Ultra-short-term prediction method and device for net load of power distribution network
By using a decoupled load generation and consumption model based on a multilayer perceptron and a long short-term memory network model, the problem of ultra-short-term net load forecasting in distribution networks was solved, achieving high-precision load forecasting and improving the operational stability of distribution networks and the absorption capacity of distributed energy.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- CHINA ELECTRIC POWER RESEARCH INSTITUTE CO LTD
- Filing Date
- 2025-12-02
- Publication Date
- 2026-04-17
AI Technical Summary
The difficulty of ultra-short-term forecasting of net load in distribution networks has increased, and existing methods have high requirements for data integrity and accuracy and are susceptible to the effects of error accumulation.
A decoupled net load generation and consumption model based on a multilayer perceptron is adopted. Historical net load data and meteorological data are used for prediction. Combined with a long short-term memory network model, the electricity load and photovoltaic output data are reconstructed to achieve high-precision prediction.
It improves the accuracy of ultra-short-term net load forecasting for distribution networks, provides a data foundation and technical support, and enhances the absorption of distributed energy resources and the safe and stable operation of distribution networks.
Smart Images
Figure CN121886338A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of load forecasting technology, and specifically to a method and apparatus for ultra-short-term net load forecasting of a distribution network. Background Technology With the large-scale integration of distributed power sources, users are gradually acquiring dual attributes of power generation and consumption, leading to a significant trend of integrated generation and consumption in distribution networks. This results in greater randomness and uncertainty in the net load of distribution networks, significantly increasing the difficulty of load forecasting. Based on different forecast time scales, net load forecasting can be categorized into ultra-short-term, short-term, medium-term, and long-term forecasts. Ultra-short-term forecasting typically refers to predicting net load for the next few minutes to hours and is widely used in important scenarios such as real-time dispatching of distribution networks, power quality management, flexible peak shaving, and energy storage optimization. Improving the accuracy of ultra-short-term net load forecasting not only helps enhance the absorption of distributed energy resources but also provides a solid guarantee for the safe, stable, and economical operation of distribution networks.
[0002] Currently, net load forecasting methods are divided into indirect and direct forecasting. Direct forecasting uses net load as the forecast target, directly building a forecasting model and combining it with weather, historical load, and renewable energy output to obtain the predicted value of future net load. Direct forecasting is more suitable for scenarios with sufficient data and complete relevant characteristics, but it has high requirements for the completeness and accuracy of historical net load data and characteristic variables. In contrast, indirect forecasting methods have a clear structure and are easy to utilize existing load and distributed power generation output forecasting models, but their results are easily affected by the cumulative effect of forecasting errors from both sources. Summary of the Invention
[0003] To overcome the above-mentioned shortcomings, this invention proposes a method and apparatus for ultra-short-term forecasting of net load in distribution networks.
[0004] Firstly, a method for ultra-short-term forecasting of net load in a distribution network is provided, the method comprising: Historical net load data and meteorological data are used as inputs to a pre-built net load generation and consumption decoupling model based on a multilayer sensor, and the historical electricity load data and photovoltaic power output data output by the pre-built net load generation and consumption decoupling model based on a multilayer sensor are obtained. The historical electricity load data and photovoltaic output data are used as inputs to the pre-built electricity load data prediction model and photovoltaic output data prediction model, respectively, to obtain the electricity load prediction data output by the pre-built electricity load data prediction model and the photovoltaic output prediction data output by the photovoltaic output data prediction model. The electricity load forecast data and photovoltaic output forecast data are reconstructed to obtain the net load forecast data of the distribution network.
[0005] Preferably, the meteorological data includes at least one of the following: temperature data and radiation data.
[0006] Preferably, the input layer of the pre-built net load generation decoupling model based on multilayer perceptron encodes the input features through a linear layer and a ReLU nonlinear unit to generate high-dimensional temporal features. The output of the ReLU nonlinear unit is max(0,Wx+b), where x is the input feature, W is the weight matrix, and b is the bias term.
