A transformer area load data resolution improvement method and device
By constructing a method for improving the resolution of transformer area load data based on neural differential equations, the problem of low resolution of transformer area load data is solved, higher accuracy load data interpolation is achieved, and the accuracy and interpretability of load forecasting are improved.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- ELECTRIC POWER RESEARCH INSTITUTE OF STATE GRID JIBEI ELECTRIC POWER CO LTD
- Filing Date
- 2026-01-28
- Publication Date
- 2026-06-09
AI Technical Summary
In existing technologies, the resolution of load data in transformer areas is low, resulting in insufficient ability of load forecasting models to capture short-term load changes, which affects the accuracy of forecasting results. Furthermore, existing methods ignore the dynamic evolution mechanism between adjacent sampling times, which limits the improvement of interpolation accuracy.
A method for improving the resolution of transformer area load data based on neural differential equations is adopted. By combining a time-series encoder, a time-series decoder, and a fully connected neural network layer, a neural differential equation is constructed. The model is trained using error backpropagation to improve the data resolution and take into account the continuous dynamic evolution of load time-series characteristics.
It effectively improves the interpolation accuracy of transformer area load data, can more accurately reflect users' electricity consumption patterns, provide richer load forecast information, and improve the interpretability and interpolation accuracy of the model.
Smart Images
Figure CN122173884A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of data processing technology, and specifically to a method and apparatus for improving the resolution of load data in a transformer substation. Background Technology
[0002] Load forecasting is crucial for the stable operation of power distribution systems. Accurate forecasting ensures supply and demand balance, achieves energy conservation and loss reduction in power distribution systems, avoids power shortages and overloads, and guarantees the stable operation of the regional power grid, preventing abnormal voltage and frequency fluctuations and significantly reducing the risk of faults. The temporal resolution of historical load data is a significant factor affecting the accuracy of load forecasting. High-resolution historical load data reflects short-term fluctuations and instantaneous changes in electricity consumption patterns, helping to capture users' actual electricity consumption behavior and providing strong data support for accurate load forecasting. However, currently, most distribution areas collect load data at a low frequency. Low-resolution historical data not only limits the load forecasting model's ability to capture short-term load changes but may also lead to the omission of important information, thus affecting the accuracy of load forecasting results.
[0003] In the field of improving the resolution of load data in distribution areas, existing technologies mainly include interpolation methods, statistical methods, and artificial intelligence methods. Interpolation methods, such as linear interpolation, spline interpolation, and polynomial interpolation, are simple but have poor ability to capture complex nonlinear changes. Statistical methods, such as autoregressive moving average models, seasonal decomposition procedures, and exponentially smoothed state-space models, can identify and model trend, seasonal, and periodic components in the data; however, statistical methods rely on assumptions about data distribution, and the actual distribution of the data will affect the model's performance. With the development of artificial intelligence technology, recurrent neural networks and generative adversarial networks have also been applied to improve the resolution of load data and have achieved good results. However, these methods directly process discrete load time series data, ignoring the dynamic evolution mechanism between sampled values at adjacent sampling times, thus limiting further improvements in interpolation accuracy. Summary of the Invention
[0004] To address the problems in the prior art, embodiments of the present invention provide a method and apparatus for improving the resolution of load data in a distribution area, which can at least partially solve the problems existing in the prior art.
[0005] On one hand, this invention proposes a method for improving the resolution of transformer area load data. This method is based on a transformer area load data resolution improvement model. The model includes a time-series encoder, a time-series decoder, and a fully connected neural network layer connected sequentially. The time-series decoder is constructed based on a neural differential equation, which includes a fully connected neural network for fitting the rate of change of the hidden state vector over time, representing the temporal features of the implicit input data. It also includes a differential equation constructed based on the parameters of the hidden state vector related to the temporal features of the implicit input data, the parameters of the fully connected neural network, the interpolation times, and the rate of change. The method for improving the resolution of transformer area load data includes: The time-series encoder is used to encode the low-resolution time-series data of the transformer area load to obtain a hidden state vector that contains the temporal features of the input data. Based on the time-series decoder, the hidden state vector of the hidden input data temporal features is decoded to obtain the hidden state vector at each interpolation time. Based on the fully connected neural network layer, the hidden state vectors at each interpolation time point are mapped to the interpolation results, and the interpolation results are inserted into the low-resolution time series data of the transformer area load to obtain the high-resolution time series data of the transformer area load.
