A power system load data super-resolution reconstruction method and system

Abstract

The application discloses a power system load data super-resolution reconstruction method and system, relates to the technical field of power system data quality improvement, and comprises the following steps: weighting weather influencing factors under different weather conditions, calculating the correlation between load data and weather influencing factor data, obtaining key influencing factor data, and performing dynamic range normalization processing by using a sliding window; high-resolution load data is obtained by using a trained time series generation model to perform super-resolution reconstruction on the normalized data; during training, the time series generation model adjusts the loss function weight according to the change of the current training loss; after training is completed, the combination features of the normalized load data and the key influencing factor data are extracted to perform super-resolution reconstruction. The defects of insufficient resolution of low-quality sampling data are overcome, the quality of reconstructed data is improved through meteorological factor constraint, and the problem that low-resolution sampling data cannot meet the training or use requirements is solved.

Description

Technical Field

[0001] This invention relates to the field of power system data quality improvement technology, and in particular to a method and system for super-resolution reconstruction of power system load data. Background Technology

[0002] The statements in this section are merely background information related to the present invention and do not necessarily constitute prior art.

[0003] With the continuous changes in the global energy structure, the power system is undergoing unprecedented transformation. On the one hand, the unexpected installation of renewable energy brings strong randomness and intermittency to the operation of the power system, which requires the system to have high-frequency data to assist in power flow scheduling and operation optimization. On the other hand, the large-scale deployment of new loads such as electric vehicles and data centers requires high-quality power supply and real-time coordination with the power system, which requires power system operators to have high-quality, real-time system operation data to meet users' additional needs.

[0004] However, existing sampling data has the following problems: On the one hand, due to reasons such as power system sensor malfunctions and transmission line faults, there is a large amount of missing data in the power data; on the other hand, due to limitations such as sampling equipment and transmission bandwidth, the data sampling frequency of the power grid in some areas is relatively low. For example, in existing monitoring and data acquisition systems, most of the meters in the smart grid are traditional sensors deployed many years ago. Traditional instruments collect measurement values ​​at a relatively low frequency, thus generating a small amount of effective data.

[0005] Currently, the solutions proposed by experts and scholars to address the problems of limited effective data and low sampling frequency can be roughly divided into two categories: one is to recover missing data, but this cannot fundamentally solve the problem of low sampling frequency; the other is data augmentation based on super-resolution reconstruction, but this is mostly based on generative adversarial networks. Traditional generative adversarial network models have certain limitations in multi-dimensional data processing and cannot efficiently process the current complex power system time series data.

[0006] Power system data are all time-series data. Due to the high proportion of renewable energy integration, the current differences between high-frequency and low-frequency data are mainly affected by weather conditions such as temperature, relative humidity, and wind speed. However, existing methods mainly start from the single time-series features of power system operation data and perform super-resolution reconstruction based on these features, without considering the impact of external features such as weather factors on data characteristics. Furthermore, traditional methods fail to capture the long-distance dependencies inherent in time-series data sampled over a long period, resulting in low-quality final data. Summary of the Invention

[0007] To address the aforementioned issues, this invention proposes a method and system for super-resolution reconstruction of power system load data. It considers the impact of weather factors, analyzes key weather-related factors most relevant to load changes, performs super-resolution reconstruction based on an improved time-series generation model, and adjusts model parameters according to the training process to achieve super-resolution reconstruction of low-resolution load data from low-voltage distribution areas.

[0008] To achieve the above objectives, the present invention adopts the following technical solution:

[0009] In a first aspect, the present invention provides a method for super-resolution reconstruction of power system load data, comprising:

[0010] After acquiring load data and weather influencing factor data, and assigning weights to each weather influencing factor under different weather conditions, the correlation between load data and weather influencing factor data is calculated to obtain key influencing factor data.

[0011] A sliding window is used to perform dynamic range normalization on load data and key influencing factor data;

[0012] The normalized load data and key influencing factor data are super-resolution reconstructed using a trained time series generation model to obtain high-resolution load data. After inverse normalization of the high-resolution load data, high-resolution load data that conforms to the actual dimensions are obtained.

