Load prediction method and terminal equipment

Abstract

The invention relates to the technical field of electric power, in particular to a load prediction method and terminal equipment. Comprising the following steps: forming a low-frequency data set and a high-frequency data set; modeling the low-frequency data set and the high-frequency data set to obtain a low-frequency component predicted value and a high-frequency component predicted value; weighting the low-frequency component predicted value and the high-frequency component predicted value to form a weight matrix; reconstructing a predicted value of each node component in the urban traffic network topological structure model through a weight matrix, and generating a prediction result; the low-frequency component prediction value and the high-frequency component prediction value are formed by predicting the low-frequency data set and the high-frequency data set respectively, and the final prediction value is formed by weighting and integrating, so that node load data can be predicted more accurately, and power grid business personnel can deploy power grid resources more accurately.

Description

Technical Field

[0001] The present invention relates to the field of electric power technology, and more specifically, to a load forecasting method and terminal equipment. Background Art

[0002] Extreme weather has always been one of the major threats to the safe and stable operation of power grids due to its low probability and high destructiveness. Load forecasting before a disaster occurs and allocating emergency resources based on it can effectively improve the power grid company's disaster prevention and mitigation capabilities in response to extreme weather, thereby reducing economic losses.

[0003] In the prior art, load forecasting methods are mainly divided into two categories, single forecasting and combined forecasting. The former uses a single mathematical model or method to forecast the load, which cannot capture complex influencing factors and nonlinear relationships. There are two types of combined forecasting. One is to use multiple groups of single methods to forecast, and then assign weights to the forecast results to obtain the final forecast results, which has poor scalability; the second tends to decompose the load first and then forecast. Load forecasting under extreme weather has a significant impact on the temporal and spatial characteristics of the load, thus creating new requirements for load forecasting. The instantaneous and destructive nature of extreme disasters may lead to local load interruptions in the region, causing sudden load changes. Summary of the invention

[0004] The purpose of the present invention is to provide a load forecasting method and terminal equipment, which first decomposes the load and then allocates weights based on the temporal and spatial characteristics of the load nodes for forecasting, so as to more accurately predict load changes and enable emergency resources to be quickly reconfigured.

[0005] To achieve the above object, the present invention adopts the following technical solution:

[0006] In a first aspect, the present application provides a load forecasting method, comprising the following steps:

[0007] S100, based on the geographical range data, forms a topological structure model of the urban transportation network;

[0008] S200, dividing the urban traffic network topology model into regions according to historical load data;

[0009] S300, performing frequency domain decomposition on total data of each area of ​​the urban traffic network topology structure model according to a Fourier function to form a low-frequency data set and a high-frequency data set;

[0010] S400, obtaining a low-frequency component prediction value and a high-frequency component prediction value by modeling the low-frequency data set and the high-frequency data set;

[0011] S500, weighting the low-frequency component prediction value and the high-frequency component prediction value to form a weight matrix;

[0012] S600, reconstructing the predicted value of each node component in the urban traffic network topology model through a weight matrix to generate a prediction result.

[0013] Specifically, the Fourier function is:

[0014]

[0015] Among them, P(t) is the continuous load sequence, a0 is the initialization amount, a i and b i is the Fourier coefficient, w i is the angular frequency, t is the time, and nT is the nth period.

[0016] Specifically, obtaining the low-frequency component prediction value includes the following steps:

[0017] L100, for low-frequency data sets, the LSSVM model is used for modeling;

[0018] L200, optimize the parameters of the LSSVM model by Rosenbrock algorithm to obtain the low-frequency component prediction value.

[0019] Specifically, the LSSVM model is for the sample (x i ,y i ) is established, and the function is to minimize the Lagrangian function L:

[0020]

[0021] Among them, x i ,y i are the x value and y value of the ith sample in the low-frequency data set, W is the weight matrix, C is the regularization parameter, m is the number of samples, and e i is the error variable, α i For the corresponding x i The Lagrange multiplier of To map to a higher dimensional region, b is the bias coefficient.