[0007] Preferably, during the training process of the pre-constructed net load generation decoupling model based on a multilayer perceptron, the guiding signal of the guiding layer is the photovoltaic theoretical output.
[0008] Preferably, during the training of the pre-built payload decoupling model based on a multilayer perceptron, the loss function is as follows:
[0009] In the above formula, The values represent the loss function, where A, B, C, and D are the first, second, third, and fourth weighting coefficients, respectively. rec To reconstruct the loss function, loss sup For weak supervision correlation functions of photovoltaic power output data, stat loss Statistical constraint loss function for electricity load data, saddle loss This is the consistency loss function.
[0010] Furthermore, the reconstruction loss function is as follows:
[0011] The weak supervision correlation function for the photovoltaic output data is as follows:
[0012] The statistical constraint loss function for the electricity load data is as follows:
[0013] The consistency loss function is as follows:
[0014] In the above formula, N represents the amount of training data for the model. For the first i Net payload data corresponding to each training data point Separate and reconstruct the data for the model corresponding to the i-th training data. For the first i The theoretical photovoltaic output data corresponding to each training data point. For the first iPhotovoltaic output data corresponding to each training data point The average of the theoretical output data for photovoltaics. This represents the average photovoltaic power output data. p 1. p 2 are the penalty functions for daily load factor and daily minimum load factor, respectively. For the first i The electricity load data corresponding to each training data point This represents the maximum daily load factor. This represents the minimum daily load factor. The maximum value of the minimum load rate. This is the minimum value of the minimum load factor. G i For the first i The irradiation data corresponding to each training data point For the first i The photovoltaic output predicted by the model for each training data point. For indicator functions, when The value is 1 when it equals 0, otherwise it is 0.
[0015] Preferably, both the pre-built electricity load data prediction model and the photovoltaic output data prediction model are long short-term memory network models.
[0016] Furthermore, during the training of the Long Short-Term Memory network model, the loss function is as follows:
[0017] In the above formula, Y i and The first i The predicted data and training data corresponding to each training data point E MRE denoted as the average relative error, and n as the number of training samples.
[0018] Furthermore, during the training of the Long Short-Term Memory network model, the loss function is as follows:
[0019] In the above formula, Y i and The first i The predicted data and training data corresponding to each training data point E The mean relative error is given by n, the number of training samples, and T, which is a preset percentage. This indicates the number of qualified prediction points.
[0020] Secondly, a distribution network net load ultra-short-term forecasting device is provided, the distribution network net load ultra-short-term forecasting device comprising: The first analysis module is used to take historical net load data and meteorological data as input to a pre-built net load generation and consumption decoupling model based on a multilayer sensor, and obtain historical electricity load data and photovoltaic power output data output by the pre-built net load generation and consumption decoupling model based on a multilayer sensor. The second analysis module is used to take the historical electricity load data and photovoltaic output data as inputs to the pre-built electricity load data prediction model and photovoltaic output data prediction model, respectively, to obtain the electricity load prediction data output by the pre-built electricity load data prediction model and the photovoltaic output prediction data output by the photovoltaic output data prediction model. The third analysis module is used to reconstruct the electricity load forecast data and photovoltaic output forecast data to obtain the net load forecast data of the distribution network.
[0021] Thirdly, a computer device is provided, comprising: one or more processors; The processor is used to execute one or more programs; When the one or more programs are executed by the one or more processors, the ultra-short-term net load forecasting method for the distribution network is implemented.
[0022] Fourthly, a computer-readable storage medium is provided having a computer program stored thereon, wherein when the computer program is executed, the method for ultra-short-term net load forecasting of the distribution network is implemented.