[0013] The time series generation model adjusts the weights of the loss function based on changes in the current training loss during training. After training, it performs super-resolution reconstruction by extracting combined features of normalized load data and key influencing factor data.

[0014] As an alternative implementation method, the correlation between load data and weather influencing factor data is:

[0015]

[0016] In the formula, r is the correlation coefficient, and w k It is the weight of the kth weather influencing factor, x i and y i Let be the i-th data value of the load data sample and the i-th data value of the weather influencing factor data sample under the k-th weather influencing factor, where n is the total number of data in the sample. and This is the average value.

[0017] As an alternative implementation method, weather influencing factors include temperature, light intensity, relative humidity, and wind speed. The process of assigning weights to each weather influencing factor under different weather conditions includes:

[0018] The temperature weight w1 is: w1 = α TE k T k S k E,TE ;

[0019] The illumination weight w2 is: w2 = α L k T k W ;

[0020] The relative humidity weight w3 is: w3 = α H k T k S ;

[0021] The wind speed weight w4 is: w4 = α W k WS k E,WS ;

[0022] Where, α TE It is the temperature reference weight; k E，TE It is the extreme temperature correction factor; α L It is the illumination baseline weight; k W It is the weather coefficient; α H It is the humidity baseline weight; α W It is the baseline weight for wind speed; k WS It is the wind speed intensity coefficient; k E，WS It is the extreme weather correction factor; k T It is the time coefficient, k S It is a seasonal coefficient.

[0023] As an alternative implementation, the time series generation model includes a generator and a discriminator. The generator uses a two-layer long short-term memory network to extract combined features of normalized load data and key influencing factor data, and adds a self-attention mechanism to each long short-term memory network layer. Then, a deconvolution layer is used to expand the normalized load data into high-resolution load data, and the generated high-resolution load data is transmitted to the discriminator.

[0024] As an alternative implementation, the total loss function of the time series generation model is composed of a weighted sum of the generator loss function, the discriminator loss function, and the time series consistency loss function, and the weight of each loss function is adjusted according to the change of the current training loss.

[0025] As an alternative implementation, the generator loss function weight x1 is:

[0026] Where, x 10λ is the generator loss baseline weight; Δgen_loss is the difference between the previous generator loss and the current loss; λ is the adjustment coefficient; gen_loss(t-1) is the previous generator loss;

[0027] The discriminator loss function weights x2 are:

[0028] Where, x 20 Δdisc_loss is the base weight of the discriminator loss; Δdisc_loss is the difference between the discriminator loss in the previous round and the current loss; disc_loss(t-1) is the discriminator loss in the previous round.

[0029] The weight x3 of the time consistency loss function is:

[0030] Where, x 30 Δts_loss is the baseline weight of the time consistency loss; Δts_loss is the difference between the time consistency loss of the previous round and the current loss; ts_loss(t-1) is the time consistency loss of the previous round.

[0031] In a second aspect, the present invention provides a power system load data super-resolution reconstruction system, comprising:

[0032] The correlation analysis module is configured to acquire load data and weather influencing factor data, assign weights to each weather influencing factor under different weather conditions, calculate the correlation between load data and weather influencing factor data, and thus obtain key influencing factor data.

[0033] The normalization module is configured to use a sliding window to perform dynamic range normalization on load data and key influencing factor data;

[0034] The reconstruction module is configured to perform super-resolution reconstruction on the normalized load data and key influencing factor data using a trained time series generation model to obtain high-resolution load data. After inverse normalization of the high-resolution load data, high-resolution load data that conforms to the actual dimensions is obtained.

[0035] The time series generation model adjusts the weights of the loss function based on changes in the current training loss during training. After training, it performs super-resolution reconstruction by extracting combined features of normalized load data and key influencing factor data.

[0036] Thirdly, the present invention provides an electronic device including a memory and a processor, and computer instructions stored in the memory and running on the processor, wherein the computer instructions, when executed by the processor, perform the method described in the first aspect.

[0037] Fourthly, the present invention provides a computer-readable storage medium for storing computer instructions, which, when executed by a processor, perform the method described in the first aspect.