[0006] The step of decoding the hidden state vector of the implicit input data temporal features based on the temporal decoder includes: The rate of change is obtained by fitting the hidden state vector of the temporal features of the implicit input data to the fully connected neural network. Using the rate of change, the initial hidden state vector corresponding to the initial time of the hidden state vector containing the temporal features of the implicit input data, the initial time, each interpolation time, and the parameters of the fully connected neural network as independent variables, the differential equation is solved to obtain the hidden state vector at each interpolation time.
[0007] The method for improving the resolution of load data in the distribution area also includes: The load data resolution enhancement model for the aforementioned transformer area is pre-trained.
[0008] The pre-training of the transformer area load data resolution enhancement model includes: During the training process based on error backpropagation, the gradient of the error with respect to the parameters of the model for improving the resolution of the load data of the transformer area is taken as the unknown quantity. The solution result of the unknown quantity is obtained by solving the triplet of the ordinary differential equation.
[0009] The method for improving the resolution of transformer area load data, following the step of obtaining high-resolution time-series data of transformer area load, further includes: The predicted and actual values of the interpolation points in the high-resolution time series data of the load of the transformer area are compared and analyzed.
[0010] The comparative analysis of the predicted and actual values of the interpolation points in the high-resolution time series data of the transformer area load includes: The predicted and actual values of the interpolation points in the high-resolution time series data of the transformer area load are compared and analyzed according to the preset error index.
[0011] On one hand, this invention proposes a transformer area load data resolution enhancement device, which is executed based on a transformer area load data resolution enhancement model. The transformer area load data resolution enhancement model includes a time-series encoder, a time-series decoder, and a fully connected neural network layer connected in sequence. The time-series decoder is constructed based on a neural differential equation, which includes a fully connected neural network for fitting the rate of change of the hidden state vector over time, representing the temporal features of the implicit input data, and a differential equation constructed based on the parameters of the hidden state vector related to the temporal features of the implicit input data, the parameters of the fully connected neural network, each interpolation time, and the rate of change. The transformer area load data resolution enhancement device includes: The first acquisition unit is used to encode the low-resolution time series data of the transformer area load based on the time encoder to obtain the hidden state vector containing the time series features of the hidden input data. The second acquisition unit is used to decode the hidden state vector of the temporal features of the implicit input data based on the temporal decoder to obtain the hidden state vector at each interpolation time. An interpolation unit is used to map the hidden state vectors at each interpolation time point to interpolation results based on the fully connected neural network layer, and insert the interpolation results into the low-resolution time series data of the transformer area load to obtain high-resolution time series data of the transformer area load.
[0012] In another aspect, embodiments of the present invention provide a computer device, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements the following method: The time-series encoder is used to encode the low-resolution time-series data of the transformer area load to obtain a hidden state vector that contains the temporal features of the input data. Based on the time-series decoder, the hidden state vector of the hidden input data temporal features is decoded to obtain the hidden state vector at each interpolation time. Based on the fully connected neural network layer, the hidden state vectors at each interpolation time point are mapped to the interpolation results, and the interpolation results are inserted into the low-resolution time series data of the transformer area load to obtain the high-resolution time series data of the transformer area load.
[0013] This invention provides a computer-readable storage medium, comprising: The computer-readable storage medium stores a computer program that, when executed by a processor, implements the following method: The time-series encoder is used to encode the low-resolution time-series data of the transformer area load to obtain a hidden state vector that contains the temporal features of the input data. Based on the time-series decoder, the hidden state vector of the hidden input data temporal features is decoded to obtain the hidden state vector at each interpolation time. Based on the fully connected neural network layer, the hidden state vectors at each interpolation time point are mapped to the interpolation results, and the interpolation results are inserted into the low-resolution time series data of the transformer area load to obtain the high-resolution time series data of the transformer area load.