[0022] Specifically, the parameter optimization of the LSSVM model by using the Rosenbrock algorithm includes the following steps:

[0023] L210, parameter initialization;

[0024] Given an initial point x (1) ∈R n , unit orthogonal direction d (1) d(2) …d (n) , step length The magnification factor μ>1, the reduction factor β∈(-1,0) and the allowable error ε>0, and y (1) =x (1) , Where i = 1, 2, ... n;

[0025] L220, Calculation The value of Is it less than f(y (i) ), if yes, then let If not, let y (j+1) =y (j) ,

[0026] L230, determine whether j reaches n, if yes, execute step L240; if no, j=j+1, execute step L220;

[0027] L240, judge f(y (n+1) ) is less than f(y (1) ), if yes, then let y (1) =y (n+1) , j=1, execute step L220; if not, execute step L250;

[0028] L250, judge f(y (n+1) ) is less than f(x (k) ), if yes, execute step L260; otherwise, execute step L270;

[0029] L260, let x (k+1) =y (n+1) , judge ||x (k+1) -x (k) ||≤ε, if yes, then take x (k+1) as the predicted value; if not, calculate λ1, λ2…λ n , and construct new orthogonal directions through the Schmidt orthogonalization model And order Let y (1) =x (k+1) , k=k+1, j=1, execute step L220;

[0030] L270, determine for j = 1, 2 ... n, Is it true? If so, then x (k) as the predicted value; if not, let y = y (n+1) , j=1, execute step L220.

[0031] Specifically, the Schmidt orthogonalization model is:

[0032]

[0033] Among them, d (j) is the unit orthogonal direction, λ i , j is the process variable.

[0034] Optionally, the LSSVM model further includes a constraint model, and the constraint model is:

[0035]

[0036] in, is a vector of all 1s, I is the identity matrix, K is the radial basis kernel function, σ is the normalization parameter, ||xx i || is the vector xx i paradigm.

[0037] Specifically, a high-frequency data set is modeled by a Transformer model to obtain a high-frequency data prediction model, and a high-frequency data prediction value is obtained by the high-frequency data prediction model; the high-frequency data set is divided into a training set and a test set, the training set is used to train the Transformer model, and the test set is used to verify the accuracy of the model; the Transformer model includes an attention module, and the function of the attention module is:

[0038]

[0039] Among them, key i For the i-th key, query j For the jth Query, e ij is the attention score of the i-th key and the j-th query, a is the function for calculating similarity, a ij is the normalized e ij ,value i is the tensor to be weighted, c j is the tensor obtained after weighting;

[0040] The Transformer model also includes a loss module for training the Transformer model, and the function of the loss module is:

[0041]

[0042] in, is the predicted value of the i-th sample, y i is the true value of the ith sample, t is the length of the input sequence, and λ is a hyperparameter.

[0043] Specifically, weighting the low-frequency component prediction value and the high-frequency component prediction value to form a weight matrix includes the following steps:

[0044] According to the weight model, the weight matrix W is assigned to the low-frequency and high-frequency data of each node. low and W high , the weight model is:

[0045]

[0046] Among them, S(t) is the total load signal of the node, L(t) is the low-frequency load data, H(t) is the high-frequency load data, a is the proportional parameter, and b is the reference parameter;

[0047] The weight model is normalized to form a normalized model, and the normalized model is:

[0048]

[0049] According to the normalized model, W is calculated high and W low ;

[0050] Through the inverse normalization model, the low-frequency component prediction value and high frequency component prediction value After the denormalization process, the position weight value is superimposed according to the weight model, and finally the normalization process is performed to obtain the final prediction result. The denormalization model is:

[0051]

[0052] Among them, f inv is the deserialization function, is the predicted value of low-frequency component, is the predicted value of high frequency component.