[0038] Fifthly, the present invention provides a computer program product, including a computer program that, when executed by a processor, implements the method described in the first aspect.

[0039] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0040] This invention proposes a super-resolution reconstruction method for low-voltage power load data that takes into account the influence of weather factors. The method involves sequentially processing outliers in the original data to obtain preprocessed data, performing correlation analysis on the preprocessed data to obtain weather factor data most relevant to load changes, and performing data super-resolution reconstruction based on an improved time series generation model. The model parameters are adjusted according to the training process to achieve super-resolution reconstruction of low-voltage load data in low-voltage areas.

[0041] This invention uses an improved Pearson correlation analysis method, which, compared to traditional correlation analysis methods, innovatively incorporates weights related to weather influencing factors. Different weights are assigned to each weather influencing factor under different weather conditions, and then the correlation between load data and weather influencing factor data is analyzed separately to reflect the correlation between load data and weather influencing factor data under different weather conditions.

[0042] This invention uses an improved normalization method. Compared with traditional correlation analysis methods, this invention takes into account the large variation of weather factors and innovatively proposes a dynamic normalization method based on a sliding window. This method dynamically adjusts the normalization range to better adapt to the dynamic changes of time series data, eliminates the dimensional differences between data features, and avoids the problem of data failure after normalization due to a fixed global normalization range.

[0043] This invention uses an improved time series generation model for super-resolution reconstruction. The improved model architecture allows the generated data to take into account the influence of weather factors. At the same time, it introduces a loss function that takes into account weights. Compared with traditional methods, the improved method can automatically update the weights according to the changing trend of the loss in each training session, so that the generated data can better match the real data.

[0044] This invention proposes a method and system for super-resolution reconstruction of power system load data. Specifically, it performs super-resolution reconstruction of low-resolution load data from low-voltage distribution areas of the power system. Taking into account the weather impact caused by the high proportion of renewable energy access, it overcomes the deficiency of insufficient resolution in low-quality sampling data. By constraining meteorological factors, it improves the quality of reconstructed data and solves the problem that the low resolution of sampling data cannot meet the requirements for training or use. It can be applied to the field of distribution network data quality improvement and enhances the development and utilization value of massive data resources in the power system.

[0045] Advantages of additional aspects of the invention will be set forth in part in the description which follows, and in part will be obvious from the description, or may be learned by practice of the invention. Attached Figure Description

[0046] 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 embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on the provided drawings without creative effort.

[0047] Figure 1 This is a flowchart of the power system load data super-resolution reconstruction method provided in Embodiment 1 of the present invention;

[0048] Figure 2 This is a schematic diagram of the workflow of the time series generation model provided in Embodiment 1 of the present invention. Detailed Implementation

[0049] The present invention will be further described below with reference to the accompanying drawings and embodiments.

[0050] It should be noted that the following detailed descriptions are exemplary and intended to provide further illustration of the invention. Unless otherwise specified, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains.

[0051] It should be noted that the terminology used herein is for the purpose of describing particular embodiments only and is not intended to limit the scope of exemplary embodiments according to the invention. As used herein, unless the context clearly indicates otherwise, the singular form is intended to include the plural form as well. Furthermore, it should be understood that the terms “comprising” and “including”, and any variations thereof, are intended to cover non-exclusive inclusion, for example, a process, method, system, product, or apparatus that includes a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such process, method, product, or apparatus.

[0052] Where there is no conflict, the embodiments and features in the embodiments of the present invention can be combined with each other.

[0053] Example 1

[0054] like Figure 1 As shown, this embodiment provides a super-resolution reconstruction method for power system load data that considers the impact of weather factors, including:

[0055] After acquiring load data and weather influencing factor data, and assigning weights to each weather influencing factor under different weather conditions, the correlation between load data and weather influencing factor data is calculated to obtain key influencing factor data.

[0056] A sliding window is used to perform dynamic range normalization on load data and key influencing factor data;

[0057] The normalized load data and key influencing factor data are super-resolution reconstructed using a trained time series generation model to obtain high-resolution load data. After inverse normalization of the high-resolution load data, high-resolution load data that conforms to the actual dimensions are obtained.

