Electric power load forecasting method, device and medium based on causal characteristics

Abstract

The present invention relates to a method, device and medium for predicting electric load based on causal characteristics. The method comprises: obtaining actual electric load influencing factor data, inputting the actual electric load influencing factor data into a trained electric load prediction model, and outputting an actual electric load prediction result; wherein the training process of the electric load prediction model is: S1, obtaining historical electric load data, S2, preprocessing the historical electric load data; S3, calculating normalized causal entropy; S4, constructing a causal analysis neural network, multiplying the cause sequence and the corresponding causal strength in the causal analysis neural network, optimizing the causal analysis neural network based on a loss function, and obtaining an electric load prediction model. Compared with the prior art, the present invention has the advantages of improving the interpretability and accuracy of load prediction results.

Description

Technical Field

[0001] The present invention relates to the technical field of power load forecasting, and in particular to a method, device and medium for forecasting power load based on causal characteristics. Background Art

[0002] In the process of predicting power load, prediction can be made by analyzing historical data. Historical data is generally a time series. A time series refers to a set of data arranged in order according to the time sequence of an observed variable in the real world. The current value of the load sequence can be predicted by the current and past values ​​of multiple sequences associated with the power load or the past values ​​of the load sequence. The power load is generally related to multiple sequences. The data collection of power load influencing factors is usually carried out from multiple different feature dimensions. The resulting time series data is mostly presented in the form of multivariate features. There are complex correlations between multiple variables, and the mutual influence is unclear. A large number of redundant and irrelevant variables will appear, which will have a negative impact on the establishment of the power load forecasting model. In order to clarify the mutual influence between the load sequence and the load factor sequence, it is necessary to analyze the causal relationship. The causal relationship refers to the change of one event or variable directly leading to the change of another event or variable. In the causal relationship, one event is considered to be the cause of another event, and there is a causal relationship. The current power load forecasting method cannot analyze a clear causal relationship, is easily interfered by redundant variables, and the prediction results are poorly interpretable and the prediction accuracy is low. Summary of the invention

[0003] The purpose of the present invention is to provide a method, device and medium for power load forecasting based on causal characteristics in order to reduce the interference of redundant variables during power load analysis and improve the interpretability and accuracy of load forecasting results.

[0004] The purpose of the present invention can be achieved by the following technical solutions:

[0005] A method for predicting power load based on causal characteristics, the method comprising:

[0006] The acquired actual power load influencing factor data is input into the trained power load forecasting model to output the actual power load forecasting result;

[0007] The training process of the power load prediction model is as follows:

[0008] S1. Acquire historical power load data, wherein the historical power load data includes historical power load influencing factor data and corresponding historical output power load data, wherein the historical power load influencing factor data is composed of N power load influencing factor time series, and each influencing factor time series corresponds to one power load influencing factor;

[0009] S2. Preprocessing the historical power load data;

[0010] S3, calculating the normalized causal entropy between each power load influencing factor time series in the preprocessed historical power load data obtained in S2 and the historical output power load data, taking the power load influencing factor time series with normalized causal entropy greater than 0 as the cause sequence, and taking the normalized causal entropy corresponding to each cause sequence as the causal strength of the cause sequence;

[0011] S4. Construct a causal analysis neural network, which includes a deep separable convolution and a prediction module. In the causal analysis neural network, the cause sequence and the corresponding causal strength are multiplied to obtain a product, and a prediction value is output according to the product. A loss function is calculated based on the prediction value and historical output power load data, and the causal analysis neural network is optimized based on the loss function to obtain a power load prediction model.

[0012] Furthermore, the specific steps of S1 and S2 are:

[0013] Obtain historical power load data, preprocess the historical power load data, and then record the preprocessed historical output power load data as ,in, Indicates t Historical power load values;

[0014] The pre-processed historical power load influencing factor data is recorded as ,in Indicates i The time series of power load influencing factors corresponding to the power load influencing factors are: N represents the total number of time series of factors affecting power load, R represents the set of real numbers, T Indicates the maximum time series number of each power load influencing factor time series and the maximum time series number of historical output power load data;

[0015] remember ,in Indicates i The first t Data on factors affecting power load.