[0053] On the other hand, the present application also provides a load forecasting system, comprising:

[0054] Basic data unit, used to form a topological structure model of the urban transportation network based on geographical range data;

[0055] A regional division unit, used for dividing the urban traffic network topology model into regions according to historical load data;

[0056] A frequency domain decomposition unit performs frequency domain decomposition on the total data of each area of ​​the urban traffic network topological structure model according to the Fourier function to form a low-frequency data set and a high-frequency data set;

[0057] An analysis unit, configured to obtain a low-frequency component prediction value and a high-frequency component prediction value by modeling the low-frequency data set and the high-frequency data set;

[0058] A weighting unit, used for weighting the low-frequency component prediction value and the high-frequency component prediction value to form a weight matrix;

[0059] The data prediction unit is used to reconstruct the predicted value of each node component in the urban traffic network topology model through a weight matrix to generate a prediction result.

[0060] Through the above method, a topological structure model of the urban transportation network is formed according to the geographical range data, and frequency domain decomposition is performed based on the historical load data of the nodes. The corresponding prediction models are used to predict the data for high-frequency data sets and low-frequency data sets. At the same time, the final prediction value is obtained by weighted integration of the prediction values. The node can be quickly predicted in combination with the spatiotemporal load information, and provided to the distribution network business personnel for reference, so that the distribution network business personnel can quickly allocate resources to quickly restore the operation of the power grid system. BRIEF DESCRIPTION OF THE DRAWINGS

[0061] The accompanying drawings constituting a part of the present invention are used to provide a further understanding of the present invention. The exemplary embodiments of the present invention and their descriptions are used to explain the present invention and do not constitute an improper limitation of the present invention. In the accompanying drawings:

[0062] Figure 1 It is a flow chart of load forecasting method;

[0063] Figure 2 This is a schematic diagram of the load forecasting system architecture;

[0064] Figure 3 This is a schematic diagram of the terminal device structure. DETAILED DESCRIPTION

[0065] The present invention will be described in detail below with reference to the accompanying drawings and in combination with embodiments. It should be noted that the embodiments and features in the embodiments of the present invention can be combined with each other without conflict.

[0066] The following detailed description is an exemplary description, which is intended to provide further detailed description of the present invention. Unless otherwise specified, all technical terms used in the present invention have the same meaning as those generally understood by those skilled in the art to which the present invention belongs. The terms used in the present invention are only for describing specific embodiments, and are not intended to limit the exemplary embodiments according to the present invention.

[0067] Reference Figure 1 The present invention provides a load forecasting method, comprising the following steps:

[0068] S100, based on the geographical range data, forms a topological structure model of the urban transportation network;

[0069] Specifically, the geographic scope data includes node data and road data. The node data includes location information such as public service facilities, transportation infrastructure and material warehouses. The road data is the roads connecting the nodes. The node data is defined as nodes and the road data is defined as edges. The adjacency matrix representation is used to form a topological structure model of the urban transportation network.

[0070] S200, dividing the urban traffic network topology model into regions according to historical load data;

[0071] Specifically, the historical load data is the historical load size of each node in the urban traffic network topology structure model, and a level division threshold is preset. The urban traffic network topology structure model is divided into regions according to the threshold interval of the historical load size of each node.

[0072] S300, performing frequency domain decomposition on the total data T of each area of ​​the urban traffic network topology structure model according to the Fourier function to form a low-frequency data set T1 and a high-frequency data set T2;

[0073] Specifically, the Fourier function is:

[0074]

[0075] Among them, P(t) is the continuous load sequence, a0 is the initialization amount, a i and b i is the Fourier coefficient, w i is the angular frequency, t is the time, nT is the nth cycle.

[0076] Through the Fourier function, the load sequence P(t) of a period of time t is subjected to Fourier transform to obtain a series of mutually orthogonal sine and cosine components after the load sequence P(t) is decomposed, and different frequency components are extracted according to the spectrum characteristics to form a low-frequency data set T1 and a high-frequency data set T2.

