stage is used to update the initial parameter group, and the linear integration stage is used to determine the weights of each of the load forecasting sub-models.

[0008] Preferably, before performing a load forecasting operation based on the pre-built load forecasting model according to the historical load data to obtain a short-term load forecasting result of the power system, the method further includes:

[0009] Construct an initial load forecasting model based on LSTM neural network;

[0010] Randomly select the initial position of each individual polar bear;

[0011] A global search is performed according to the individual polar bear movement behavior formula to update each of the initialization positions to obtain the global optimal position of each of the individual polar bears. The individual polar bear movement behavior formula is as follows:

[0012]

[0013] in, represents the motion of the i-th polar bear at coordinate j in the t-th iteration, x s represents the motion coordinates of the polar bear, w represents the Euclidean distance between the current position of the polar bear and the best polar bear in the population, γ represents a random number in the interval [0, w], sign(w) represents a piecewise function, and the result of the sign(w) function is {-1, 0, 1} according to the value of w, and α represents a random number in the interval (0, 1);

[0014] A local search is performed according to a search radius formula to update each of the global optimal positions to obtain the optimal position for completing the local search. The search radius formula is as follows:

[0015]

[0016] Among them, a represents the field of vision of the polar bear, Represents a random number in the interval (0,π / 2);

[0017] Repeat the above global search step and local search step until the preset number of iterations is met to obtain the global optimal position of each polar bear individual.

[0018] Preferably, after obtaining the global optimal position, the method further comprises:

[0019] Grouping the global optimal positions based on a grouping algorithm;

[0020] When the group members and the number of groups remain unchanged, the grouping is stopped to obtain the initial parameter group of the load forecasting model.

[0021] Preferably, after obtaining the initial parameter group of the load forecasting model, the method further includes:

[0022] Searching for a loss function overflow point along a preset search direction based on the initial parameter group to determine an adjacent stable region;

[0023] Determine the best position within the adjacent stable area based on the polar bear algorithm;

[0024] An optimal solution is selected from the best position in the adjacent stable region and the initial parameter group based on a loss function to obtain an updated parameter group.

[0025] Preferably, after selecting the optimal solution from the best position in the adjacent stable region and the initial parameter group based on the loss function to obtain the updated parameter group, the method further comprises:

[0026] The weights of each load forecasting sub-model in the updated parameter group are calculated based on the following formula:

[0027]

[0028] Among them, H lass (i) represents the loss function of the i-th load forecasting sub-model;

[0029] The load forecasting sub-models and corresponding weights are linearly combined to obtain a load forecasting model.

[0030] Preferably, after acquiring the historical load data of the power system, the method further includes:

[0031] The initial cluster center v of the historical load data is calculated based on the initial membership of each sample based on the following formula: j ,

[0032]

[0033] Among them, u ij represents the initial membership of the sample, m represents the fuzzy factor, x i represents the i-th sample, and n represents the number of samples;

[0034] Calculating the Pearson distance between each of the samples and the initial cluster center;

[0035] Based on the following formula, each of the initial memberships is updated according to the Pearson distance to obtain an updated membership.

[0036]

[0037] Among them, d pearson represents the Pearson distance function, c represents the number of categories, x j represents the jth sample, v k represents the kth cluster center;

[0038] The initial cluster center is updated according to the updated membership degree based on the following formula to obtain the updated cluster center

[0039]

[0040] Based on the above steps, the cluster center is iteratively updated until the preset accuracy is met, and multiple groups of historical load data sets are obtained, and the user power consumption behavior of each of the historical load data sets is different;

[0041] The load forecasting operation based on the pre-built load forecasting model is performed according to the historical load data to obtain the short-term load forecasting result of the power system, specifically including:

[0042] Based on the pre-built load forecasting model, load forecasting operations are performed respectively according to each of the historical load data sets to obtain multiple groups of short-term load forecasting results, and the short-term load forecasting results have the same user power consumption behavior as the corresponding historical load data sets.

[0043] According to another aspect of the present application, a short-term load forecasting device is provided, comprising:

[0044] A load data acquisition module is used to acquire historical load data of the power system;

[0045] The short-term load forecasting module is used to perform load forecasting operations according to the historical load data based on a pre-built load forecasting model to obtain a short-term load forecasting result for the power system. The load forecasting model is composed of a mixture of multiple load forecasting sub-models, and is obtained through the polar bear algorithm training stage, the global optimization stage, and the linear integration stage in sequence. The polar bear algorithm training stage is used to generate an initial parameter group for the load forecasting model, the global optimization stage is used to update the initial parameter group, and the linear integration stage is used to determine the weights of each of the load forecasting sub-models.

[0046] Preferably, before the short-term load forecasting module, the device further includes a training module for:

[0047] Construct an initial load forecasting model based on LSTM neural network;

[0048] Randomly select the initial position of each individual polar bear;

[0049] A global search is performed according to the individual polar bear movement behavior formula to update each of the initialization positions to obtain the global optimal position of each of the individual polar bears. The individual polar bear movement behavior formula is as follows:

[0050]

[0051] in, represents the motion of the i-th polar bear at coordinate j in the t-th iteration, x srepresents the motion coordinates of the polar bear, w represents the Euclidean distance between the current position of the polar bear and the best polar bear in the population, γ represents a random number in the interval [0, w], sign(w) represents a piecewise function, and the result of the sign(w) function is {-1, 0, 1} according to the value of w, and α represents a random number in the interval (0, 1);

[0052] A local search is performed according to a search radius formula to update each of the global optimal positions to obtain the optimal position for completing the local search. The search radius formula is as follows:

[0053]

[0054] Among them, a represents the field of vision of the polar bear, Represents a random number in the interval (0,π / 2);

[0055] Repeat the above global search step and local search step until the preset number of iterations is met to obtain the global optimal position of each polar bear individual.