[0016] Furthermore, the normalized causal entropy between the power load influencing factor time series and the historical output power load data is:

[0017] ;

[0018] in, Represents the time series of factors affecting power load With historical output power load data Y The normalized causal entropy between K represents a set of conditions, Indicates the historical output power load data in t The value of the moment, Represents the time series of factors affecting power load exist t Before l The value within the delay, Represents a condition set K exist t Before l The value within the delay, I represents the conditional mutual information, h Indicates historical output power load data Y Self-information;

[0019] condition set K The calculation method is:

[0020] (1) Setup K is an empty set;

[0021] (2) Treat the time series of power load influencing factors corresponding to each power load influencing factor as a node, traverse all nodes, and if there is a node j , so that , then add j For collection K After the traversal is completed, the final condition set is obtained K .

[0022] Furthermore, historical output power load data Y The self-information is:

[0023] ;

[0024] in, Indicates when Y The value of The probability of an event occurring is:

[0025] ;

[0026] The Gaussian kernel function It is used to characterize the actual distribution of each sample point, specifically:

[0027] ;

[0028] in, θ represents the distribution parameter.

[0029] Furthermore, the conditional mutual information is:

[0030] ;

[0031] in,

[0032] ;

[0033] ;

[0034] ;

[0035] ;

[0036] in, Indicates i The data of power load influencing factors in the power load influencing factors time series are Before The value within the delay, Represents a condition set exist Before The value within the delay.

[0037] Furthermore, the depthwise separable convolution includes a depthwise convolution, a time-domain convolutional network, a first batch of normalization layers, a first activation layer, a point-by-point convolutional layer, a second batch of normalization layers, and a second activation layer;

[0038] The specific steps of multiplying the cause sequence and the corresponding cause-effect strength in the causal analysis neural network to obtain the product and outputting the prediction value according to the product are as follows:

[0039] The cause sequence and the corresponding causal strength are input into the deep convolution. The deep convolution treats each input cause sequence as a separate channel. In each channel, the cause sequence is multiplied by the causal strength corresponding to the cause sequence. The deep convolution outputs the product corresponding to each channel. Each product passes through the time domain convolution network, the first batch of normalization layers and the first activation layer of the corresponding channel in sequence. The first feature output by the first activation layer of all channels is input into the point-by-point convolution layer. The point-by-point convolution layer outputs the combined feature. The combined feature passes through the second batch of normalization layers and the second activation layer to output the second feature. The second feature is input into the prediction module, and the prediction module outputs the predicted value.

[0040] Furthermore, the first features output by the first activation layers of all channels are input into the point-by-point convolution layer, and the specific steps of the point-by-point convolution layer outputting the combined features are:

[0041] The first feature output by the first activation layer of all channels is input into the point-by-point convolution layer. The point-by-point convolution layer uses a 1x1 convolution kernel to linearly combine the first feature, and the point-by-point convolution layer outputs the combined feature.

[0042] Furthermore, the preprocessing is specifically as follows:

[0043] The historical power load data is standardized by standard deviation to obtain the preprocessed historical power load data.

[0044] Another aspect of the present invention provides a power load forecasting device based on causal characteristics, comprising a memory, a processor, and a program stored in the memory, wherein the processor implements the above method when executing the program.

[0045] Another aspect of the present invention provides a computer-readable storage medium having a program stored thereon, wherein the program implements the above method when executed.