[0014] This invention also provides a computer program product, which includes a computer program that, when executed by a processor, implements the following method: The time-series encoder is used to encode the low-resolution time-series data of the transformer area load to obtain a hidden state vector that contains the temporal features of the input data. Based on the time-series decoder, the hidden state vector of the hidden input data temporal features is decoded to obtain the hidden state vector at each interpolation time. Based on the fully connected neural network layer, the hidden state vectors at each interpolation time point are mapped to the interpolation results, and the interpolation results are inserted into the low-resolution time series data of the transformer area load to obtain the high-resolution time series data of the transformer area load.
[0015] The method and apparatus for improving the resolution of transformer area load data provided in this invention encode low-resolution time series data of transformer area load based on the time encoder to obtain a hidden state vector containing the temporal features of the implicit input data; decode the hidden state vector containing the temporal features of the implicit input data based on the time decoder to obtain the hidden state vector at each interpolation time; map the hidden state vector at each interpolation time to the interpolation result based on the fully connected neural network layer, and insert the interpolation result into the low-resolution time series data of transformer area load to obtain high-resolution time series data of transformer area load, which can effectively improve the interpolation accuracy of transformer area load data. Attached Figure Description
[0016] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort. In the drawings: Figure 1 This is a flowchart illustrating a method for improving the resolution of transformer load data according to an embodiment of the present invention.
[0017] Figure 2 This is a schematic diagram of the model structure provided in an embodiment of the present invention.
[0018] Figure 3 This is a comparative analysis diagram provided in the embodiments of the present invention.
[0019] Figure 4 This is a schematic diagram of the structure of a transformer area load data resolution improvement device provided in an embodiment of the present invention.
[0020] Figure 5 This is a schematic diagram of the physical structure of a computer device provided in an embodiment of the present invention. Detailed Implementation
[0021] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the embodiments of the present invention will be further described in detail below with reference to the accompanying drawings. Here, the illustrative embodiments and descriptions of the present invention are used to explain the present invention, but are not intended to limit the present invention. It should be noted that, unless otherwise specified, the embodiments and features in the embodiments of this application can be arbitrarily combined with each other.
[0022] Figure 1 This is a flowchart illustrating a method for improving the resolution of transformer load data according to an embodiment of the present invention, as shown below. Figure 1 As shown, the transformer area load data resolution improvement method provided in this embodiment of the invention is executed based on a transformer area load data resolution improvement model; the transformer area load data resolution improvement model includes a time-series encoder, a time-series decoder, and a fully connected neural network layer connected in sequence; the time-series decoder is constructed based on a neural differential equation, which includes a fully connected neural network for fitting the rate of change of the hidden state vector of the latent input data time-series features over time, and a differential equation constructed based on the relevant parameters of the hidden state vector of the latent input data time-series features, the parameters of the fully connected neural network, each interpolation time, and the rate of change; the transformer area load data resolution improvement method includes: Step S1: Encode the low-resolution time series data of the transformer area load based on the time encoder to obtain the hidden state vector containing the time series features of the implicit input data.
[0023] Step S2: Based on the time-series decoder, the hidden state vector of the hidden input data temporal features is decoded to obtain the hidden state vector at each interpolation time.
[0024] Step S3: Based on the fully connected neural network layer, the hidden state vectors at each interpolation time point are mapped to interpolation results, and the interpolation results are inserted into the low-resolution time series data of the transformer area load to obtain high-resolution time series data of the transformer area load.
[0025] In step S1 above, the device encodes the low-resolution time-series data of the transformer load based on the time encoder to obtain a hidden state vector that implies the time-series characteristics of the input data. The device can be a computer device, such as a server, that executes this method. The acquisition, storage, use, and processing of data in this application all comply with relevant regulations.
[0026] like Figure 2 The following is an explanation of the model for improving the resolution of load data in the distribution area: The timing encoder is used to output a hidden state vector that implies the timing characteristics of the input data based on the historical time series data of the input transformer area load. The timing decoder is used to output the hidden state vector at each interpolation time based on the hidden state vector of the implicit temporal features of the input data.
[0027] The fully connected neural network layer is used to insert the hidden state vector of each interpolation time into the historical time series data of the transformer area load according to the target interpolation time period, so as to obtain the historical time series interpolated data of the transformer area load.