[0056] Preferably, after the training module, the device further comprises an optimization module for:

[0057] Grouping the global optimal positions based on a grouping algorithm;

[0058] When the group members and the number of groups remain unchanged, the grouping is stopped to obtain the initial parameter group of the load forecasting model.

[0059] Preferably, after obtaining the initial parameter group of the load forecasting model, the method further includes:

[0060] Searching for a loss function overflow point along a preset search direction based on the initial parameter group to determine an adjacent stable region;

[0061] Determine the best position within the adjacent stable area based on the polar bear algorithm;

[0062] An optimal solution is selected from the best position in the adjacent stable region and the initial parameter group based on a loss function to obtain an updated parameter group.

[0063] Preferably, after the optimization module, the device further includes an integration module for:

[0064] The weights of each load forecasting sub-model in the updated parameter group are calculated based on the following formula:

[0065]

[0066] Among them, H lass (i) represents the loss function of the i-th load forecasting sub-model;

[0067] The load forecasting sub-models and corresponding weights are linearly combined to obtain a load forecasting model.

[0068] Preferably, after the load data acquisition module, the device further includes a clustering module for:

[0069] The initial cluster center v of the historical load data is calculated based on the initial membership of each sample based on the following formula: j ,

[0070]

[0071] Among them, u ij represents the initial membership of the sample, m represents the fuzzy factor, x i represents the i-th sample, and n represents the number of samples;

[0072] Calculating the Pearson distance between each of the samples and the initial cluster center;

[0073] Based on the following formula, each of the initial memberships is updated according to the Pearson distance to obtain an updated membership.

[0074]

[0075] Among them, d pearson represents the Pearson distance function, c represents the number of categories, x j represents the jth sample, v k represents the kth cluster center;

[0076] The initial cluster center is updated according to the updated membership degree based on the following formula to obtain the updated cluster center

[0077]

[0078] Based on the above steps, the cluster center is iteratively updated until the preset accuracy is met, and multiple groups of historical load data sets are obtained, and the user power consumption behavior of each of the historical load data sets is different;

[0079] The load forecasting operation based on the pre-built load forecasting model is performed according to the historical load data to obtain the short-term load forecasting result of the power system, specifically including:

[0080] Based on the pre-built load forecasting model, load forecasting operations are performed respectively according to each of the historical load data sets to obtain multiple groups of short-term load forecasting results, and the short-term load forecasting results have the same user power consumption behavior as the corresponding historical load data sets.

[0081] According to another aspect of the present application, a storage medium is provided, wherein at least one executable instruction is stored in the storage medium, and the executable instruction enables a processor to perform operations corresponding to the above-mentioned short-term load forecasting method.

[0082] According to another aspect of the present application, a terminal is provided, comprising: a processor, a memory, a communication interface and a communication bus, wherein the processor, the memory and the communication interface communicate with each other through the communication bus;

[0083] The memory is used to store at least one executable instruction, and the executable instruction enables the processor to execute operations corresponding to the above-mentioned short-term load forecasting method.

[0084] By means of the above technical solution, the technical solution provided by the embodiment of the present application has at least the following advantages:

[0085] The present application provides a short-term load forecasting method and device, storage medium, and terminal. First, the historical load data of the power system is obtained; secondly, a load forecasting operation is performed based on the historical load data based on a pre-built load forecasting model to obtain a short-term load forecasting result of the power system. The load forecasting model is composed of a mixture of multiple load forecasting sub-models, which are obtained through a polar bear algorithm training stage, a global optimization stage, and a linear integration stage in sequence. The polar bear algorithm training stage is used to generate an initial parameter group of the load forecasting model, the global optimization stage is used to update the initial parameter group, and the linear integration stage is used to determine the weights of each of the load forecasting sub-models. Compared with the prior art, the embodiment of the present application generates an initial parameter group of the load forecasting model through the polar bear algorithm training stage, and updates the initial parameter group through the global optimization stage, so that multiple optimal load forecasting sub-models can be obtained; further, the weights of each optimal load forecasting sub-model are determined through the linear integration stage to obtain a load forecasting model composed of a mixture of multiple load forecasting sub-models, which can make full use of the advantages of each sub-model to improve the prediction accuracy of the load forecasting model, so that the accuracy of the short-term load forecasting result obtained based on the load forecasting model according to the historical load data set is improved.