[0046] Compared with the prior art, the present invention has the following beneficial effects:

[0047] The present invention uses normalized causal entropy to mine the causal relationship and corresponding causal strength between variables, which helps to reduce redundant information and noise in the data, thereby improving the ability to identify truly meaningful features of the data. By understanding the causal relationship between variables, it is possible to better grasp the dynamics behind the data, make more accurate predictions, and reduce the interference of redundant variables during power load analysis. At the same time, the present invention integrates causal relationships and causal strength into the deep network model, which not only plays a role in improving the prediction accuracy of the model, but also gives the prediction results of the model a stronger interpretability. The information of causal relationships and causal strength can intuitively explain how the model predicts based on the causal associations in the data. The present invention also has the advantages of a small number of parameters and low computational complexity. The deep separable convolutional model usually has a lightweight structure, which enables the model to work more efficiently in prediction tasks. BRIEF DESCRIPTION OF THE DRAWINGS

[0048] Figure 1 is a flow chart of the present invention;

[0049] Figure 2 A training flow chart of the power load forecasting model of the present invention;

[0050] Figure 3 This is an example diagram of normalized causal entropy between sequences of the present invention;

[0051] Figure 4A network architecture diagram of the power load prediction model of the present invention;

[0052] Figure 5 This is a structural diagram of the prediction device of the present invention. DETAILED DESCRIPTION

[0053] The present invention is described in detail below in conjunction with the accompanying drawings and specific embodiments. This embodiment is implemented based on the technical solution of the present invention, and provides a detailed implementation method and specific operation process, but the protection scope of the present invention is not limited to the following embodiments.

[0054] Embodiment 1:

[0055] The present invention proposes a method for predicting power load based on causal characteristics. The flowchart of the method is as follows: Figure 1 The training flow chart of the power load forecasting model is shown in Figure 2 The method comprises the following steps:

[0056] The acquired actual power load influencing factor data is input into the trained power load forecasting model to output the actual power load forecasting result;

[0057] The training process of the power load prediction model is as follows:

[0058] S1. Acquire historical power load data, wherein the historical power load data includes historical power load influencing factor data and corresponding historical output power load data, wherein the historical power load influencing factor data is composed of N power load influencing factor time series, and each influencing factor time series corresponds to one power load influencing factor;

[0059] S2. Preprocessing the historical power load data;

[0060] S3, calculating the normalized causal entropy between each power load influencing factor time series in the preprocessed historical power load data obtained in S2 and the historical output power load data, taking the power load influencing factor time series with normalized causal entropy greater than 0 as the cause sequence, and taking the normalized causal entropy corresponding to each cause sequence as the causal strength of the cause sequence;

[0061] S4. Construct a causal analysis neural network, which includes a deep separable convolution and prediction module. In the causal analysis neural network, the cause sequence and the corresponding causal strength are multiplied to obtain a product, and a predicted value is output according to the product. A loss function is calculated based on the predicted value and the historical output power load data, and the causal analysis neural network is optimized based on the loss function to obtain a power load prediction model. The causal analysis neural network of the present invention is a deep separable temporal convolutional network (CF-DSTCN).

[0062] First, in S1, the historical power load influencing factor data belongs to an exogenous series, defined as ,in N is the number of exogenous sequences, T is the length of the sequence; Represents a time series matrix X Middle i The expression vector of the sequence, that is, i Time series of factors affecting power load.

[0063] Define the power load data as the target time series and record the historical output power load data as . is the true value vector of load history data.

[0064] The goal of power load forecasting is to When , predict the next value of the target sequence ,in Represents a nonlinear mapping function.

[0065] In the present invention, the historical power load data includes historical power load influencing factor data and corresponding historical output power load data, and the historical power load data can be recorded through the data of the photovoltaic power station operation history. The historical power load influencing factor data includes data such as temperature, humidity, precipitation, wind speed, light intensity and weather type, that is, the historical power load influencing factor data is composed of multiple power load influencing factor time series, and the power load influencing factor time series includes a temperature time series, a humidity time series, a precipitation time series, a wind speed time series, a light intensity time series and a weather type time series.

[0066] In S2, the preprocessing is specifically to normalize the standard deviation of the historical power load data.

[0067] Since data samples may be obtained from multiple different sources, and data from different sources use different dimensions and magnitudes, in order to avoid the impact of data scale on model training, the experimental data needs to be normalized. The standard deviation normalization (Z-score normalization) method is used to normalize the input data. The calculation formula is:

[0068]

[0069] in, For the corresponding sequence i The mean of all sample data, For sequence i The standard deviation of all sample data, For The results are normalized by standard deviation.