[0028] Before implementing the method of this invention, modeling can be performed first. Specifically, a temporal encoder can be constructed based on a recurrent neural network, as explained below: Constructing a time-series encoder based on a recurrent neural network (RNN) is an effective method for processing sequential data. Its core idea is to use the temporal expansion properties of RNNs to encode variable-length input sequences into fixed-length vector representations.
[0029] The input sequence can be processed step by step to update the hidden state, as shown in equation (1): = RNN ( , (1) in, Represents a recurrent neural network. For low-resolution time series data of transformer area load in Power value at time, for The hidden state vector at time step 1. for The hidden state vector at time step 1. The final hidden state, or the aggregation of all hidden states, is encoded as a sequence.
[0030] The temporal decoder is constructed based on neural differential equations, which include a fully connected neural network for fitting the rate of change of the hidden state vector over time to the temporal features of the hidden input data, and differential equations constructed based on the parameters of the hidden state vector related to the temporal features of the hidden input data, the parameters of the fully connected neural network, each interpolation time, and the rate of change.
[0031] The neural differential equation based on a fully connected neural network can be expressed by equation (2): (2) in, Hidden state vector at discrete time points The continuous expression, where t represents the time variable. This represents the parameters of a fully connected neural network. This represents the rate of change of the hidden state vector, which implies the temporal characteristics of the input data, over time. A fully connected neural network can have one hidden layer or multiple hidden layers; the model accuracy of a fully connected neural network with multiple hidden layers is better than that of a fully connected neural network with one hidden layer.
[0032] The parameters related to the hidden state vector that implies the temporal features of the input data include the initial time of the hidden state vector. It refers to the starting time of the low-resolution time series data of the transformer area load, and also includes the hidden state vector at the initial time. The differential equation can be expressed by equation (3): (3) in, This represents the hidden state vector at the k-th interpolation time. This represents a solver for ordinary differential equations. This represents the k-th interpolation time. The interpolation time interval can be set independently according to the actual situation, and can be selected as 15 minutes. (The initial time is...) It is also definite, therefore, at the initial time By taking the time starting point as the interpolation time interval, an interpolation time can be obtained.
[0033] Unlike traditional neural networks that discretely solve for the hidden state vector at the interpolation time of transformer load, neural differential equations combine neural networks with differential equations. By using neural networks to parameterize the continuous dynamic evolution of the hidden state vector at the interpolation time to construct differential equations, the hidden state at the interpolation time can be obtained by solving the differential equations. This modeling method can better adapt to the continuous dynamic evolution of the transformer load time series characteristics and has stronger interpretability. The continuous dynamic evolution refers to the continuous evolution of time series characteristics over time.
[0034] In step S2 above, the device decodes the hidden state vector of the temporal features of the implicit input data based on the temporal decoder to obtain the hidden state vector at each interpolation time. The decoding of the hidden state vector of the temporal features of the implicit input data based on the temporal decoder includes: The rate of change is obtained by fitting the hidden state vector of the temporal features of the implicit input data to the fully connected neural network; this rate of change is... .
[0035] Using the rate of change, the initial hidden state vector corresponding to the initial time of the hidden state vector containing the temporal characteristics of the implicit input data, the initial time, each interpolation time, and the parameters of the fully connected neural network as independent variables, the differential equation is solved to obtain the hidden state vector at each interpolation time. A differential equation solver can be used to continuously solve for each interpolation time. Hidden state vector .
[0036] In step S3 above, the device maps the hidden state vectors at each interpolation time point to interpolation results based on the fully connected neural network layer, and inserts the interpolation results into the low-resolution time series data of the transformer area load to obtain high-resolution time series data of the transformer area load. The interpolation time points can be determined according to actual needs. For example, the existing low-resolution time series data of the transformer area load can be understood as time series data with low refinement, where the time interval between two adjacent sampling times is 1 hour. However, in practice, it is necessary to obtain high-resolution time series data with a time interval between two adjacent sampling times of 15 minutes. This requires introducing two interpolation times between every two adjacent sampling times of the time series data with a time resolution of 1 hour.
[0037] The pre-trained model for improving the resolution of the transformer area load data includes: During the training process based on error backpropagation, the gradient of the error with respect to the parameters of the model for improving the resolution of the load data of the transformer area is taken as the unknown quantity. The solution result of the unknown quantity is obtained by solving the triplet of the ordinary differential equation.