[0023] The above-described technical solutions of the present invention have at least one or more of the following beneficial effects: This invention provides a method and apparatus for ultra-short-term net load forecasting of a distribution network, comprising: using historical net load data and meteorological data as inputs to a pre-constructed net load generation-consumption decoupling model based on a multilayer sensor, to obtain historical electricity load data and photovoltaic power output data output by the pre-constructed net load generation-consumption decoupling model based on a multilayer sensor; using the historical electricity load data and photovoltaic power output data as inputs to a pre-constructed electricity load data prediction model and a photovoltaic power output data prediction model, respectively, to obtain electricity load forecast data output by the pre-constructed electricity load data prediction model and photovoltaic power output forecast data output by the pre-constructed photovoltaic power output data prediction model; and reconstructing the electricity load forecast data and photovoltaic power output forecast data to obtain distribution network net load forecast data. This scheme achieves high-precision ultra-short-term forecasting of distribution network net load, specifically: The net load generation and decoupling model of this invention introduces a guiding layer to guide the separation process with weak supervision, ensuring the independence of each component of the separation result and good physical interpretability. An objective function and an input layer are designed, and a deep representation and decoupling of historical net load measurement data are achieved through nonlinear feature mapping, effectively extracting the electricity load signal and distributed photovoltaic power output signal that have both physical meaning and statistical characteristics.
[0024] The ultra-short-term forecasting model for distribution network net load of this invention constructs independent LSTM time-series forecasting models for the separated components of electricity load and distributed photovoltaic power output. It fully explores the dynamic evolution law of load and photovoltaic power output, obtains high-precision ultra-short-term forecasts of electricity load and photovoltaic components, and reconstructs the component forecasting results to achieve high-precision ultra-short-term forecasts of distribution network net load. This provides a data foundation and technical support for subsequent scientific research and business applications such as generating typical scenarios for distribution network simulation and estimating distributed photovoltaic capacity for distributed photovoltaic absorption calculation. Attached Figure Description
[0025] Figure 1 This is a schematic diagram of the main steps of the ultra-short-term net load forecasting method for distribution networks according to an embodiment of the present invention; Figure 2 This is a block diagram of the net load generation decoupling model according to an embodiment of the present invention; Figure 3 This is a schematic diagram of the net load ultra-short-term model according to an embodiment of the present invention. Detailed Implementation
[0026] The specific embodiments of the present invention will be further described in detail below with reference to the accompanying drawings.
[0027] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.
[0028] Example 1 See appendix Figure 1 , Figure 1 This is a schematic flowchart illustrating the main steps of an ultra-short-term net load forecasting method for distribution networks according to an embodiment of the present invention. Figure 1 As shown, the ultra-short-term net load forecasting method for distribution networks in this embodiment of the invention mainly includes the following steps: Step S101: Use historical net load data and meteorological data as input to the pre-built net load generation and consumption decoupling model based on multilayer sensor to obtain the historical electricity load data and photovoltaic output data output by the pre-built net load generation and consumption decoupling model based on multilayer sensor; Step S102: Use the historical electricity load data and photovoltaic output data as inputs to the pre-built electricity load data prediction model and photovoltaic output data prediction model, respectively, to obtain the electricity load prediction data output by the pre-built electricity load data prediction model and the photovoltaic output prediction data output by the photovoltaic output data prediction model. Step S103: Reconstruct the electricity load forecast data and photovoltaic output forecast data to obtain the net load forecast data of the distribution network.
[0029] In this embodiment, the meteorological data includes at least one of the following: temperature data and radiation data.
[0030] In this embodiment, as Figure 2 As shown, the pre-built payload generation and decoupling model based on multilayer perceptron achieves efficient separation of complex measured payloads by integrating main task reconstruction, auxiliary statistical information and physical prior knowledge, and using a small amount of labeled data for guidance. The data processing structure includes an input layer and a guidance layer. The design of the input layer and the guidance layer are described separately. The input layer encodes the input features through a linear layer and a ReLU nonlinear unit to generate high-dimensional time-series features. The output of the ReLU nonlinear unit is max(0,Wx+b), where x is the input feature, W is the weight matrix, and b is the bias term.