[0070] In S3, the causal relationship between the exogenous time series and the target time series is calculated.

[0071] In time series data, there may be superficial associations between variables, but not all of them are true causal relationships. Considering causal relationships can avoid being influenced by misleading associations, thereby improving the reliability of predictions. Information theory methods have been proven to be effective in inferring causal relationships. In this invention, normalized causal entropy is introduced to identify causal relationships between time series. This method can detect and filter out false or indirect causal relationships. j Pair Sequence i The normalized causal entropy of is defined as follows:

[0072]

[0073] Among them, the node i The parent of the conditional set K , which can be excluded by i Indirect influence transmitted from parents. Representation Node i exist t The value of the moment, , Respectively represent nodes j and condition set K exist t Before l The value within the delay. represents the conditional mutual information, Represents a child node i The self-information of solves the dimensional difference problem caused by different node self-information. In the above normalized causal entropy formula, if the condition set is removed, the formula is:

[0074]

[0075] condition set K The calculation method is:

[0076] (1) Setup K is an empty set;

[0077] (2) Treat the time series of power load influencing factors corresponding to each power load influencing factor as a node, traverse all nodes, and if there is a node j , so that , then add j For collection K After the traversal is completed, the final condition set is obtained K .

[0078] For causal entropy and Adoption and sequence j Pair Sequence i The normalized causal entropy is calculated using the same formula, except that the conditional set is removed.

[0079] Combined with power load data, the series Y Output power load data for history, series i is any power load influencing factor sequence in the historical power load influencing factor data, so the normalized causal entropy between the historical output power load data and the power load influencing factor sequence is:

[0080]

[0081] The causal strength between nodes is measured by normalized causal entropy, which takes a value between 0 and 1, and the cause sequence that has a causal relationship with the target time series is obtained. and the corresponding causal strength , n Represents the number of cause sequences that have a causal relationship with the target time series.

[0082] An example of normalized causal entropy between sequences is shown below: Figure 3 As shown, Y Fathers K For the conditional set and Y Normalized causal entropy If it is equal to 0, it means and Y There is no causal relationship between them.

[0083] The conditional mutual information is:

[0084]

[0085] in,

[0086]

[0087]

[0088]

[0089]

[0090] in, Indicates i The data of power load influencing factors in the power load influencing factors time series are Before The value within the delay, Represents a condition set exist Before The value within the delay.

[0091] Historical output power load data Y The self-information is:

[0092]

[0093] in, Indicates when Y The value of The probability of an event occurring at a certain time;

[0094] Given a time series Y The discrete sequence of ( ), the distribution probability based on the kernel function is estimated as follows:

[0095]

[0096] The Gaussian kernel function It is used to characterize the actual distribution of each sample point, specifically:

[0097]

[0098] in, θ represents the distribution parameter.

[0099] In S4, the network architecture of the power load forecasting model is shown in Figure 4 As shown. Construct a causal analysis neural network. The causal analysis neural network includes a depth-wise separable convolution and a prediction module. First, the depth-wise separable convolution includes a depth-wise convolution, a time-domain convolution network, a first batch of normalization layers, a first activation layer, a point-wise convolution layer, a second batch of normalization layers, and a second activation layer;

[0100] The specific steps of multiplying the cause sequence and the corresponding cause-effect strength in the causal analysis neural network to obtain the product and outputting the prediction value according to the product are as follows:

[0101] The cause sequence and the corresponding causal strength are input into the deep convolution. The deep convolution treats each input cause sequence as a separate channel. In each channel, the cause sequence is multiplied by the causal strength corresponding to the cause sequence. The deep convolution outputs the product corresponding to each channel. Each product passes through the time domain convolution network, the first batch of normalization layers and the first activation layer of the corresponding channel in sequence. The first feature output by the first activation layer of all channels is input into the point-by-point convolution layer. The point-by-point convolution layer outputs the combined feature. The combined feature passes through the second batch of normalization layers and the second activation layer to output the second feature. The second feature is input into the prediction module, and the prediction module outputs the predicted value.