[0038] Before using the distribution area load data resolution enhancement model for interpolation, the model needs to be trained. Training methods can be conventional in the field, or the adjoint sensitivity method (ASM) can be used to achieve error backpropagation. The ASM is an important method for analyzing the impact of system parameter changes on output. This method constructs an analytical framework based on mathematical principles and calculates sensitivity values through specific algorithms, enabling efficient assessment of the specific effects of changes in different factors in the system. The core of the ASM lies in solving the adjoint equation, achieving accurate sensitivity calculation through adjoint variables.
[0039] The solution to the unknown quantity can be obtained by solving the triplet of the ordinary differential equation consisting of equations (2), (4) and (5).
[0040] (4) (5) in, Represents the accompanying state auxiliary variable, Indicates about parameters Auxiliary variables, express The transpose of . The solution process is performed in the order of equation (6), equation (7) and equation (8).
[0041] (6) (7) (8) in, This represents the gradient of the training loss function of the load data resolution improvement model for the transformer substation with respect to the model parameters.
[0042] After the step of obtaining high-resolution time-series data of the distribution area load, the method for improving the resolution of the distribution area load data further includes: The predicted and actual values of the interpolation points in the high-resolution time series data of the load of the transformer area are compared and analyzed.
[0043] The comparative analysis of the predicted and actual values of the interpolation points in the high-resolution time series data of the transformer area load includes: The predicted and actual values of interpolation points in the high-resolution time series data of the transformer area load are compared and analyzed based on preset error indices. The preset error indices can be selected from the three indices in Table 1, where MAE represents the mean absolute error, RMSE represents the weighted root mean square error, and R... 2 This represents the correlation coefficient.
[0044] Comparative analysis results are as follows Figure 3 As shown, the time resolution of load data for a certain distribution area is increased from 1 hour to 15 minutes. The comparison curve between the interpolated predicted value and the actual value is as follows. Figure 3 As shown in Table 1, the error in the interpolation data of this invention is small and within a controllable range.
[0045] Table 1
[0046] The method for improving the resolution of transformer area load data provided in this invention utilizes neural differential equations to simulate the dynamic behavior of transformer area loads. It performs high-precision interpolation processing on the original low-frequency acquired load data, which helps to more accurately depict user electricity consumption patterns, better reflect the true load fluctuation characteristics, and provide richer information for load forecasting. Furthermore, compared to traditional deep learning model-based methods for refining and improving the resolution of transformer area load data, the neural differential equation-based method fully considers the dynamic evolution mechanism of load power at different interpolation times, improving the interpretability and accuracy of the interpolation model.
[0047] The method for improving the resolution of transformer area load data provided in this invention is based on a transformer area load data resolution improvement model. The model includes a time-series encoder, a time-series decoder, and a fully connected neural network layer connected sequentially. The time-series decoder is constructed based on a neural differential equation, which includes a fully connected neural network for fitting the rate of change of the hidden state vector over time, representing the temporal features of the implicit input data, and a differential equation constructed based on the parameters of the hidden state vector related to the temporal features of the implicit input data, the parameters of the fully connected neural network, each interpolation time, and the rate of change. The method for improving the resolution of transformer area load data includes: encoding low-resolution time-series data of transformer area load based on the time encoder to obtain a hidden state vector containing the temporal features of the implicit input data; decoding the hidden state vector containing the temporal features of the implicit input data based on the time decoder to obtain the hidden state vector at each interpolation time; mapping the hidden state vector at each interpolation time to an interpolation result based on the fully connected neural network layer, and inserting the interpolation result into the low-resolution time-series data of transformer area load to obtain high-resolution time-series data of transformer area load, which can effectively improve the interpolation accuracy of transformer area load data.
[0048] Further, the decoding process of the hidden state vector of the implicit input data temporal features based on the temporal decoder includes: The rate of change is obtained by fitting the hidden state vector of the temporal features of the implicit input data to the fully connected neural network; the above embodiments can be referred to for explanation, and will not be repeated here.
[0049] Using the rate of change, the initial hidden state vector corresponding to the initial time of the hidden state vector containing the temporal characteristics of the implicit input data, the initial time, each interpolation time, and the parameters of the fully connected neural network as independent variables, the differential equation is solved to obtain the hidden state vector at each interpolation time. This can be referred to the above embodiment for explanation, and will not be repeated here.