[0086] The above description is only an overview of the technical solution of the present application. In order to more clearly understand the technical means of the present application, it can be implemented in accordance with the contents of the specification. In order to make the above and other purposes, features and advantages of the present application more obvious and easy to understand, the specific implementation methods of the present application are listed below. BRIEF DESCRIPTION OF THE DRAWINGS

[0087] Various other advantages and benefits will become apparent to those of ordinary skill in the art by reading the detailed description of the preferred embodiments below. The accompanying drawings are only for the purpose of illustrating the preferred embodiments and are not to be considered as limiting the present application. Also, the same reference symbols are used throughout the accompanying drawings to represent the same components. In the accompanying drawings:

[0088] Figure 1 A flow chart of a short-term load forecasting method provided by an embodiment of the present application is shown;

[0089] Figure 2 Another short-term load forecasting method flow chart provided by an embodiment of the present application is shown;

[0090] Figure 3 A schematic diagram of a search path in the global optimization phase provided by an embodiment of the present application is shown;

[0091] Figure 4 A block diagram of a short-term load forecasting device provided in an embodiment of the present application is shown;

[0092] Figure 5 A schematic diagram of the structure of a terminal provided in an embodiment of the present application is shown. DETAILED DESCRIPTION

[0093] The exemplary embodiments of the present disclosure will be described in more detail below with reference to the accompanying drawings. Although the exemplary embodiments of the present disclosure are shown in the accompanying drawings, it should be understood that the present disclosure can be implemented in various forms and should not be limited by the embodiments set forth herein. On the contrary, these embodiments are provided to enable a more thorough understanding of the present disclosure and to fully convey the scope of the present disclosure to those skilled in the art.

[0094] At the same time, it should be understood that for the convenience of description, the sizes of the various parts shown in the drawings are not drawn according to the actual proportional relationship.

[0095] The following description of at least one exemplary embodiment is merely illustrative in nature and is in no way intended to limit the present application, its application, or uses.

[0096] Technologies, methods, and equipment known to ordinary technicians in the relevant art may not be discussed in detail, but where appropriate, the technologies, methods, and equipment should be considered as part of the specification.

[0097] It should be noted that like reference numerals and letters refer to similar items in the following figures, and therefore, once an item is defined in one figure, it need not be further discussed in subsequent figures.

[0098] The embodiments of the present application can be applied to computer systems / servers, which can operate with many other general or special computing system environments or configurations. Examples of well-known computing systems, environments and / or configurations suitable for use with computer systems / servers include, but are not limited to: personal computer systems, server computer systems, thin clients, thick clients, handheld or laptop devices, microprocessor-based systems, set-top boxes, programmable consumer electronics, network personal computers, small computer systems, large computer systems, and distributed cloud computing technology environments including any of the above systems, etc.

[0099] Computer systems / servers may be described in the general context of computer system executable instructions (such as program modules) executed by a computer system. In general, program modules may include routines, programs, object programs, components, logic, data structures, etc., which perform specific tasks or implement specific abstract data types. Computer systems / servers may be implemented in a distributed cloud computing environment where tasks are performed by remote processing devices linked through a communication network. In a distributed cloud computing environment, program modules may be located on local or remote computing system storage media including storage devices.

[0100] The present application embodiment provides a short-term load forecasting method, such as Figure 1 As shown, the method includes:

[0101] 101. Obtain historical load data of the power system.

[0102] The historical load data is used to characterize the load data in the power system before the current time. For example, the historical load data for the first 8 months of the year is obtained to predict the load trend for the next 4 months. In the embodiment of the present application, the current execution end may be a load forecasting module in the power system, which obtains the historical load data within a preset time before the current time.

[0103] 102. Based on the pre-built load forecasting model, load forecasting operations are performed according to historical load data to obtain short-term load forecasting results for the power system.

[0104] Among them, the load forecasting model is composed of a mixture of multiple load forecasting sub-models, which are obtained in sequence through the polar bear algorithm training stage, the global optimization stage, and the linear integration stage. The polar bear algorithm training stage is used to generate the initial parameter group of the load forecasting model, the global optimization stage is used to update the initial parameter group, and the linear integration stage is used to determine the weights of each load forecasting sub-model. Due to the volatility, sensitivity and time correlation of power load data, the prediction results obtained based on a single artificial intelligence method are of low accuracy. In order to solve this problem, in the embodiment of the present application, the initial parameter group of the load forecasting model is generated in the polar bear algorithm training stage, the initial parameter group is updated in the global optimization stage, and the weights of each load forecasting sub-model are determined in the linear integration stage to obtain a load forecasting model composed of a mixture of multiple load forecasting sub-models. The calculation method and data processing technology are effectively combined, so that the diversity and generalization of the load forecasting model are higher, and further, the accuracy of the short-term load forecasting results of the power system obtained based on the load forecasting model according to the historical load data is higher.

[0105] It should be noted that due to the difference between users' electricity consumption behavior during holidays and on weekdays, the load characteristics of the power system change. In order to improve the accuracy of short-term load forecasting results, in an embodiment of the present application, the acquired historical load data can also be clustered to divide the historical load data into multiple groups of historical load data sets according to load characteristics, and then use the load forecasting model to perform load forecasting operations on each historical load data set, so that the predicted short-term load forecasting results have the same user power consumption behavior as the corresponding historical load data set, thereby improving the accuracy of the short-term load forecasting results.

[0106] Compared with the prior art, the embodiment of the present application generates an initial parameter group of the load forecasting model through the polar bear algorithm training phase, and updates the initial parameter group through the global optimization phase, thereby obtaining multiple optimal load forecasting sub-models; further, the weights of each optimal load forecasting sub-model are determined through the linear integration phase to obtain a load forecasting model composed of a mixture of multiple load forecasting sub-models, which can fully utilize the advantages of each sub-model to improve the prediction accuracy of the load forecasting model, so that the accuracy of the short-term load forecasting results obtained based on the load forecasting model according to the historical load data set is improved.