[0031] In this embodiment, during the blind source separation model training process, a guiding layer can be introduced to achieve weakly supervised learning of the model under limited labeled data, thereby more accurately fitting the actual data distribution. This invention, based on the pvlib photovoltaic modeling tool, combines geographical location information and equipment parameters to construct a photovoltaic power generation theoretical model, and uses temperature and irradiance data to estimate the theoretical photovoltaic output. Finally, the obtained theoretical photovoltaic output is used as the guiding signal for model training, effectively improving the model's separation performance and physical interpretability. Therefore, during the training process of the pre-constructed net load generation-use decoupling model based on a multilayer perceptron, the guiding signal of the guiding layer is the theoretical photovoltaic output.
[0032] In this embodiment, physical interpretability refers to the ability of the model's internal variables and prediction results to establish a clear and concrete correspondence with real physical processes or quantities, thereby truly reflecting observable and understandable physical phenomena in reality. Physical interpretability is mainly reflected in the following two aspects: (1) There is a significant correlation between photovoltaic output and irradiance and temperature data, and when irradiance is zero, photovoltaic output should be zero; (2) Electricity load should be able to reflect the changing patterns of user electricity consumption behavior, presenting actual characteristics such as intraday fluctuations, seasonality, and holiday effects. In combination with the convergence requirements of the model, the loss function in the training process of the pre-constructed net load generation and decoupling model based on multilayer perceptron is as follows:
[0033] In the above formula, The values represent the loss function, where A, B, C, and D are the first, second, third, and fourth weighting coefficients, respectively. rec To reconstruct the loss function, loss sup For weak supervision correlation functions of photovoltaic power output data, stat loss Statistical constraint loss function for electricity load data, saddle loss This is the consistency loss function.
[0034] In one implementation, the reconstruction loss function is as follows:
[0035] The weak supervision correlation function for the photovoltaic output data is as follows:
[0036] The statistical constraint loss function for the electricity load data is as follows:
[0037] The consistency loss function is as follows:
[0038] In the above formula, N represents the amount of training data for the model. For the first i Net payload data corresponding to each training data point Separate and reconstruct the data for the model corresponding to the i-th training data. For the first i The theoretical photovoltaic output data corresponding to each training data point. For the first i Photovoltaic output data corresponding to each training data point The average of the theoretical output data for photovoltaics. This represents the average photovoltaic power output data. p 1.p 2 are the penalty functions for daily load factor and daily minimum load factor, respectively. For the first i The electricity load data corresponding to each training data point This represents the maximum daily load factor. This represents the minimum daily load factor. The maximum value of the minimum load rate. This is the minimum value of the minimum load factor. G i For the first i The irradiation data corresponding to each training data point For the first i The photovoltaic output predicted by the model for each training data point. For indicator functions, when The value is 1 when it equals 0, otherwise it is 0.
[0039] In this embodiment, both the pre-built electricity load data prediction model and the photovoltaic output data prediction model are long short-term memory network models.
[0040] Specifically, the net load curve is decomposed using a net load generation-consumption decoupling model to obtain photovoltaic and electricity load time series; then, a Long Short-Term Memory (LSTM) network model is applied separately for ultra-short-term prediction. This method can better capture the statistical characteristics and patterns of electricity load and photovoltaic time series data, improving the accuracy of prediction.
[0041] For electricity load forecasting, LSTM can effectively capture historical load variation patterns, thus providing accurate forecasts of electricity demand in the short term. In this study, the model input data consists of historical data from the 24 hours prior to the forecast point and meteorological data from the next 4 hours. The output is the electricity load data for the next 4 hours. Taking a certain domain dataset as an example, with a data point time interval of 15 minutes, the input and output of the forecast model can be represented as follows:
[0042] In the formula X load , Y load These are the input and output values of the model, respectively. L t-k , T t-k (Time step) tk Historical electricity load values and corresponding time values, C t+k Stepping into the future t+k Temperature forecast, L t+k Stepping into the future t+kThe predicted electricity load values.