[0050] Furthermore, the method for improving the resolution of transformer area load data also includes: The load data resolution enhancement model for the aforementioned distribution area is pre-trained. This can be referred to the above embodiments for explanation, and will not be repeated here.
[0051] Furthermore, the pre-training of the transformer area load data resolution enhancement model includes: During the error backpropagation training process, the gradient of the error with respect to the parameters of the model for improving the resolution of the transformer area load data is taken as the unknown quantity. The solution result of the unknown quantity is obtained by solving the triplet of the ordinary differential equation. Refer to the above embodiment for further details.
[0052] Furthermore, after the step of obtaining high-resolution time-series data of the distribution area load, the method for improving the resolution of the distribution area load data further includes: The predicted and actual values of the interpolation points in the high-resolution time series data of the transformer area load are compared and analyzed. This can be referred to the above embodiments for explanation, and will not be repeated here.
[0053] Furthermore, the comparative analysis of the predicted and actual values of the interpolation points in the high-resolution time series data of the transformer area load includes: The predicted and actual values of interpolation points in the high-resolution time series data of the transformer area load are compared and analyzed according to a preset error index. This can be referred to the above embodiment for explanation, and will not be repeated here.
[0054] Figure 4 This is a schematic diagram of the structure of a transformer area load data resolution improvement device provided in an embodiment of the present invention, as shown below. Figure 4As shown, the transformer area load data resolution enhancement device provided in this embodiment of the invention is executed based on a transformer area load data resolution enhancement model; the transformer area load data resolution enhancement model includes a time-series encoder, a time-series decoder, and a fully connected neural network layer connected in sequence; the time-series decoder is constructed based on a neural differential equation, which includes a fully connected neural network for fitting the rate of change of the hidden state vector of the latent input data time-series features over time, and a differential equation constructed based on the relevant parameters of the hidden state vector of the latent input data time-series features, the parameters of the fully connected neural network, each interpolation time, and the rate of change; the transformer area load data resolution enhancement device includes a first acquisition unit 401, a second acquisition unit 402, and an interpolation unit 403, wherein: The first acquisition unit 401 is used to encode the low-resolution time series data of the transformer area load based on the time encoder to obtain the hidden state vector of the time series features of the hidden input data; the second acquisition unit 402 is used to decode the hidden state vector of the time series features of the hidden input data based on the time decoder to obtain the hidden state vector of each interpolation time; the interpolation unit 403 is used to map the hidden state vector of each interpolation time to the interpolation result based on the fully connected neural network layer, and insert the interpolation result into the low-resolution time series data of the transformer area load to obtain the high-resolution time series data of the transformer area load.
[0055] Specifically, the first acquisition unit 401 in the device is used to encode the low-resolution time series data of the transformer area load based on the time encoder to obtain the hidden state vector of the time series features of the implicit input data; the second acquisition unit 402 is used to decode the hidden state vector of the time series features of the implicit input data based on the time decoder to obtain the hidden state vector of each interpolation time; the interpolation unit 403 is used to map the hidden state vector of each interpolation time to the interpolation result based on the fully connected neural network layer, and insert the interpolation result into the low-resolution time series data of the transformer area load to obtain the high-resolution time series data of the transformer area load.
[0056] The transformer area load data resolution enhancement device provided in this embodiment of the invention is executed based on a transformer area load data resolution enhancement model. The transformer area load data resolution enhancement model includes a time-series encoder, a time-series decoder, and a fully connected neural network layer connected in sequence. The time-series decoder is constructed based on a neural differential equation, which includes a fully connected neural network for fitting the rate of change of the hidden state vector over time, representing the temporal features of the implicit input data, and a differential equation constructed based on the parameters of the hidden state vector related to the temporal features of the implicit input data, the parameters of the fully connected neural network, each interpolation time, and the rate of change. The transformer area load data resolution enhancement device is used to: encode low-resolution time series data of transformer area load based on the time encoder to obtain a hidden state vector containing the time series features of the implicit input data; decode the hidden state vector containing the time series features of the implicit input data based on the time decoder to obtain the hidden state vector at each interpolation time; map the hidden state vector at each interpolation time to an interpolation result based on the fully connected neural network layer, and insert the interpolation result into the low-resolution time series data of transformer area load to obtain high-resolution time series data of transformer area load, which can effectively improve the interpolation accuracy of transformer area load data.