[0058] The time series generation model adjusts the weights of the loss function based on changes in the current training loss during training. After training, it performs super-resolution reconstruction by extracting combined features of normalized load data and key influencing factor data.

[0059] The method of this embodiment will be described in detail below.

[0060] S1: Obtain load data and weather influencing factor data; First, use a random forest model to preprocess outliers in the load data and weather influencing factor data to identify and replace outliers in the low-resolution load data and weather influencing factor data.

[0061] The weather factors include temperature, light intensity, relative humidity, and wind speed; the load data uses low-resolution load data from low-voltage distribution areas (low-voltage distribution areas refer to areas supplied by transformers at low voltage (380V)).

[0062] Then, using the improved Pearson correlation analysis method, the correlation between load data and weather influencing factor data was analyzed to identify key influencing factor data, and preprocessed load data and key influencing factor data were obtained.

[0063] Specifically:

[0064] In the correlation analysis, an improved Pearson correlation analysis method was used. This improved method innovatively incorporates weights related to weather influencing factors, assigning different weights to each weather influencing factor under different weather conditions. k Then, the correlation between load data and weather influencing factor data was analyzed to reflect the correlation between load data and weather influencing factor data under different weather conditions.

[0065] The correlation coefficient is:

[0066]

[0067] In the formula, r is the correlation coefficient, and w k It is the weight of the kth weather influencing factor, x i and y i Let be the i-th data value of the load data sample and the i-th data value of the weather influencing factor data sample under the k-th weather influencing factor, where n is the total number of data in the sample. and This is the average value.

[0068] w k The settings are as follows:

[0069] (1) Temperature weighting: Temperature usually has a significant impact on load data, especially during extreme weather (such as high temperatures in summer or cold temperatures in winter), when it has a particularly significant impact on loads such as air conditioning and heating.

[0070] w1=α TE k T k S k E,TE ;

[0071] Where: w1 is the temperature weight; α TE It is the temperature reference weight; k T This is a time coefficient, divided into daytime and nighttime, with higher values ​​during the day and lower values ​​at night. Its value is automatically generated by the model. The model identifies the season corresponding to the input temperature data and divides the data range into daytime and nighttime. For example, when the model identifies that the temperature data was sampled in spring, it sets the daytime range to 6:00-18:00; k S This is a seasonal coefficient, divided into spring / autumn and winter / summer. It is higher in winter / summer and lower in spring / autumn. Its value is automatically generated by the model, which identifies the season corresponding to the input temperature data and generates the corresponding value. k E，TE It is the extreme temperature correction factor, which is divided into high temperature and low temperature. The value is higher for high temperature and lower for low temperature. Its value is automatically generated by the model. The model determines the low temperature and high temperature by identifying whether the input temperature data exceeds the set threshold, and then generates the corresponding value.

[0072] (2) Sunlight weighting: Sunlight mainly affects the daytime load, especially the photovoltaic power generation load. When the daytime temperature is high, sunlight has a greater impact on the load.

[0073] w2=α L k T k W ;

[0074] Where: w2 is the illumination weight; α L It is the illumination baseline weight; k T This is a time coefficient, divided into daytime and nighttime. Its value is automatically generated by the model. The model identifies the season corresponding to the input illumination data and divides the data interval into daytime and nighttime. For example, when the model identifies that the illumination data was sampled in spring, it sets the daytime period to 6:00-18:00, with a value of 1 for daytime and 0 for nighttime; k W It is the sunshine / cloudy coefficient, which is divided into sunny and cloudy days. The value is higher on sunny days and lower on cloudy days. The value is automatically generated by the model. The model identifies the input light data, and strong light indicates sunny days, while weak light indicates cloudy days. Based on this, the data is divided into sunny and cloudy days, and the corresponding value is generated.

[0075] (3) Relative humidity weight: Relative humidity has a certain impact on the load in summer and winter, especially on the performance of air conditioning load (the air conditioning load is large when the humidity is high), but the range of humidity variation is relatively small, and usually has a small direct impact on the load.