[0102] The above detailed deep convolution treats each cause sequence of the input as a separate channel, multiplies each channel by a weight coefficient, and then applies the convolution kernel to each channel separately. Separate channels allow us to correctly interpret the relationship between the cause time series and the target time series without mixing the inputs. The deep convolution layer uses a stacked time convolution network TCN. TCN can capture longer-range temporal dependencies by using multiple layers of convolution kernels and dilated convolutions, which helps to better understand and predict complex patterns in sequences. In point-by-point convolution, a 1x1 convolution kernel is used to linearly combine the outputs of the deep convolution. This is equivalent to combining the outputs of each channel.

[0103] The prediction module is a multi-layer perceptron (MLP). The second feature is input into the multi-layer perceptron (MLP) to predict the target time series. Y , the prediction results are as follows:

[0104]

[0105] Then, the loss function of mean square error (MSE) is calculated based on the predicted value and historical output power load data to measure the predicted value. With historical output power load data , the difference between them, the expression of mean square error is:

[0106]

[0107] in, is the predicted value, is the true value. The smaller the mean square error, the better the performance of the model.

[0108] Figure 4 In the above example, the causal strength vector is expressed as , which is element-wise identical to n The input time series (historical power load influencing factor time series) are multiplied, such as Figure 4 The top of the express. The higher the value of means the stronger the causal relationship between the historical power load influencing factor time series and the historical output power load data. Causal strength can help the model select important features and relationships more targeted. The model can decide whether to include the information of a certain variable in the prediction process based on the size of the causal strength, thereby improving the efficiency and accuracy of the model in dealing with complex variable relationships. In the time convolution network TCN, the expansion factor is selected , convolution kernel size , the number of layers of stacked TCN The output features of the depthwise separable convolution are fed into an MLP model to predict Y .

[0109] Causal prediction emphasizes identifying causal relationships from past data and then making predictions based on these relationships without using information about future events. This paper uses the temporal convolutional network (TCN) as the backbone network of the deep convolution module and stacks it L Get L layers. TCN is designed with key concepts such as causal convolution, dilated convolution, and residual connection in mind to process long sequence data and achieve accurate time series modeling and prediction. In TCN, the convolution layer uses causal dilated convolution, in which the size of the convolution kernel and the dilation factor are selected. The dilation factor determines the number of time steps that the convolution kernel skips on the input sequence. Dilated convolution makes the size of the effective window grow exponentially with the number of layers. In this way, the convolutional network can use fewer layers to obtain a large receptive field. For a convolution kernel size of and the expansion factor is The convolution kernel needs to be padded at the beginning of the sequence Zeros are added to ensure that the convolution operation does not leak future information. In order to control the number of parameters, a fixed convolution kernel size is usually chosen. k , and increase the expansion factor exponentially d (Right now ),in j Represents the level of the network. TCN uses residual connections to solve the gradient vanishing problem in deep networks and accelerate the training process. F Adding it to the output processed by the backbone TCN network makes it easier for the network to learn residual features.

[0110] The set that will have a causal relationship with the target time series , causal strength vector Input to the CF-DSTCN model to obtain the target time series Y The prediction results For example, suppose the factor that has a causal relationship with the power load is temperature , light intensity , Weather type , and the corresponding causal strengths are , , , then the predicted value . Use mean square error MSE as an evaluation indicator to measure the predicted value With the true value The gap between them is then calculated, and the model parameters are optimized through the adaptive motion estimation algorithm (Adam). The iteration is continued until the preset number of iterations is reached to achieve better model performance.

[0111] After the training is completed, the actual power load influencing factor data is obtained, and the actual power load influencing factor data is input into the trained power load prediction model to output the actual power load prediction result.

[0112] Embodiment 2:

[0113] The present invention also provides a corresponding Figure 1 A schematic structural diagram of a power load prediction device based on causal characteristics. The structural diagram of the prediction device of the present invention is as follows Figure 5 As shown. At the hardware level, the power load forecasting device based on causal characteristics may include multiple processors, internal buses, network interfaces, memory, and a non-volatile memory, and of course may also include hardware required for other services. The processor reads the corresponding computer program from the non-volatile memory into the memory and then runs it to achieve the above Figure 1 Of course, in addition to the software implementation, the present invention does not exclude other implementations, such as logic devices or a combination of software and hardware, etc., that is, the execution subject of the following processing flow is not limited to each logic unit, but can also be hardware or logic devices.