[0077] S400, obtaining a low-frequency component prediction value by modeling the low-frequency data set T1 and the high-frequency data set T2 and high frequency component prediction value

[0078] Specifically, the low-frequency data set T1 is modeled using the LSSVM model, the target parameters in the LSSVM model are optimized using the Rosenbrock algorithm to obtain a low-frequency data prediction model, and the low-frequency component prediction value is obtained using the low-frequency data prediction model. As a preferred embodiment, the method comprises the following steps:

[0079] L100, for the low-frequency data set T1, the LSSVM model is used to model;

[0080] Specifically, the LSSVM model is for the sample (x i ,y i ) is established, and the function is to minimize the Lagrangian function L:

[0081]

[0082] Among them, x i ,y i are the x value and y value of the ith sample in the low-frequency data set T1, W is the weight matrix, C is the regularization parameter, m is the number of samples, and e i is the error variable, α i For the corresponding x i The Lagrange multiplier of To map to a higher dimensional region, b is the bias coefficient.

[0083] L200, optimizing the parameters of the LSSVM model by Rosenbrock algorithm to obtain the predicted value of the low-frequency component;

[0084] Specifically, the following steps are included:

[0085] L210, parameter initialization;

[0086] Specifically, given the initial point x (1) ∈R n , unit orthogonal direction d (1) d (2) …d (n) Take the coordinate direction and step size The magnification factor μ>1, the reduction factor β∈(-1,0) and the allowable error ε>0, and y (1) =x (1) , Wherein i=1, 2...n, as a preferred implementation, n=2, and k=1, j=1.

[0087] L220, Calculation The value of Is it less than f(y (i) ), if yes, then let If not, let y (j+1) =y (j) ,

[0088] L230, determine whether j reaches n, if yes, execute step L240; if not, j=j+1, execute step L220.

[0089] L240, judge f(y (n+1) ) is less than f(y (1) ), if yes, then let y (1) =y (n+1) , j=1, execute step L220; if not, execute step L250.

[0090] L250, judge f(y (n+1) ) is less than f(x (k) ), if yes, execute step L260; otherwise, execute step L270.

[0091] L260, let x (k+1) =y (n+1) , judge ||x (k+1) -x (k) ||≤ε, if yes, then take x (k+1) as the predicted value; if not, calculate λ1, λ2…λ n , and construct new orthogonal directions through the Schmidt orthogonalization model And order Let y (1) =x (k+1) , k=k+1, j=1, execute step L220;

[0092] Specifically, the Schmidt orthogonalization model is:

[0093]

[0094] Among them, d (j) is the unit orthogonal direction, λ i , j is the process variable.

[0095] L270, determine for j = 1, 2 ... n, Is it true? If so, then x (k) as the predicted value; if not, let y = y (n+1) , j=1, execute step L220.

[0096] As a preferred implementation, the LSSVM model also includes a constraint model, which is:

[0097]

[0098] in, is a vector of all 1s, I is the identity matrix, K is the radial basis kernel function, σ is the normalization parameter, ||xx i || is the vector xx i paradigm.

[0099] When the Rosenbrock algorithm is used to optimize the parameters of the LSSVM model through the constraint model, the objects to be optimized are the regularization parameter C and the standardization parameter σ.

[0100] The high-frequency data set T2 is modeled by the Transformer model to obtain a high-frequency data prediction model, and the high-frequency data prediction value is obtained by the high-frequency data prediction model.

[0101] Specifically, the high-frequency data set T2 is divided into a training set and a test set, the training set is used to train the Transformer model, and the test set is used to verify the accuracy of the model; the Transformer model includes an encoder module, a decoder module, an attention module, and a forward propagation module. The encoder module weights the input sequence features through the attention module, and converts them into a latent vector sequence containing the input information after the forward propagation module; the decoder module weights the target sequence and the latent vector sequence output by the encoder module through the attention module, and outputs the entire load sequence to be predicted in sequence.