[0107] The present application embodiment provides another short-term load forecasting method, such as Figure 2 As shown, the method includes:

[0108] 201. Perform clustering processing on historical load data.

[0109] It should be noted that due to the difference between the electricity consumption behavior of users on holidays and on weekdays, the load characteristics of the power system change. In order to improve the accuracy of the short-term load forecast results, in the embodiment of the present application, the acquired historical load data is clustered in advance to divide the historical load data into multiple groups of historical load data sets according to the load characteristics. Furthermore, the load forecasting model is used to perform load forecasting operations according to each historical load data set, so that the predicted short-term load forecast results have the same user power consumption behavior as the corresponding historical load data set, thereby improving the accuracy of the short-term load forecast results. Specifically, an improved FCM clustering algorithm can be used. The FCM clustering algorithm aims to find the cluster center of each category to minimize the objective function

[0110]

[0111] Among them, c represents the number of categories, u ij Represents the membership degree of the sample (that is, the degree to which the i-th sample belongs to the j-th cluster center, satisfying m represents the fuzzy factor (it can take the value of 2, indicating that the historical load data is divided into two categories: holidays and working days), d eu represents the Euclidean distance, x i represents the i-th sample, n represents the number of samples, v j represents the jth cluster center. It should be noted that due to the Euclidean distance d eu represents the actual distance between two points in the P-dimensional space, ignoring the trend of the time series, so it is impossible to accurately perform clustering. Based on this, in the embodiment of the present application, the Pearson distance d is used pearson Instead of the Euclidean distance d eu , since the Pearson distance uses covariance to describe the longitudinal shape similarity between load curves, it can reduce the similarity between objects classified into different groups and maximize the similarity between load curves. The Pearson distance can be calculated according to the following formula

[0112]

[0113] d pearson (x i ,v j )=1-R(x i ,v j )

[0114] Among them, x ik represents the k-th dimension value of the i-th sample, represents the average value of all dimensions of the i-th sample, v jk represents the kth dimension value of the jth cluster center, represents the average value of all dimensions of the jth cluster center, and p represents the number of samples.

[0115] Accordingly, step 201 of the embodiment specifically includes: calculating the initial cluster center v of the historical load data according to the initial membership degree of each sample based on the following formula: j ,

[0116]

[0117] Among them, u ij represents the initial membership of the sample, m represents the fuzzy factor, x i represents the i-th sample, and n represents the number of samples; calculate the Pearson distance between each sample and the initial cluster center; update each initial membership according to the Pearson distance based on the following formula to obtain the updated membership

[0118]

[0119] Among them, d pearson represents the Pearson distance function, c represents the number of categories, x j represents the jth sample, v k represents the kth cluster center; based on the following formula, the initial cluster center is updated according to the updated membership degree to obtain the updated cluster center

[0120]

[0121] Based on the above steps, the cluster center is iteratively updated until the preset accuracy is met, and multiple groups of historical load data sets are obtained, and the user power consumption behavior of each historical load data set is different.

[0122] 202. Generate an initial parameter set of the load forecasting model through the polar bear algorithm training phase.

[0123] In order to consider the correlation of time series, in the embodiment of the present application, an initial load forecasting model is constructed based on LSTM (Long Short Term Memory) neural network. Specifically, LSTM is mainly composed of three gates at each time step, namely, forget gate f t , input gate i t and output gate o t , if the LSTM input is represented as p t , the generated hidden output is expressed as Then there is an LSTM block at each time step, and the storage unit c in the LSTM module t Will interact with the following inputs to transfer important information throughout the time series, which can be described by the following mathematical formula,

[0124] f t =σ(w f ·[h t-1 ,p t ]+b f )

[0125] i t =σ(w i ·[h t-1 ,p t ]+b i )

[0126] o t =σ(w o ·[h t-1 ,p t ]+b o )

[0127]

[0128] h t =o t ⊙tanh(c t )

[0129] in, represents the candidate storage unit, σ represents the activation function, w f 、w i 、w o 、w c Both represent learnable weights, b f , b i , b o , b c All represent learnable deviations. Then, the initial parameter group of the load forecasting model is generated through the polar bear algorithm training phase. Specifically, first, the initialization position of each individual polar bear is randomly selected. Secondly, a global search is performed. Specifically, a global search is performed according to the individual polar bear movement behavior formula to update the initialization position and obtain the global optimal position of each individual polar bear. The individual polar bear movement behavior formula is as follows:

[0130]

[0131] in, represents the motion of the i-th polar bear at coordinate j in the t-th iteration, x s represents the motion coordinates of the polar bear, w represents the Euclidean distance between the current position of the polar bear and the best polar bear in the population, γ represents a random number in the interval [0, w], sign(w) represents a piecewise function, and according to the value of w, the result of the sign(w) function is {-1, 0, 1}, α represents a random number in the interval (0, 1], and w can be calculated according to the following formula,

[0132]

[0133] Among them, n represents the individual dimension, represents the best individual in the jth dimension, Represents the jth dimension of the current individual. Further, a local search is performed. Specifically, a local search is performed according to the search radius formula to update the global best position and obtain the best position to complete the local search. The search radius formula is as follows:

[0134]

[0135] Among them, a represents the field of vision of the polar bear, represents a random number in the interval (0,π / 2); the following equation for each spatial coordinate describes each individual movement.