[0114] For the improvement of a technology, it can be clearly distinguished whether it is a hardware improvement (for example, improvement of the circuit structure of diodes, transistors, switches, etc.) or a software improvement (improvement of the method flow). However, with the development of technology, many improvements of the method flow today can be regarded as direct improvements of the hardware circuit structure. Designers almost always obtain the corresponding hardware circuit structure by programming the improved method flow into the hardware circuit. Therefore, it cannot be said that the improvement of a method flow cannot be implemented with a hardware entity module. For example, a programmable logic device (PLD) (such as a field programmable gate array (FPGA)) is such an integrated circuit whose logical function is determined by the user's programming of the device. Designers can "integrate" a digital system on a PLD by programming themselves, without having to ask chip manufacturers to design and make dedicated integrated circuit chips. Moreover, nowadays, instead of manually making integrated circuit chips, this kind of programming is mostly implemented by "logic compiler" software, which is similar to the software compiler used when developing and writing programs, and the original code before compilation must also be written in a specific programming language, which is called hardware description language (HDL). There is not only one kind of HDL, but many kinds, such as ABEL (Advanced Boolean Expression Language), AHDL (Altera Hardware Description Language), Confluence, CUPL (Cornell University Programming Language), HDCal, JHDL (Java Hardware Description Language), Lava, Lola, MyHDL, PALASM, RHDL (Ruby Hardware Description Language), etc. The most commonly used ones are VHDL (Very-High-Speed ​​Integrated Circuit Hardware Description Language) and Verilog. Those skilled in the art should also know that it is only necessary to program the method flow slightly in the above-mentioned hardware description languages ​​and program it into the integrated circuit, and then it is easy to obtain the hardware circuit that implements the logic method flow.

[0115] The controller may be implemented in any suitable manner, for example, the controller may take the form of a microprocessor or processor and a computer-readable medium storing a computer-readable program code (e.g., software or firmware) executable by the (micro)processor, a logic gate, a switch, an application-specific integrated circuit (ASIC), a programmable logic controller, and an embedded microcontroller, examples of which include but are not limited to the following microcontrollers: ARC 625D, Atmel AT91SAM, Microchip PIC18F26K20, and Silicone Labs C8051F320, and the memory controller may also be implemented as part of the control logic of the memory. It is also known to those skilled in the art that, in addition to implementing the controller in a purely computer-readable program code manner, the controller may be implemented in the form of a logic gate, a switch, an application-specific integrated circuit, a programmable logic controller, and an embedded microcontroller by logically programming the method steps. Therefore, such a controller may be considered as a hardware component, and the devices for implementing various functions included therein may also be considered as structures within the hardware component. Or even, the devices for implementing various functions may be considered as both software modules for implementing the method and structures within the hardware component.

[0116] The systems, devices, modules or units described in the above embodiments may be implemented by computer chips or entities, or by products with certain functions. A typical implementation device is a computer. Specifically, the computer may be, for example, a personal computer, a laptop computer, a cellular phone, a camera phone, a smart phone, a personal digital assistant, a media player, a navigation device, an email device, a game console, a tablet computer, a wearable device, or a combination of any of these devices.

[0117] For the convenience of description, the above device is described as being divided into various units according to their functions. Of course, when implementing the present invention, the functions of each unit can be implemented in the same or multiple software and / or hardware.

[0118] It will be appreciated by those skilled in the art 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. Furthermore, 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.

[0119] 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.

[0120] 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 including 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.

[0121] 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.

[0122] In a typical configuration, a computing device includes one or more processors (CPU), input / output interfaces, network interfaces, and memory.

[0123] The memory may include non-permanent storage in a computer-readable medium, random access memory (RAM) and / or non-volatile memory in the form of read-only memory (ROM) or flash RAM. The memory is an example of a computer-readable medium.