[0076] w3=α H k T k S ;

[0077] Where: w3 is the relative humidity weight; α H It is the humidity baseline weight; k T This is a time coefficient, divided into daytime and nighttime, with higher values ​​during the day and lower values ​​at night. Its value is automatically generated by the model. The model identifies the season corresponding to the input humidity data and divides the data range into daytime and nighttime. For example, when the model identifies that the humidity data was sampled in spring, it sets the daytime range to 6:00-18:00; k S It is a seasonal coefficient, divided into summer, winter and spring / autumn. It is higher in summer, followed by winter, and lower in spring / autumn. Its value is automatically generated by the model. The model generates the corresponding value by identifying the season corresponding to the input humidity data.

[0078] (4) Wind speed weighting:

[0079] w4=α W k WS k E,WS ;

[0080] Where: w4 is the wind speed weight; α W It is the baseline weight for wind speed; k WS This is the wind speed intensity coefficient, categorized into strong wind, normal wind, and calm / light wind. Strong wind has a higher value, normal wind has a value of 1, and calm / light wind has a lower value. Its value is automatically generated by the model. The model identifies which range of strong wind, normal wind, or calm / light wind the input wind speed data belongs to, determines the wind speed accordingly, and generates the corresponding value. E，WS It is an extreme weather correction factor, divided into strong wind and normal wind. It is higher for strong wind and 1 for normal wind. Its value is automatically generated by the model. The model identifies which interval of strong wind or normal wind the input wind speed data belongs to, and determines the wind speed accordingly, thereby generating the corresponding value.

[0081] S2: For time series data, the range of the data may change over time. At the same time, considering the large variation of weather factors, in order to avoid deviations and invalid data caused by fixed range normalization, this embodiment proposes an improved normalization method to normalize the load data and key influencing factor data. By using a fixed-size sliding window, the minimum and maximum values ​​within the current time window are calculated to perform dynamic range normalization processing on low-resolution load data and key influencing factor data.

[0082] The size of the sliding window can be manually adjusted according to the number of data points. As the input load data and key influencing factor data are traversed, the window slides forward, thereby dynamically updating the normalization parameters. By dynamically adjusting the normalization range, it can better adapt to the dynamic changes of time series data, eliminate the dimensional differences between data features, and avoid the failure of normalized data due to a fixed global normalization range.

[0083] S3: Combine the normalized low-resolution load data with the key influencing factor data at the dimensional level. First, the normalized load data and key influencing factor data are input into the generator. The generator learns the combined time-series characteristics of the normalized load data and key influencing factor data to obtain high-resolution load data.

[0084] Specifically, in the generator, the generator architecture is improved by comprehensively considering the influence of weather factors and the temporal characteristics of the load, and a self-attention mechanism layer is added. The multi-dimensional combined data is used to learn the multi-dimensional fusion features of time series data. That is, a two-layer long short-term memory network (LSTM) is used to extract the combined features of low-resolution load data and key influencing factor data, and a self-attention mechanism is added to each LSTM layer. Then, a deconvolution layer is used to expand the low-resolution load data into high-resolution load data, and then the generated high-resolution load data is transmitted to the discriminator.

[0085] S4: Secondly, in the discriminator, the high-resolution payload data generated by the generator and the real high-resolution payload data are received. The discriminator learns to distinguish between real samples and generated samples and provides feedback parameters for the generated samples to the generator to update the generator parameters. By continuously updating the parameters, a well-trained super-resolution reconstruction model based on the time series generative model (TimeGAN) is obtained.

[0086] Figure 2 This diagram illustrates the workflow of the discriminator and generator in the TimeGAN model. Specifically, the generator learns the features of low-resolution data to generate high-resolution payload data. The discriminator distinguishes between real and generated samples, updates the generator parameters, and the loss function reaches its minimum. (Analysis) Figure 2 It can be seen that the generator and discriminator are trained interactively, constantly modifying their respective parameters until the loss function reaches its minimum value.