[0102] The function of the attention module is:

[0103]

[0104] Among them, key i For the i-th key, query j For the jth Query, e ij is the attention score of the i-th key and the j-th query, a is the function for calculating similarity, a ij is the normalized e ij ,value i is the tensor to be weighted, c j is the tensor obtained after weighting.

[0105] As a preferred implementation, the Transformer model further includes a loss module for training the Transformer model, and the function of the loss module is:

[0106]

[0107] in, is the predicted value of the i-th sample, y iis the true value of the i-th sample, t is the length of the input sequence, λ is a hyperparameter, and the model parameters with the smallest loss are found according to the loss function to obtain the trained high-frequency data prediction model.

[0108] S500, predicted value of low frequency component and high frequency component prediction value Perform weighting to form a weight matrix;

[0109] Specifically, according to the weight model, for the nodes in the traffic network, the corresponding relationship between the environmental variables and the weights is preset, and the weights of the nodes are allocated according to the environmental variables of the nodes to form a weight matrix, which includes a low-frequency weight matrix W formed based on the predicted value of the low-frequency component of each node low and the high-frequency weight matrix W formed based on the high-frequency component prediction value of each node high In this embodiment, for nodes in the transportation network, location weights are assigned according to factors such as distance, terrain, and environment.

[0110] The weight model is:

[0111]

[0112] Among them, S(t) is the total load signal of the node, L(t) is the low-frequency load data, H(t) is the high-frequency load data, a is the proportional parameter, and b is the reference parameter.

[0113] The weight model is normalized to form a normalized model, and the normalized model is:

[0114]

[0115] According to the normalized model, W is calculated high and W low .

[0116] S600, reconstructing the predicted value of each node component in the urban traffic network topology model through a weight matrix to generate a prediction result;

[0117] Specifically, the low-frequency component prediction value is transformed into and high frequency component prediction value After the denormalization process, the position weight value is superimposed according to the weight model, and finally the normalization process is performed to obtain the final prediction result. The denormalization model is:

[0118]

[0119] Among them, f inv is the deserialization function, is the predicted value of low-frequency component, is the predicted value of high frequency component.

[0120] Through the above implementation method, a topological structure model of the urban transportation network is formed according to the geographical scope data, and frequency domain decomposition is performed based on the historical load data of the nodes. The corresponding prediction models are used to predict the data for the high-frequency data set and the low-frequency data set. At the same time, the final prediction value is obtained by weighted integration of the prediction values. The nodes can be quickly predicted in combination with the spatiotemporal load information, and provided to the distribution network business personnel for reference, so that the distribution network business personnel can quickly allocate resources to quickly restore the operation of the power grid system.

[0121] Reference Figure 2 On the other hand, the present application also provides a load forecasting system, comprising:

[0122] Basic data unit, used to form a topological structure model of the urban transportation network based on geographical range data;

[0123] A regional division unit, used for dividing the urban traffic network topology model into regions according to historical load data;

[0124] A frequency domain decomposition unit performs frequency domain decomposition on the total data of each area of ​​the urban traffic network topological structure model according to the Fourier function to form a low-frequency data set and a high-frequency data set;

[0125] An analysis unit, configured to obtain a low-frequency component prediction value and a high-frequency component prediction value by modeling the low-frequency data set and the high-frequency data set;

[0126] A weighting unit, used for weighting the low-frequency component prediction value and the high-frequency component prediction value to form a weight matrix;

[0127] The data prediction unit is used to reconstruct the predicted value of each node component in the urban traffic network topology model through a weight matrix to generate a prediction result.

[0128] Reference Figure 3 , On the other hand, the present application also provides a terminal device, the terminal device comprising: a processor 1, a communication interface 2, a memory 3 and a communication bus 4, wherein the processor 1, the communication interface 2, and the memory 3 communicate with each other via the communication bus 4;

[0129] Memory 3, used for storing computer programs;

[0130] The processor 1 is used to execute the program stored in the memory 3 to implement the steps of the above method embodiment.