[0124] Embodiment 3:

[0125] The present invention also provides a computer-readable storage medium on which a program is stored. When the program is executed, the method in the embodiment is implemented.

[0126] Computer readable media include permanent and non-permanent, removable and non-removable media that can be implemented by any method or technology to store information. Information can be computer readable instructions, data structures, program modules or other data. Examples of computer storage media include, but are not limited to, phase change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technology, compact disk read-only memory (CD-ROM), digital versatile disk (DVD) or other optical storage, magnetic cassettes, magnetic tape disk storage or other magnetic storage devices or any other non-transmission media that can be used to store information that can be accessed by a computing device. As defined herein, computer readable media does not include temporary computer readable media (transitory media), such as modulated data signals and carrier waves.

[0127] It should also be noted that the terms "include", "comprises" or any other variations thereof are intended to cover non-exclusive inclusion, so that a process, method, commodity or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, commodity or device. In the absence of more restrictions, the elements defined by the sentence "comprises a ..." do not exclude the existence of other identical elements in the process, method, commodity or device including the elements.

[0128] It should be understood by those skilled in the art that the 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 aspects. 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.

[0129] The present invention may be described in the general context of computer-executable instructions executed by a computer, such as program modules. Generally, program modules include routines, programs, objects, components, data structures, etc. that perform specific tasks or implement specific abstract data types. The present invention may also be practiced in distributed computing environments where tasks are performed by remote processing devices connected through a communication network. In a distributed computing environment, program modules may be located in local and remote computer storage media, including storage devices.

[0130] Each embodiment of the present invention is described in a progressive manner, and the same or similar parts between the embodiments can be referred to each other, and each embodiment focuses on the differences from other embodiments. In particular, for the system embodiment, since it is basically similar to the method embodiment, the description is relatively simple, and the relevant parts can be referred to the partial description of the method embodiment.

[0131] The above description is only an embodiment of the present invention and is not intended to limit the present invention. For those skilled in the art, the present invention may have various modifications and variations. Any modification, equivalent substitution, improvement, etc. made within the spirit and principle of the present invention should be included in the scope of the claims of the present invention.

[0132] The preferred specific embodiments of the present invention are described in detail above. It should be understood that a person skilled in the art can make many modifications and changes based on the concept of the present invention without creative work. Therefore, any technical solution that can be obtained by a person skilled in the art through logical analysis, reasoning or limited experiments based on the concept of the present invention on the basis of the prior art should be within the scope of protection determined by the claims.

[0133] The preferred specific embodiments of the present invention have been described. It should be understood that a person skilled in the art can make many modifications and changes based on the concept of the present invention without creative work. Therefore, any technical solution that can be obtained by a person skilled in the art through logical analysis, reasoning or limited experiments based on the concept of the present invention on the basis of the prior art should be within the scope of protection determined by the claims.

Claims

1. A method for predicting power load based on causal characteristics, characterized in that: Methods include: The acquired actual power load influencing factor data is input into the trained power load forecasting model to output the actual power load forecasting result; The training process of the power load prediction model is as follows: S1. Acquire historical power load data, wherein the historical power load data includes historical power load influencing factor data and corresponding historical output power load data, wherein the historical power load influencing factor data is composed of N power load influencing factor time series, and each influencing factor time series corresponds to one power load influencing factor; S2. Preprocessing the historical power load data; S3, calculating the normalized causal entropy between each power load influencing factor time series in the preprocessed historical power load data obtained in S2 and the historical output power load data, taking the power load influencing factor time series with normalized causal entropy greater than 0 as the cause sequence, and taking the normalized causal entropy corresponding to each cause sequence as the causal strength of the cause sequence; S4. Construct a causal analysis neural network, which includes a deep separable convolution and a prediction module. In the causal analysis neural network, the cause sequence and the corresponding causal strength are multiplied to obtain a product, and a prediction value is output according to the product. A loss function is calculated based on the prediction value and historical output power load data, and the causal analysis neural network is optimized based on the loss function to obtain a power load prediction model.