[0087] In the discriminator, the architecture is improved by using a two-layer LSTM to process time-series data and adding a self-attention mechanism layer to learn the long-range dependencies of the time-series data. A sigmoid activation function is used to output the binary judgment result. Interactive training between the generator and the discriminator is achieved by feeding the feedback parameters of the generated samples to the generator.

[0088] In the TimeGAN model, the parameters transmitted between the generator and the discriminator are called the loss function. In this embodiment, the total loss function consists of the generator loss function, the discriminator loss function, and the temporal consistency loss function ts_loss. The generator loss function gen_loss uses mean squared error loss, and the discriminator loss function disc_loss uses binary cross-entropy loss.

[0089] The time-series consistency loss function ts_loss is:

[0090]

[0091] In the formula, T is the length of the time series, and y_true t and y_pred t y_true represents the actual data and generated data at time t. t+1 and y_pred t+1 It consists of the actual data and generated data at time t+1.

[0092] Based on the model training bias and the characteristics of the generated data, the weights of the three loss functions are set as x1, x2, and x3, respectively. At the same time, the weights are automatically updated according to the changing trend of the loss in each training session to avoid premature saturation or neglect of a certain loss term.

[0093] The total loss function, total_loss, is:

[0094] total_loss=x1×gen_loss+x2×disc_loss+x3×ts_loss.

[0095] The generator loss function weight x1 is:

[0096]

[0097] Where, x 10 Δgen_loss is the generator loss baseline weight; Δgen_loss is the difference between the previous generator loss and the current loss; λ is the adjustment coefficient, which controls the rate of weight change; gen_loss(t-1) is the previous generator loss.

[0098] The discriminator loss function weights x2 are:

[0099]

[0100] Where, x 20 Δdisc_loss is the base weight of the discriminator loss; Δdisc_loss is the difference between the discriminator loss in the previous round and the current loss; disc_loss(t-1) is the discriminator loss in the previous round.

[0101] The weight x3 of the time consistency loss function is:

[0102]

[0103] Where, x 30 Δts_loss is the baseline weight of the time consistency loss; Δts_loss is the difference between the time consistency loss of the previous round and the current loss; ts_loss(t-1) is the time consistency loss of the previous round.

[0104] S5: By interactively training the generator and discriminator, the loss function is continuously updated to finally obtain the trained TimeGAN model; using the trained model, the actual low-resolution low-voltage substation load data is super-resolution reconstructed to obtain high-resolution low-voltage substation load data, and the generated data is inversely normalized to obtain high-resolution data that conforms to the actual dimensions.

[0105] This embodiment proposes a method for super-resolution reconstruction of low-resolution load data in low-voltage distribution areas of the power system. Taking into account the weather impact caused by the high proportion of renewable energy access, it overcomes the defect of insufficient resolution of low-quality sampling data, improves the quality of reconstructed data by constraining meteorological factors, and solves the problem that the low resolution of sampling data cannot meet the requirements for training or use. It can be applied to the field of distribution network data quality improvement and enhance the development and utilization value of massive data resources in the power system.

[0106] Example 2

[0107] This embodiment provides a power system load data super-resolution reconstruction system, including:

[0108] The correlation analysis module is configured to acquire load data and weather influencing factor data, assign weights to each weather influencing factor under different weather conditions, calculate the correlation between load data and weather influencing factor data, and thus obtain key influencing factor data.

[0109] The normalization module is configured to use a sliding window to perform dynamic range normalization on load data and key influencing factor data;

[0110] The reconstruction module is configured to perform super-resolution reconstruction on the normalized load data and key influencing factor data using a trained time series generation model to obtain high-resolution load data. After inverse normalization of the high-resolution load data, high-resolution load data that conforms to the actual dimensions is obtained.

[0111] The time series generation model adjusts the weights of the loss function based on changes in the current training loss during training. After training, it performs super-resolution reconstruction by extracting combined features of normalized load data and key influencing factor data.

[0112] It should be noted that the above modules correspond to the steps described in Embodiment 1, and the examples and application scenarios implemented by the above modules and the corresponding steps are the same, but are not limited to the content disclosed in Embodiment 1. It should also be noted that the above modules, as part of the system, can be executed in a computer system such as a set of computer-executable instructions.