[0043] The photovoltaic forecasting model can also use the LSTM model for prediction. The model input consists of photovoltaic output data for the 24 hours prior to the prediction point and meteorological data for the next 4 hours. The input and output of the forecasting model can be represented as follows:
[0044] In the formula X pv 、Y pv These are the model's input and output values, respectively. R t+k Stepping into the future t+k The irradiance prediction value is given, and the remaining data are similar to those of the electricity load prediction model. The specific structure of the ultra-short-term prediction model constructed in this invention is as follows: Figure 3 As shown.
[0045] In one embodiment, this invention uses the average relative error (MRE) and prediction accuracy pass rate, which are commonly used in load forecasting, as evaluation indicators to measure the performance of the prediction model. These indicators respectively reflect the average level of the overall prediction error and the effectiveness of the prediction data in meeting the needs of actual power production applications. Therefore, the loss function during the training process of the Long Short-Term Memory network model is as follows:
[0046] In the above formula, Y i and The first i The predicted data and training data corresponding to each training data point E MRE denoted as the average relative error, and n as the number of training samples.
[0047] In one implementation, the loss function during the training of the Long Short-Term Memory network model is as follows:
[0048] In the above formula, Y i and The first i The predicted data and training data corresponding to each training data point E The mean relative error is given by n, the number of training samples, and T, which is a preset percentage. This indicates the number of qualified prediction points.
[0049] Example 2 Based on the same inventive concept, the present invention also provides a distribution network net load ultra-short-term forecasting device, the distribution network net load ultra-short-term forecasting device comprising: The first analysis module is used to take historical net load data and meteorological data as input to a pre-built net load generation and consumption decoupling model based on a multilayer sensor, and obtain historical electricity load data and photovoltaic power output data output by the pre-built net load generation and consumption decoupling model based on a multilayer sensor. The second analysis module is used to take the historical electricity load data and photovoltaic output data as inputs to the pre-built electricity load data prediction model and photovoltaic output data prediction model, respectively, to obtain the electricity load prediction data output by the pre-built electricity load data prediction model and the photovoltaic output prediction data output by the photovoltaic output data prediction model. The third analysis module is used to reconstruct the electricity load forecast data and photovoltaic output forecast data to obtain the net load forecast data of the distribution network.
[0050] In this embodiment, the meteorological data includes at least one of the following: temperature data and radiation data.
[0051] In this embodiment, the input layer of the pre-built net load generation decoupling model based on multilayer perceptron encodes the input features through a linear layer and a ReLU nonlinear unit to generate high-dimensional time-series features. The output of the ReLU nonlinear unit is max(0,Wx+b), where x is the input feature, W is the weight matrix, and b is the bias term.
[0052] In this embodiment, during the training process of the pre-built net load generation decoupling model based on a multilayer perceptron, the guiding signal of the guiding layer is the photovoltaic theoretical output.
[0053] In this embodiment, the loss function during the training of the pre-built payload decoupling model based on a multilayer perceptron is as follows:
[0054] In the above formula, The values represent the loss function, where A, B, C, and D are the first, second, third, and fourth weighting coefficients, respectively. rec To reconstruct the loss function, loss sup For weak supervision correlation functions of photovoltaic power output data, stat loss Statistical constraint loss function for electricity load data, saddle loss This is the consistency loss function.