[0136]

[0137] in, represents the jth coordinate of the polar bear individual after local movement at the tth iteration, j∈[0,n-1], represents the jth coordinate of the polar bear individual before local movement at the tth iteration, j∈[0,n-1], represents a random number uniformly distributed from [0,π], as well as Both represent random numbers uniformly distributed in [0,π]. + means the polar bear searches forward, and - means the polar bear searches backward.

[0138] Accordingly, step 202 of the embodiment specifically includes: constructing an initial load forecasting model based on an LSTM neural network; randomly selecting the initialization position of each individual polar bear; performing a global search according to the individual polar bear movement behavior formula to update each initialization position and obtain the global optimal position of each individual polar bear. The individual polar bear movement behavior formula is as follows: in, represents the motion of the i-th polar bear at coordinate j in the t-th iteration, x s represents the motion coordinates of the polar bear, w represents the Euclidean distance between the current position of the polar bear and the best polar bear in the population, γ represents a random number in the interval [0, w], sign(w) represents a piecewise function, and the result of the sign(w) function is {-1, 0, 1} according to the value of w, and α represents a random number in the interval (0, 1); perform local search according to the search radius formula to update each global optimal position and obtain the optimal position to complete the local search. The search radius formula is as follows: Among them, a represents the field of vision of the polar bear, represents a random number in the interval (0,π / 2); repeat the above global search steps and local search steps until the preset number of iterations is met to obtain the global optimal position of each polar bear individual.

[0139] 203. The global optimal positions are grouped based on a grouping algorithm; when the group members and the number of groups remain unchanged, the grouping is stopped to obtain an initial parameter group of the load forecasting model.

[0140] In order to ensure the diversity of the obtained neural network architecture and reduce the amount of calculation involved in finding higher quality solutions in the subsequent global optimization stage, in the embodiment of the present application, each global optimal position is grouped by a clustering algorithm. Specifically, the obtained members are clustered into one or more high-quality sub-areas. When the group members and the number of groups remain unchanged, all individuals are considered to have reached a consensus and the clustering is stopped. Finally, the best three points and the weighted centroid in each cluster are substituted into the neural network as high-quality initial solutions to generate the initial parameter set of the load forecasting model.

[0141] 204. Update the initial parameter set through the global optimization phase.

[0142] In the embodiment of the present application, the quality of the training parameters of the load forecasting model is improved by constructing a power system F(x), and the optimization problem of the load forecasting model parameters is transformed into a problem of finding a stable balance of the nonlinear power system. The local optimal solution is systematically and layer by layer escaped and the global optimal solution is found. The power system can be expressed by the following formula:

[0143]

[0144] Among them, DH loss (x s ) T Represents the Jacobian matrix of the loss function, H loss (x s ) represents the loss function of the neural network. Specifically, based on the initial parameter set obtained in step 203 of the embodiment Represents the Tier-0 solution. From the Tier-0 solution, we search for the overflow point of the loss function along the preset search direction. During the search process, the loss function keeps increasing. When the loss function starts to decrease, the turning point is the overflow point of the loss function, indicating that the search path approaches the adjacent stable area and crosses the stable boundary. The search path diagram in the global optimization stage is as follows: Figure 3 At this time, the polar bear algorithm in step 203 of the embodiment can be used to converge to the best position in the adjacent stable area. Then we get the Tier-1 solution Finally, the solution with the smallest loss function among Tier-0 and Tier-1 is selected as the optimal solution to obtain the updated parameter set.

[0145] Correspondingly, step 204 of the embodiment specifically includes: searching for the overflow point of the loss function along a preset search direction based on the initial parameter group to determine an adjacent stable area; determining the best position in the adjacent stable area based on the polar bear algorithm; and selecting the optimal solution from the best position in the adjacent stable area and the initial parameter group based on the loss function to obtain an updated parameter group.

[0146] 205. The weights of each load forecasting sub-model are determined through the linear integration stage.

[0147] In the present application embodiment, firstly according to the formula Calculate the weights of each load forecasting sub-model, and then linearly combine each load forecasting sub-model and the corresponding weights to obtain the load forecasting model h load =W s (i)*h t (i)i=1,2,...,n.

[0148] Accordingly, step 205 of the embodiment specifically includes: calculating the weight of each load forecasting sub-model in the updated parameter group based on the following formula: Among them, H lass (i) represents the loss function of the i-th load forecasting sub-model; each load forecasting sub-model and the corresponding weight are linearly combined to obtain the load forecasting model.