2. The method for predicting power load based on causal characteristics according to claim 1, characterized in that: The specific steps of S1 and S2 are: Obtain historical power load data, preprocess the historical power load data, and then record the preprocessed historical output power load data as ,in, Indicates t Historical power load values; The pre-processed historical power load influencing factor data is recorded as ,in Indicates i The time series of power load influencing factors corresponding to the power load influencing factors are: N represents the total number of time series of factors affecting power load, R represents the set of real numbers, T Indicates the maximum time series number of each power load influencing factor time series and the maximum time series number of historical output power load data; remember ,in Indicates i The first t Data on factors affecting power load.

3. The method for predicting power load based on causal characteristics according to claim 2 is characterized in that: The normalized causal entropy between the power load influencing factor time series and the historical output power load data is: ； in, Represents the time series of factors affecting power load With historical output power load data Y The normalized causal entropy between K represents a set of conditions, Indicates historical output power load data in t The value of the moment, Represents the time series of factors affecting power load exist t Before l The value within the delay, Represents a condition set K exist t Before l The value within the delay, I represents the conditional mutual information, h Indicates historical output power load data Y Self-information; condition set K The calculation method is: (1) Settings K is an empty set; (2) Treat the time series of power load influencing factors corresponding to each power load influencing factor as a node, traverse all nodes, and if there is a node j , so that , then add j For collection K After the traversal is completed, the final condition set is obtained K .

4. The method for predicting power load based on causal characteristics according to claim 3 is characterized in that: Historical output power load data Y The self-information is: ； in, Indicates when Y The value of The probability of an event occurring is: ； The Gaussian kernel function It is used to characterize the actual distribution of each sample point, specifically: ； in, θ represents the distribution parameter.

5. The method for predicting power load based on causal characteristics according to claim 4 is characterized in that: The conditional mutual information is: ； in, ； ； ； ； in, Indicates i The data of power load influencing factors in the power load influencing factors time series are Before The value within the delay, Represents a condition set exist Before The value within the delay.

6. The method for predicting power load based on causal characteristics according to claim 1, characterized in that: The depthwise separable convolution includes a depthwise convolution, a time-domain convolutional network, a first batch of normalization layers, a first activation layer, a point-by-point convolutional layer, a second batch of normalization layers, and a second activation layer; The specific steps of multiplying the cause sequence and the corresponding cause-effect strength in the causal analysis neural network to obtain the product and outputting the prediction value according to the product are as follows: The cause sequence and the corresponding causal strength are input into the deep convolution. The deep convolution treats each input cause sequence as a separate channel. In each channel, the cause sequence is multiplied by the causal strength corresponding to the cause sequence. The deep convolution outputs the product corresponding to each channel. Each product passes through the time domain convolution network, the first batch of normalization layers and the first activation layer of the corresponding channel in sequence. The first feature output by the first activation layer of all channels is input into the point-by-point convolution layer. The point-by-point convolution layer outputs the combined feature. The combined feature passes through the second batch of normalization layers and the second activation layer to output the second feature. The second feature is input into the prediction module, and the prediction module outputs the predicted value.

7. The method for predicting power load based on causal characteristics according to claim 6, characterized in that: The first feature output by the first activation layer of all channels is input into the point-by-point convolution layer, and the specific steps of the point-by-point convolution layer outputting the combined feature are: The first feature output by the first activation layer of all channels is input into the point-by-point convolution layer. The point-by-point convolution layer uses a 1x1 convolution kernel to linearly combine the first feature, and the point-by-point convolution layer outputs the combined feature.

8. The method for predicting power load based on causal characteristics according to claim 6, characterized in that: The pre-processing is specifically as follows: The historical power load data is standardized by standard deviation to obtain the preprocessed historical power load data.

9. A power load forecasting device based on causal characteristics, comprising a memory, a processor, and a program stored in the memory, characterized in that: When the processor executes the program, the method according to any one of claims 1 to 8 is implemented.

10. A computer-readable storage medium having a program stored thereon, characterized in that: When the program is executed, the method according to any one of claims 1 to 8 is implemented.

Images