[0055] Furthermore, the reconstruction loss function is as follows:
[0056] The weak supervision correlation function for the photovoltaic output data is as follows:
[0057] The statistical constraint loss function for the electricity load data is as follows:
[0058] The consistency loss function is as follows:
[0059] In the above formula, N represents the amount of training data for the model. For the first i Net payload data corresponding to each training data point Separate and reconstruct the data for the model corresponding to the i-th training data. For the first i The theoretical photovoltaic output data corresponding to each training data point. For the first i Photovoltaic output data corresponding to each training data point The average of the theoretical output data for photovoltaics. This represents the average photovoltaic power output data. p 1. p 2 are the penalty functions for daily load factor and daily minimum load factor, respectively. For the first i The electricity load data corresponding to each training data point This represents the maximum daily load factor. This represents the minimum daily load factor. The maximum value of the minimum load rate. This is the minimum value of the minimum load factor. G i For the first i The irradiation data corresponding to each training data point For the first i The photovoltaic output predicted by the model for each training data point. For indicator functions, when The value is 1 when it equals 0, otherwise it is 0.
[0060] In this embodiment, both the pre-built electricity load data prediction model and the photovoltaic output data prediction model are long short-term memory network models.
[0061] Furthermore, during the training of the Long Short-Term Memory network model, the loss function is as follows:
[0062] In the above formula, Y i and The first i The predicted data and training data corresponding to each training data point E MREdenoted as the average relative error, and n as the number of training samples.
[0063] Furthermore, during the training of the Long Short-Term Memory network model, the loss function is as follows:
[0064] In the above formula, Y i and The first i The predicted data and training data corresponding to each training data point E The mean relative error is given by n, the number of training samples, and T, which is a preset percentage. This indicates the number of qualified prediction points.
[0065] Example 3 Based on the same inventive concept, this invention also provides a computer device, which includes a processor and a memory. The memory stores a computer program, which includes program instructions. The processor executes the program instructions stored in the computer storage medium. The processor may be a Central Processing Unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. It is the computing and control core of the terminal, and is suitable for implementing one or more instructions. Specifically, it is suitable for loading and executing one or more instructions in the computer storage medium to realize the corresponding method flow or corresponding function, so as to realize the steps of the ultra-short-term net load forecasting method for distribution networks in the above embodiments.
[0066] Example 4 Based on the same inventive concept, this invention also provides a storage medium, specifically a computer-readable storage medium (Memory), which is a memory device in a computer device used to store programs and data. It is understood that the computer-readable storage medium here can include both the built-in storage medium in the computer device and extended storage media supported by the computer device. The computer-readable storage medium provides storage space that stores the terminal's operating system. Furthermore, this storage space also stores one or more instructions suitable for loading and execution by a processor. These instructions can be one or more computer programs (including program code). It should be noted that the computer-readable storage medium here can be high-speed RAM or non-volatile memory, such as at least one disk storage device. The processor can load and execute one or more instructions stored in the computer-readable storage medium to implement the steps of the ultra-short-term net load forecasting method for a distribution network described in the above embodiments.
[0067] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied 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.
[0068] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.
[0069] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.
[0070] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.
[0071] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit it. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the specific implementation of the present invention. Any modifications or equivalent substitutions that do 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 method for ultra-short-term forecasting of net load in a distribution network, characterized in that, The method includes: Historical net load data and meteorological data are used as inputs to a pre-built net load generation and consumption decoupling model based on a multilayer sensor, and the historical electricity load data and photovoltaic power output data output by the pre-built net load generation and consumption decoupling model based on a multilayer sensor are obtained. The historical electricity load data and photovoltaic output data are used as inputs to the pre-built electricity load data prediction model and photovoltaic output data prediction model, respectively, to obtain the electricity load prediction data output by the pre-built electricity load data prediction model and the photovoltaic output prediction data output by the photovoltaic output data prediction model. The electricity load forecast data and photovoltaic output forecast data are reconstructed to obtain the net load forecast data of the distribution network.
2. The method as described in claim 1, characterized in that, The meteorological data includes at least one of the following: temperature data and radiation data.