[0149] The present application provides a short-term load forecasting method, firstly, the historical load data of the power system is obtained; secondly, based on the pre-built load forecasting model, a load forecasting operation is performed according to the historical load data to obtain the short-term load forecasting result of the power system, the load forecasting model is composed of a mixture of multiple load forecasting sub-models, and is obtained in sequence through the polar bear algorithm training stage, the global optimization stage, and the linear integration stage, the polar bear algorithm training stage is used to generate the initial parameter group of the load forecasting model, the global optimization stage is used to update the initial parameter group, and the linear integration stage is used to determine the weight of each load forecasting sub-model. Compared with the prior art, the embodiment of the present application generates the initial parameter group of the load forecasting model through the polar bear algorithm training stage, and updates the initial parameter group through the global optimization stage, so as to obtain multiple optimal load forecasting sub-models; further, the weights of each optimal load forecasting sub-model are determined through the linear integration stage to obtain a load forecasting model composed of a mixture of multiple load forecasting sub-models, which can make full use of the advantages of each sub-model to improve the prediction accuracy of the load forecasting model, so that the accuracy of the short-term load forecasting result obtained by predicting the historical load data set based on the load forecasting model is improved.

[0150] Furthermore, as a response to the above Figure 1 The implementation of the method shown in the figure, the embodiment of the present application provides a short-term load forecasting device, such as Figure 4 As shown, the device comprises:

[0151] Load data acquisition module 31, short-term load forecasting module 32.

[0152] A load data acquisition module 31 is used to acquire historical load data of the power system;

[0153] The short-term load forecasting module 32 is used to perform load forecasting operations according to the historical load data based on a pre-constructed load forecasting model to obtain a short-term load forecasting result of the power system. The load forecasting model is composed of a mixture of multiple load forecasting sub-models, and is obtained through the polar bear algorithm training stage, the global optimization stage, and the linear integration stage in sequence. The polar bear algorithm training stage is used to generate an initial parameter group of the load forecasting model, the global optimization stage is used to update the initial parameter group, and the linear integration stage is used to determine the weights of each of the load forecasting sub-models.

[0154] Preferably, before the short-term load forecasting module, the device further includes a training module for:

[0155] Construct an initial load forecasting model based on LSTM neural network;

[0156] Randomly select the initial position of each individual polar bear;

[0157] A global search is performed according to the individual polar bear movement behavior formula to update each of the initialization positions to obtain the global optimal position of each of the individual polar bears. The individual polar bear movement behavior formula is as follows:

[0158]

[0159] in, represents the motion of the i-th polar bear at coordinate j in the t-th iteration, x s represents the motion coordinates of the polar bear, w represents the Euclidean distance between the current position of the polar bear and the best polar bear in the population, γ represents a random number in the interval [0, w], sign(w) represents a piecewise function, and the result of the sign(w) function is {-1, 0, 1} according to the value of w, and α represents a random number in the interval (0, 1);

[0160] A local search is performed according to a search radius formula to update each of the global optimal positions to obtain the optimal position for completing the local search. The search radius formula is as follows:

[0161]

[0162] Among them, a represents the field of vision of the polar bear, Represents a random number in the interval (0,π / 2);

[0163] Repeat the above global search step and local search step until the preset number of iterations is met to obtain the global optimal position of each polar bear individual.

[0164] Preferably, after the training module, the device further comprises an optimization module for:

[0165] Grouping the global optimal positions based on a grouping algorithm;

[0166] When the group members and the number of groups remain unchanged, the grouping is stopped to obtain the initial parameter group of the load forecasting model.

[0167] Preferably, after obtaining the initial parameter group of the load forecasting model, the method further includes:

[0168] Searching for a loss function overflow point along a preset search direction based on the initial parameter group to determine an adjacent stable region;

[0169] Determine the best position within the adjacent stable area based on the polar bear algorithm;

[0170] An optimal solution is selected from the best position in the adjacent stable region and the initial parameter group based on a loss function to obtain an updated parameter group.

[0171] Preferably, after the optimization module, the device further includes an integration module for:

[0172] The weights of each load forecasting sub-model in the updated parameter group are calculated based on the following formula:

[0173]

[0174] Among them, H lass (i) represents the loss function of the i-th load forecasting sub-model;

[0175] The load forecasting sub-models and corresponding weights are linearly combined to obtain a load forecasting model.

[0176] Preferably, after the load data acquisition module, the device further includes a clustering module for:

[0177] The initial cluster center v of the historical load data is calculated based on the initial membership of each sample based on the following formula: j ,

[0178]

[0179] Among them, u ij represents the initial membership of the sample, m represents the fuzzy factor, x i represents the i-th sample, and n represents the number of samples;

[0180] Calculating the Pearson distance between each of the samples and the initial cluster center;

[0181] Based on the following formula, each of the initial memberships is updated according to the Pearson distance to obtain an updated membership.

[0182]

[0183] Among them, d pearson represents the Pearson distance function, c represents the number of categories, x j represents the jth sample, v k represents the kth cluster center;

[0184] The initial cluster center is updated according to the updated membership degree based on the following formula to obtain the updated cluster center

[0185]

[0186] Based on the above steps, the cluster center is iteratively updated until the preset accuracy is met, and multiple groups of historical load data sets are obtained, and the user power consumption behavior of each of the historical load data sets is different;

[0187] The load forecasting operation based on the pre-built load forecasting model is performed according to the historical load data to obtain the short-term load forecasting result of the power system, specifically including:

[0188] Based on the pre-built load forecasting model, load forecasting operations are performed respectively according to each of the historical load data sets to obtain multiple groups of short-term load forecasting results, and the short-term load forecasting results have the same user power consumption behavior as the corresponding historical load data sets.