[0113] In further embodiments, the following is also provided:

[0114] An electronic device includes a memory and a processor, as well as computer instructions stored in the memory and running on the processor, wherein the computer instructions, when executed by the processor, perform the method described in Embodiment 1. For brevity, further details are omitted here.

[0115] It should be understood that in this embodiment, the processor can be a central processing unit (CPU), or it can be other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor can be a microprocessor or any conventional processor, etc.

[0116] Memory may include read-only memory and random access memory, and provides instructions and data to the processor. A portion of memory may also include non-volatile random access memory. For example, memory may also store information about the device type.

[0117] A computer-readable storage medium for storing computer instructions, which, when executed by a processor, perform the method described in Embodiment 1.

[0118] The method in Example 1 can be directly implemented by a hardware processor, or implemented by a combination of hardware and software modules within the processor. The software modules can reside in readily available storage media in the field, such as random access memory, flash memory, read-only memory, programmable read-only memory, electrically erasable programmable memory, or registers. This storage medium is located in memory, and the processor reads information from the memory and, in conjunction with its hardware, completes the steps of the above method. To avoid repetition, a detailed description is not provided here.

[0119] A computer program product includes a computer program that, when executed by a processor, implements the method described in Embodiment 1.

[0120] The present invention also provides at least one computer program product tangibly stored on a non-transitory computer-readable storage medium. The computer program product includes computer-executable instructions, such as instructions included in program modules, which execute in a device on a target real or virtual processor to perform the processes / methods described above. Typically, program modules include routines, programs, libraries, objects, classes, components, data structures, etc., that perform specific tasks or implement specific abstract data types. In various embodiments, the functionality of program modules can be combined or divided among program modules as needed. The machine-executable instructions for the program modules can execute within a local or distributed device. In a distributed device, the program modules can reside in both local and remote storage media.

[0121] The computer program code used to implement the methods of the present invention may be written in one or more programming languages. This computer program code may be provided to a processor of a general-purpose computer, a special-purpose computer, or other programmable data processing device, such that when executed by the computer or other programmable data processing device, the program code causes the functions / operations specified in the flowcharts and / or block diagrams to be implemented. The program code may be executed entirely on a computer, partially on a computer, as a stand-alone software package, partially on a computer and partially on a remote computer, or entirely on a remote computer or server.

[0122] In the context of this invention, computer program code or related data may be carried by any suitable carrier to enable a device, apparatus, or processor to perform the various processes and operations described above. Examples of carriers include signals, computer-readable media, and the like. Examples of signals may include electrical, optical, radio, sound, or other forms of propagation signals, such as carrier waves, infrared signals, etc.

[0123] Those skilled in the art will recognize that the units and algorithm steps described in conjunction with the embodiments herein can be implemented in electronic hardware or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0124] While the specific embodiments of the present invention have been described above in conjunction with the accompanying drawings, this is not intended to limit the scope of protection of the present invention. Those skilled in the art should understand that various modifications or variations that can be made by those skilled in the art without creative effort based on the technical solutions of the present invention are still within the scope of protection of the present invention.

Claims

1. A method for super-resolution reconstruction of power system load data, characterized in that, include: After acquiring load data and weather influencing factor data, and assigning weights to each weather influencing factor under different weather conditions, the correlation between load data and weather influencing factor data is calculated to obtain key influencing factor data. A sliding window is used to perform dynamic range normalization on load data and key influencing factor data; The normalized load data and key influencing factor data are super-resolution reconstructed using a trained time series generation model to obtain high-resolution load data. After inverse normalization of the high-resolution load data, high-resolution load data that conforms to the actual dimensions are obtained. The time series generation model adjusts the weights of the loss function based on changes in the current training loss during training. After training, it performs super-resolution reconstruction by extracting combined features of normalized load data and key influencing factor data. The total loss function of the time series generation model is composed of a weighted average of the generator loss function, the discriminator loss function, and the temporal consistency loss function, and the weight of each loss function is adjusted according to the change of the current training loss. The time series generation model includes a generator and a discriminator. The generator uses a two-layer long short-term memory network to extract combined features of normalized load data and key influencing factor data. A self-attention mechanism is added to each long short-term memory network layer. Then, a deconvolution layer is used to expand the normalized load data into high-resolution load data, and the generated high-resolution load data is transmitted to the discriminator.