[0057] The embodiments of the present invention provide a device for improving the resolution of load data in the transformer area, which can be used to execute the processing flow of the above-described method embodiments. Its functions will not be repeated here, but can be referred to the detailed description of the above-described method embodiments.
[0058] Figure 5 This is a schematic diagram of the physical structure of a computer device provided in an embodiment of the present invention, such as... Figure 5 As shown, the computer device includes: a memory 501, a processor 502, and a computer program stored in the memory 501 and executable on the processor 502. When the processor 502 executes the computer program, it implements the following method: The time-series encoder is used to encode the low-resolution time-series data of the transformer area load to obtain a hidden state vector that contains the temporal features of the input data. Based on the time-series decoder, the hidden state vector of the hidden input data temporal features is decoded to obtain the hidden state vector at each interpolation time. Based on the fully connected neural network layer, the hidden state vectors at each interpolation time point are mapped to the interpolation results, and the interpolation results are inserted into the low-resolution time series data of the transformer area load to obtain the high-resolution time series data of the transformer area load.
[0059] This embodiment discloses a computer program product, which includes a computer program that, when executed by a processor, implements the following method: The time-series encoder is used to encode the low-resolution time-series data of the transformer area load to obtain a hidden state vector that contains the temporal features of the input data. Based on the time-series decoder, the hidden state vector of the hidden input data temporal features is decoded to obtain the hidden state vector at each interpolation time. Based on the fully connected neural network layer, the hidden state vectors at each interpolation time point are mapped to the interpolation results, and the interpolation results are inserted into the low-resolution time series data of the transformer area load to obtain the high-resolution time series data of the transformer area load.
[0060] This embodiment provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the following method: The time-series encoder is used to encode the low-resolution time-series data of the transformer area load to obtain a hidden state vector that contains the temporal features of the input data. Based on the time-series decoder, the hidden state vector of the hidden input data temporal features is decoded to obtain the hidden state vector at each interpolation time. Based on the fully connected neural network layer, the hidden state vectors at each interpolation time point are mapped to the interpolation results, and the interpolation results are inserted into the low-resolution time series data of the transformer area load to obtain the high-resolution time series data of the transformer area load.
[0061] Compared with existing technologies, the present invention provides a method for improving the resolution of transformer area load data. This method is based on a transformer area load data resolution improvement model. The model includes a time-series encoder, a time-series decoder, and a fully connected neural network layer connected sequentially. The time-series decoder is constructed based on a neural differential equation, which includes a fully connected neural network for fitting the rate of change of the hidden state vector over time, representing the temporal features of the implicit input data. The model also includes parameters related to the hidden state vector based on the temporal features of the implicit input data, parameters of the fully connected neural network, and interpolation times. The differential equation constructed by the rate of change, and the method for improving the resolution of transformer area load data, include: encoding the low-resolution time series data of transformer area load based on the time encoder to obtain a hidden state vector containing the temporal features of the implicit input data; decoding the hidden state vector containing the temporal features of the implicit input data based on the time decoder to obtain the hidden state vector at each interpolation time; mapping the hidden state vector at each interpolation time to an interpolation result based on the fully connected neural network layer, and inserting the interpolation result into the low-resolution time series data of transformer area load to obtain high-resolution time series data of transformer area load, which can effectively improve the interpolation accuracy of transformer area load data.
[0062] 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.
[0063] 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.
[0064] 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.
[0065] 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.
[0066] In the description of this specification, the references to terms such as "an embodiment," "a specific embodiment," "some embodiments," "for example," "example," "specific example," or "some examples," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of the invention. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples.
[0067] The specific embodiments described above further illustrate the purpose, technical solution, and beneficial effects of the present invention. It should be understood that the above descriptions are merely specific embodiments of the present invention and are not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.