[0189] The present application provides a short-term load forecasting device, which first obtains the historical load data of the power system; secondly, based on the pre-built load forecasting model, performs load forecasting operations according to the historical load data to obtain the short-term load forecasting result of the power system, wherein the load forecasting model is composed of a mixture of multiple load forecasting sub-models, and is obtained in sequence through the polar bear algorithm training stage, the global optimization stage, and the linear integration stage, wherein the polar bear algorithm training stage is used to generate the initial parameter group of the load forecasting model, the global optimization stage is used to update the initial parameter group, and the linear integration stage is used to determine the weights of each of the load forecasting sub-models. Compared with the prior art, the embodiment of the present application generates the initial parameter group of the load forecasting model through the polar bear algorithm training stage, and updates the initial parameter group through the global optimization stage, so as to obtain multiple optimal load forecasting sub-models; further, the weights of each optimal load forecasting sub-model are determined through the linear integration stage to obtain a load forecasting model composed of a mixture of multiple load forecasting sub-models, which can make full use of the advantages of each sub-model to improve the prediction accuracy of the load forecasting model, so that the accuracy of the short-term load forecasting result obtained by predicting the historical load data set based on the load forecasting model is improved.

[0190] According to an embodiment of the present application, a storage medium is provided, wherein the storage medium stores at least one executable instruction, and the computer executable instruction can execute the interface access test method in any of the above method embodiments.

[0191] Based on this understanding, the technical solution of the present application can be embodied in the form of a software product, which can be stored in a non-volatile storage medium (which can be a CD-ROM, a USB flash drive, a mobile hard disk, etc.), and includes a number of instructions for enabling a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods described in each implementation scenario of the present application.

[0192] Figure 5 A schematic diagram of the structure of a terminal provided according to an embodiment of the present application is shown, and the specific embodiment of the present application does not limit the specific implementation of the terminal.

[0193] like Figure 5 As shown, the terminal may include: a processor (processor) 402 , a communication interface (Communications Interface) 404 , a memory (memory) 406 , and a communication bus 408 .

[0194] The processor 402 , the communication interface 404 , and the memory 406 communicate with each other via a communication bus 408 .

[0195] The communication interface 404 is used to communicate with other devices such as clients or other servers.

[0196] The processor 402 is used to execute the program 410, and specifically can execute the relevant steps in the above-mentioned short-term load forecasting method embodiment.

[0197] Specifically, the program 410 may include program codes, which include computer operation instructions.

[0198] The processor 402 may be a central processing unit (CPU), or an application-specific integrated circuit (ASIC), or one or more integrated circuits configured to implement the embodiments of the present application. The one or more processors included in the computer device may be processors of the same type, such as one or more CPUs; or processors of different types, such as one or more CPUs and one or more ASICs.

[0199] The memory 406 is used to store the program 410. The memory 406 may include a high-speed RAM memory, and may also include a non-volatile memory (non-volatile memory), such as at least one disk memory.

[0200] The program 410 may be specifically configured to enable the processor 402 to perform the following operations:

[0201] Obtain historical load data of the power system;

[0202] Based on the pre-constructed load forecasting model, load forecasting operations are performed according to the historical load data to obtain a short-term load forecasting result of the power system. The load forecasting model is composed of a mixture of multiple load forecasting sub-models, and is obtained through the polar bear algorithm training stage, the global optimization stage, and the linear integration stage in sequence. The polar bear algorithm training stage is used to generate an initial parameter group for the load forecasting model, the global optimization stage is used to update the initial parameter group, and the linear integration stage is used to determine the weights of each of the load forecasting sub-models.

[0203] The storage medium may also include an operating system and a network communication module. The operating system is a program that manages the hardware and software resources of the physical device of the short-term load forecasting method, and supports the operation of the information processing program and other software and / or programs. The network communication module is used to realize the communication between the components inside the storage medium, and the communication with other hardware and software in the information processing physical device.

[0204] Each embodiment in this specification is described in a progressive manner, and each embodiment focuses on the differences from other embodiments. The same or similar parts between the embodiments can be referred to each other. For the system embodiment, since it basically corresponds to the method embodiment, the description is relatively simple, and the relevant parts can be referred to the partial description of the method embodiment.

[0205] The method and system of the present application may be implemented in many ways. For example, the method and system of the present application may be implemented by software, hardware, firmware, or any combination of software, hardware, and firmware. The above order of steps for the method is for illustration only, and the steps of the method of the present application are not limited to the order specifically described above, unless otherwise specifically stated. In addition, in some embodiments, the present application may also be implemented as a program recorded in a recording medium, which includes machine-readable instructions for implementing the method according to the present application. Therefore, the present application also covers a recording medium storing a program for executing the method according to the present application.

[0206] Obviously, those skilled in the art should understand that the above modules or steps of the present application can be implemented by a general computing device, they can be concentrated on a single computing device, or distributed on a network composed of multiple computing devices, and optionally, they can be implemented by a program code executable by a computing device, so that they can be stored in a storage device and executed by the computing device, and in some cases, the steps shown or described can be executed in a different order from that herein, or they can be made into individual integrated circuit modules, or multiple modules or steps therein can be made into a single integrated circuit module for implementation. Thus, the present application is not limited to any specific combination of hardware and software.

[0207] The above description is only the preferred embodiment of the present application and is not intended to limit the present application. For those skilled in the art, the present application may have various modifications and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application shall be included in the protection scope of the present application.