2. The method for super-resolution reconstruction of power system load data as described in claim 1, characterized in that, The correlation between load data and weather influencing factor data is as follows: ； In the formula, r It is the correlation coefficient. w k It is the weight of the kth weather influencing factor. x i and y i For the first load data sample i The data sample of the weather influencing factors under the kth weather influencing factor and the data value of the kth weather influencing factor. i One data value, n The total number of data points in the sample. and This is the average value.

3. The method for super-resolution reconstruction of power system load data as described in claim 2, characterized in that, Weather influencing factors include temperature, sunlight, relative humidity, and wind speed. The process of assigning weights to these factors under different weather conditions includes: Temperature weighting w 1 is: ; Illumination weight w 2 is: ; relative humidity weight w 3 is: ; Wind speed weight w 4 is: ; in, α TE It is a temperature reference weight; k E，TE It is an extreme temperature correction factor; α L It is the baseline weight of illumination; k W It is the weather coefficient; α H It is the humidity baseline weight; α W It is the wind speed baseline weight; k WS It is the wind speed intensity coefficient; k E，WS It is an extreme weather correction factor; k T It is a time coefficient. k S It is a seasonal coefficient.

4. The method for super-resolution reconstruction of power system load data as described in claim 1, characterized in that, Generator loss function weights x 1 is: ; in, x 10 It is the generator loss baseline weight; Δ gen_loss It is the difference between the generator loss in the previous round and the current loss; λ It is an adjustment factor; gen_loss ( t -1) is the generator loss from the previous round; Discriminator loss function weights x 2 is: ; in, x 20 It is the discriminator loss baseline weight; Δ disc_loss It is the difference between the discriminator loss in the previous round and the current loss; disc_loss ( t -1) is the loss of the discriminator in the previous round; Timing consistency loss function weights x 3 is: ; in, x 30 It is the baseline weight for time-series consistency loss; Δ ts_loss It is the difference between the time sequence consistency loss of the previous round and the current loss; ts_loss ( t -1) is the timing consistency loss from the previous round.

5. A power system load data super-resolution reconstruction system, characterized in that, include: The correlation analysis module is configured to acquire load data and weather influencing factor data, assign weights to each weather influencing factor under different weather conditions, calculate the correlation between load data and weather influencing factor data, and thus obtain key influencing factor data. The normalization module is configured to use a sliding window to perform dynamic range normalization on load data and key influencing factor data; The reconstruction module is configured to perform super-resolution reconstruction on the normalized load data and key influencing factor data using a trained time series generation model to obtain high-resolution load data. After inverse normalization of the high-resolution load data, high-resolution load data that conforms to the actual dimensions is obtained. The time series generation model adjusts the weights of the loss function based on changes in the current training loss during training. After training, it performs super-resolution reconstruction by extracting combined features of normalized load data and key influencing factor data. The total loss function of the time series generation model is composed of a weighted average of the generator loss function, the discriminator loss function, and the temporal consistency loss function, and the weight of each loss function is adjusted according to the change of the current training loss. The time series generation model includes a generator and a discriminator. The generator uses a two-layer long short-term memory network to extract combined features of normalized load data and key influencing factor data. A self-attention mechanism is added to each long short-term memory network layer. Then, a deconvolution layer is used to expand the normalized load data into high-resolution load data, and the generated high-resolution load data is transmitted to the discriminator.

6. An electronic device, characterized in that, It includes a memory and a processor, as well as computer instructions stored in the memory and running on the processor, which, when executed by the processor, perform the method according to any one of claims 1-4.

7. A computer-readable storage medium, characterized in that, Used to store computer instructions, which, when executed by a processor, perform the method described in any one of claims 1-4.

8. A computer program product, characterized in that, Includes a computer program, which, when executed by a processor, implements the method described in any one of claims 1-4.

Images