3. The method as described in claim 1, characterized in that, The input layer of the pre-built net load generation decoupling model based on multilayer perceptron encodes the input features through a linear layer and a ReLU nonlinear unit to generate high-dimensional time-series features. The output of the ReLU nonlinear unit is max(0,Wx+b), where x is the input feature, W is the weight matrix, and b is the bias term.
4. The method as described in claim 1, characterized in that, During the training process of the pre-built net load generation decoupling model based on multilayer perceptron, the guiding signal of the guiding layer is the photovoltaic theoretical output.
5. The method as described in claim 1, characterized in that, During the training of the pre-built payload decoupling model based on a multilayer perceptron, the loss function is as follows: In the above formula, The values represent the loss function, where A, B, C, and D are the first, second, third, and fourth weighting coefficients, respectively. rec To reconstruct the loss function, loss sup For weak supervision correlation functions of photovoltaic power output data, stat loss Statistical constraint loss function for electricity load data, saddle loss This is the consistency loss function.
6. The method as described in claim 5, characterized in that, The reconstruction loss function is as follows: The weak supervision correlation function for the photovoltaic output data is as follows: The statistical constraint loss function for the electricity load data is as follows: The consistency loss function is as follows: In the above formula, N represents the amount of training data for the model. For the first i Net payload data corresponding to each training data point Separate and reconstruct the data for the model corresponding to the i-th training data. For the first i The theoretical photovoltaic output data corresponding to each training data point. For the first i Photovoltaic output data corresponding to each training data point The average of the theoretical output data for photovoltaic power. This represents the average photovoltaic power output data. p 1. p 2 are the penalty functions for daily load factor and daily minimum load factor, respectively. For the first i The electricity load data corresponding to each training data point This represents the maximum daily load factor. This represents the minimum daily load factor. The maximum value of the minimum load rate. This is the minimum value of the minimum load factor. G i For the first i Irradiation data corresponding to each training data point For the first i The photovoltaic output predicted by the model for each training data point. For indicator functions, when The value is 1 when it equals 0, otherwise it is 0.
7. The method as described in claim 1, characterized in that, Both the pre-built electricity load data prediction model and the photovoltaic output data prediction model are long short-term memory network models.
8. The method as described in claim 7, characterized in that, During the training of the Long Short-Term Memory network model, the loss function is as follows: In the above formula, Y i and The first i The predicted data and training data corresponding to each training data point E MRE denoted as the average relative error, and n as the number of training samples.
9. The method as described in claim 7, characterized in that, During the training of the Long Short-Term Memory network model, the loss function is as follows: In the above formula, Y i and The first i The predicted data and training data corresponding to each training data point E The mean relative error is given by n, the number of training samples, and T, which is a preset percentage. This indicates the number of qualified prediction points.
10. An apparatus based on the ultra-short-term net load forecasting method for distribution networks according to any one of claims 1-9, characterized in that, The device includes: The first analysis module is used to take historical net load data and meteorological data as input to a pre-built net load generation and consumption decoupling model based on a multilayer sensor, and obtain historical electricity load data and photovoltaic power output data output by the pre-built net load generation and consumption decoupling model based on a multilayer sensor. The second analysis module is used to take the historical electricity load data and photovoltaic output data as inputs to the pre-built electricity load data prediction model and photovoltaic output data prediction model, respectively, to obtain the electricity load prediction data output by the pre-built electricity load data prediction model and the photovoltaic output prediction data output by the photovoltaic output data prediction model. The third analysis module is used to reconstruct the electricity load forecast data and photovoltaic output forecast data to obtain the net load forecast data of the distribution network.
11. A computer device, characterized in that, include: One or more processors; The processor is used to execute one or more programs; When the one or more programs are executed by the one or more processors, the ultra-short-term net load forecasting method for distribution networks as described in any one of claims 1 to 9 is implemented.
12. A computer-readable storage medium, characterized in that, It contains a computer program, which, when executed, implements the ultra-short-term net load forecasting method for distribution networks as described in any one of claims 1 to 9.