Claims

1. A short-term load forecasting method, characterized in that: include: Obtain historical load data of the power system; Based on the pre-constructed load forecasting model, load forecasting operations are performed according to the historical load data to obtain a short-term load forecasting result of the power system. The load forecasting model is composed of a mixture of multiple load forecasting sub-models, and is obtained through the polar bear algorithm training stage, the global optimization stage, and the linear integration stage in sequence. The polar bear algorithm training stage is used to generate an initial parameter group for the load forecasting model, the global optimization stage is used to update the initial parameter group, and the linear integration stage is used to determine the weights of each of the load forecasting sub-models.

2. The method according to claim 1, characterized in that Before performing a load forecasting operation based on the historical load data based on the pre-built load forecasting model to obtain a short-term load forecasting result of the power system, the method further includes: Construct an initial load forecasting model based on LSTM neural network; Randomly select the initial position of each individual polar bear; A global search is performed according to the individual polar bear movement behavior formula to update each of the initialization positions to obtain the global optimal position of each of the individual polar bears. The individual polar bear movement behavior formula is as follows: in, represents the motion of the i-th polar bear at coordinate j in the t-th iteration, x s represents the motion coordinates of the polar bear, w represents the Euclidean distance between the current position of the polar bear and the best polar bear in the population, γ represents a random number in the interval [0, w], sign(w) represents a piecewise function, and the result of the sign(w) function is {-1, 0, 1} according to the value of w, and α represents a random number in the interval (0, 1); A local search is performed according to a search radius formula to update each of the global optimal positions to obtain the optimal position for completing the local search. The search radius formula is as follows: Among them, a represents the field of vision of the polar bear, Represents a random number in the interval (0,π / 2); Repeat the above global search step and local search step until the preset number of iterations is met to obtain the global optimal position of each polar bear individual.

3. The method according to claim 2, characterized in that After obtaining the global optimal position, the method further comprises: Grouping the global optimal positions based on a grouping algorithm; When the group members and the number of groups remain unchanged, the grouping is stopped to obtain the initial parameter group of the load forecasting model.

4. The method according to claim 3, characterized in that: After obtaining the initial parameter group of the load forecasting model, the method further includes: Searching for a loss function overflow point along a preset search direction based on the initial parameter group to determine an adjacent stable region; Determine the best position within the adjacent stable area based on the polar bear algorithm; An optimal solution is selected from the best position in the adjacent stable region and the initial parameter group based on a loss function to obtain an updated parameter group.

5. The method according to claim 4, characterized in that After selecting the optimal solution from the best position in the adjacent stable region and the initial parameter group based on the loss function to obtain the updated parameter group, the method further includes: The weights of each load forecasting sub-model in the updated parameter group are calculated based on the following formula: Among them, H lass (i) represents the loss function of the i-th load forecasting sub-model; The load forecasting sub-models and corresponding weights are linearly combined to obtain a load forecasting model.

6. The method according to claim 1, characterized in that After acquiring the historical load data of the power system, the method further includes: The initial cluster center v of the historical load data is calculated based on the initial membership of each sample based on the following formula: j , Among them, u ij represents the initial membership of the sample, m represents the fuzzy factor, x i represents the i-th sample, and n represents the number of samples; Calculating the Pearson distance between each of the samples and the initial cluster center; Based on the following formula, each of the initial memberships is updated according to the Pearson distance to obtain an updated membership. Among them, d pearson represents the Pearson distance function, c represents the number of categories, x j represents the jth sample, v k represents the kth cluster center; The initial cluster center is updated according to the updated membership degree based on the following formula to obtain the updated cluster center Based on the above steps, the cluster center is iteratively updated until the preset accuracy is met, and multiple groups of historical load data sets are obtained, and the user power consumption behavior of each of the historical load data sets is different; The load forecasting operation based on the pre-built load forecasting model is performed according to the historical load data to obtain the short-term load forecasting result of the power system, specifically including: Based on the pre-built load forecasting model, load forecasting operations are performed respectively according to each of the historical load data sets to obtain multiple groups of short-term load forecasting results, and the short-term load forecasting results have the same user power consumption behavior as the corresponding historical load data sets.

7. A short-term load forecasting device, characterized in that: include: A load data acquisition module is used to acquire historical load data of the power system; The short-term load forecasting module is used to perform load forecasting operations according to the historical load data based on a pre-built load forecasting model to obtain a short-term load forecasting result for the power system. The load forecasting model is composed of a mixture of multiple load forecasting sub-models, and is obtained through the polar bear algorithm training stage, the global optimization stage, and the linear integration stage in sequence. The polar bear algorithm training stage is used to generate an initial parameter group for the load forecasting model, the global optimization stage is used to update the initial parameter group, and the linear integration stage is used to determine the weights of each of the load forecasting sub-models.

8. A storage medium, wherein at least one executable instruction is stored in the storage medium, characterized in that: The executable instructions enable the processor to perform operations corresponding to the short-term load forecasting method according to any one of claims 1-6.

9. A terminal, comprising: A processor, a memory, a communication interface and a communication bus, wherein the processor, the memory and the communication interface communicate with each other via the communication bus; The memory is used to store at least one executable instruction, wherein the executable instruction enables the processor to perform operations corresponding to the short-term load forecasting method according to any one of claims 1-6.

Images