Claims
1. A method for improving the resolution of load data in a distribution area, characterized in that, The transformer area load data resolution improvement method is executed based on a transformer area load data resolution improvement model. The transformer area load data resolution improvement model includes a time-series encoder, a time-series decoder, and a fully connected neural network layer connected sequentially. The time-series decoder is constructed based on a neural differential equation, which includes a fully connected neural network for fitting the rate of change of the hidden state vector over time, representing the temporal features of the implicit input data, and a differential equation constructed based on the parameters of the hidden state vector related to the temporal features of the implicit input data, the parameters of the fully connected neural network, each interpolation time, and the rate of change. The transformer area load data resolution improvement method includes: The time-series encoder is used to encode the low-resolution time-series data of the transformer area load to obtain a hidden state vector that contains the temporal features of the input data. Based on the time-series decoder, the hidden state vector of the hidden input data temporal features is decoded to obtain the hidden state vector at each interpolation time. Based on the fully connected neural network layer, the hidden state vectors at each interpolation time point are mapped to the interpolation results, and the interpolation results are inserted into the low-resolution time series data of the transformer area load to obtain the high-resolution time series data of the transformer area load.
2. The method for improving the resolution of transformer area load data according to claim 1, characterized in that, The decoding process of the hidden state vector of the implicit input data temporal features based on the temporal decoder includes: The rate of change is obtained by fitting the hidden state vector of the temporal features of the implicit input data to the fully connected neural network. Using the rate of change, the initial hidden state vector corresponding to the initial time of the hidden state vector containing the temporal features of the implicit input data, the initial time, each interpolation time, and the parameters of the fully connected neural network as independent variables, the differential equation is solved to obtain the hidden state vector at each interpolation time.
3. The method for improving the resolution of transformer area load data according to claim 1, characterized in that, The method for improving the resolution of transformer area load data also includes: The load data resolution enhancement model for the aforementioned transformer area is pre-trained.
4. The method for improving the resolution of transformer area load data according to claim 3, characterized in that, The pre-trained model for improving the resolution of the transformer area load data includes: During the training process based on error backpropagation, the gradient of the error with respect to the parameters of the model for improving the resolution of the load data of the transformer area is taken as the unknown quantity. The solution result of the unknown quantity is obtained by solving the triplet of the ordinary differential equation.
5. The method for improving the resolution of transformer area load data according to any one of claims 1 to 4, characterized in that, After the step of obtaining high-resolution time-series data of the distribution area load, the method for improving the resolution of the distribution area load data further includes: The predicted and actual values of the interpolation points in the high-resolution time series data of the load of the transformer area are compared and analyzed.
6. The method for improving the resolution of transformer area load data according to claim 5, characterized in that, The comparative analysis of the predicted and actual values of the interpolation points in the high-resolution time series data of the transformer area load includes: The predicted and actual values of the interpolation points in the high-resolution time series data of the transformer area load are compared and analyzed according to the preset error index.
7. A device for improving the resolution of load data in a distribution area, characterized in that, The transformer area load data resolution enhancement device is executed based on the transformer area load data resolution enhancement model; the transformer area load data resolution enhancement model includes a time-series encoder, a time-series decoder, and a fully connected neural network layer connected in sequence; the time-series decoder is constructed based on a neural differential equation, which includes a fully connected neural network for fitting the rate of change of the hidden state vector of the latent input data time-series features over time, and a differential equation constructed based on the relevant parameters of the hidden state vector of the latent input data time-series features, the parameters of the fully connected neural network, each interpolation time, and the rate of change; the transformer area load data resolution enhancement device includes: The first acquisition unit is used to encode the low-resolution time series data of the transformer area load based on the time encoder to obtain the hidden state vector containing the time series features of the hidden input data. The second acquisition unit is used to decode the hidden state vector of the temporal features of the implicit input data based on the temporal decoder to obtain the hidden state vector at each interpolation time. An interpolation unit is used to map the hidden state vectors at each interpolation time point to interpolation results based on the fully connected neural network layer, and insert the interpolation results into the low-resolution time series data of the transformer area load to obtain high-resolution time series data of the transformer area load.
8. A computer device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the method of any one of claims 1 to 6.
9. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program that, when executed by a processor, implements the method of any one of claims 1 to 6.
10. A computer program product, characterized in that, The computer program product includes a computer program that, when executed by a processor, implements the method of any one of claims 1 to 6.