[0131] The bus mentioned in the terminal device can be a Peripheral Component Interconnect (PCI) bus or an Extended Industry Standard Architecture (EISA) bus, etc. The bus can be divided into an address bus, a data bus, a control bus, etc. For ease of representation, only one thick line is used in the figure, but it does not mean that there is only one bus or one type of bus.

[0132] The communication interface is used for communication between the above terminal device and other devices.

[0133] The memory may include a random access memory (RAM) or a non-volatile memory (NVM), such as at least one disk memory. Optionally, the memory may also be at least one storage device located away from the aforementioned processor.

[0134] The above-mentioned processor can be a general-purpose processor, including a central processing unit (CPU), a network processor (NP), etc.; it can also be a digital signal processor (DSP), an application specific integrated circuit (ASIC), a field programmable gate array (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components.

[0135] An embodiment of the present application further provides a storage medium, the storage medium includes a stored program, wherein when the program is run, the method steps of the above method embodiment are executed.

[0136] It is known from common technical knowledge that the present invention can be implemented by other embodiments that do not deviate from its spirit or essential features. Therefore, the above disclosed embodiments are only illustrative in all respects and are not exclusive. All changes within the scope of the present invention or within the scope equivalent to the present invention are included in the present invention.

[0137] Those skilled in the art will appreciate that embodiments of the present invention may be provided as methods, systems, or computer program products. Therefore, the present invention may take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Moreover, the present invention may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0138] The present invention is described with reference to flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to embodiments of the present invention. It should be understood that each process and / or block in the flowchart and / or block diagram, as well as the combination of processes and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0139] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0140] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1 A step that specifies a function in one or more boxes.

[0141] Finally, 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 above embodiments, ordinary technicians in the relevant field should understand that the specific implementation methods of the present invention can still be modified or replaced by equivalents. Any modification or equivalent replacement that does not depart from the spirit and scope of the present invention should be covered within the scope of protection of the claims of the present invention.

Claims

1. A load forecasting method, characterized in that: The following steps are involved: S100, based on the geographical range data, forms a topological structure model of the urban transportation network; S200, dividing the urban traffic network topology model into regions according to historical load data; S300, performing frequency domain decomposition on total data of each area of ​​the urban traffic network topology structure model according to a Fourier function to form a low-frequency data set and a high-frequency data set; S400, obtaining a low-frequency component prediction value and a high-frequency component prediction value by modeling the low-frequency data set and the high-frequency data set; S500, weighting the low-frequency component prediction value and the high-frequency component prediction value to form a weight matrix; S600, reconstructing the predicted value of each node component in the urban traffic network topology model through a weight matrix to generate a prediction result.

2. A load forecasting method according to claim 1, characterized in that: The Fourier function is: Among them, P(t) is the continuous load sequence, a0 is the initialization amount, a i and b i is the Fourier coefficient, w i is the angular frequency, t is the time, and nT is the nth period.

3. A load forecasting method according to claim 1, characterized in that: The step of obtaining the low-frequency component prediction value comprises the following steps: L100, for low-frequency data sets, the LSSVM model is used for modeling; L200, optimize the parameters of the LSSVM model by Rosenbrock algorithm to obtain the low-frequency component prediction value.

4. A load forecasting method according to claim 3, characterized in that: The LSSVM model is used for the samples (x i ,y i ) is established, and the function is to minimize the Lagrangian function L: Among them, x i ,y i are the x value and y value of the ith sample in the low-frequency data set, W is the weight matrix, C is the regularization parameter, m is the number of samples, and e i is the error variable, α i For the corresponding x i The Lagrange multiplier of To map to a higher dimensional region, b is the bias coefficient.

5. A load forecasting method according to claim 3, characterized in that: The method of optimizing the parameters of the LSSVM model by using the Rosenbrock algorithm comprises the following steps: L210, parameter initialization; Given an initial point x (1) ∈R n , unit orthogonal direction d (1) ,d (2) …d (n) , step length The magnification factor μ>1, the reduction factor β∈(-1,0) and the allowable error ε>0, and y (1) =x (1) , Where i = 1, 2, ... n; L220, Calculation The value of Is it less than f(y (i) ), if yes, then let If not, let y (j+1) =y (j) , L230, determine whether j reaches n, if yes, execute step L240; if no, j=j+1, execute step L220; L240, judge f(y (n+1) ) is less than f(y (1) ), if yes, then let y (1) =y (n+1) , j=1, execute step L220; if not, execute step L250; L250, judge f(y (n+1) ) is less than f(x (k) ), if yes, execute step L260; otherwise, execute step L270; L260, let x (k+1) =y (n+1) , judge ||x (k+1) -x (k) ||≤ε, if yes, then take x (k+1) as the predicted value; if not, calculate λ1, λ2…λ n , and construct new orthogonal directions through the Schmidt orthogonalization model And order Let y (1) =x (k+1) , k=k+1, j=1, execute step L220; L270, determine for j = 1, 2 ... n, Is it true? If so, then x (k) as the predicted value; if not, let y = y (n+1) , j=1, execute step L220.

6. A load forecasting method according to claim 5, characterized in that: The Schmidt orthogonalization model is: Among them, d (j) is the unit orthogonal direction, λ i , j is the process variable.

7. A load forecasting method according to claim 5, characterized in that: The LSSVM model also includes a constraint model, which is: in, is a vector of all 1s, I is the identity matrix, K is the radial basis kernel function, σ is the normalization parameter, ||xx i || is the vector xx i paradigm.

8. A load forecasting method according to claim 1, characterized in that: The high-frequency data set is modeled by a Transformer model to obtain a high-frequency data prediction model, and a high-frequency data prediction value is obtained by the high-frequency data prediction model; the high-frequency data set is divided into a training set and a test set, the training set is used to train the Transformer model, and the test set is used to verify the accuracy of the model; the Transformer model includes an attention module, and the function of the attention module is: Among them, key i For the i-th key, query j For the jth Query, e ij is the attention score of the i-th key and the j-th query, a is the function for calculating similarity, a ij is the normalized e ij ,value i is the tensor to be weighted, c j is the tensor obtained after weighting; The Transformer model also includes a loss module for training the Transformer model, and the function of the loss module is: in, is the predicted value of the i-th sample, y i is the true value of the i-th sample, t is the length of the input sequence, and λ is a hyperparameter.

9. A load forecasting method according to claim 1, characterized in that: The step of weighting the low-frequency component prediction value and the high-frequency component prediction value to form a weight matrix includes the following steps: Assign a weight matrix W to each node's low-frequency and high-frequency data low and W high : Among them, S(t) is the total load signal of the node, L(t) is the low-frequency load data, H(t) is the high-frequency load data, a is the proportional parameter, and b is the reference parameter; Perform normalization processing to form a normalized model, which is: According to the normalized model, W is calculated high and W low ; Through the anti-normalization model, the low-frequency component prediction value and high frequency component prediction value After the denormalization process, the position weight value is superimposed, and finally the normalization process is performed to obtain the final prediction result. The denormalization model is: Among them, f inv is the deserialization function, is the predicted value of low-frequency component, is the predicted value of high frequency component.

10. A load forecasting system, characterized in that: include: Basic data unit, used to form a topological structure model of the urban transportation network based on geographical range data; A regional division unit, used for dividing the urban traffic network topology model into regions according to historical load data; A frequency domain decomposition unit performs frequency domain decomposition on the total data of each area of ​​the urban traffic network topological structure model according to the Fourier function to form a low-frequency data set and a high-frequency data set; An analysis unit, configured to obtain a low-frequency component prediction value and a high-frequency component prediction value by modeling the low-frequency data set and the high-frequency data set; A weighting unit, used for weighting the low-frequency component prediction value and the high-frequency component prediction value to form a weight matrix; The data prediction unit is used to reconstruct the predicted value of each node component in the urban traffic network topology model through a weight matrix to generate a prediction result.

Images