Weak power grid power load prediction method and system based on Informer model improved by time sequence embedding mechanism

Abstract

The invention provides a weak power grid power load prediction method and system based on an Informer model improved by a time sequence embedding mechanism, and belongs to the technical field of power grid power. Comprising the following steps: S1, for a micro-balance area power grid of a weak power grid, acquiring load measuring point operation parameters of the micro-balance area power grid and meteorological operation parameters of a micro-balance area to obtain a multi-source time sequence data set; s2, obtaining a load change parameter representing a load change degree based on the multi-source time sequence data set; s3, constructing a load prediction model; s4, building a holiday influence coefficient and an intra-week workday switching influence coefficient; s5, constructing a comprehensive time influence coefficient; and S6, performing result output on the comprehensive time influence coefficient. According to the method, the sensing capability of the abnormal load change of the weak power grid is improved, and the deployability and maintainability of the method in the actual operation of the weak power grid are improved.

Description

Technical Field

[0001] This invention relates to the field of power grid technology, and more specifically, to a method and system for predicting power load in weak power grids based on an improved Informer model with a time-series embedding mechanism. Background Technology

[0002] With the rapid integration of distributed new energy, electric vehicles, and flexible loads into the distribution side, the operating characteristics of micro-balanced power grids in weak power grids are becoming increasingly complex. Their loads exhibit frequent fluctuations, strong randomness, and significant time characteristics. In micro-balanced power grids, due to the relatively weak grid structure and limited regulation capacity, the accuracy of load forecasting results directly affects the safety, economy, and rationality of grid operation and dispatching decisions. Therefore, high-precision load forecasting for micro-balanced power grids has significant engineering application value.

[0003] Existing power load forecasting methods are mostly based on historical load data to build forecasting models. Some methods introduce meteorological factors or time features to improve forecast accuracy, but they still have certain shortcomings in practical applications. Traditional statistical methods or conventional machine learning models have limited ability to model long-term series dependencies and cannot effectively characterize the variation of load at different time scales. Furthermore, when processing load data in micro-balance zones, existing deep learning models often fail to fully consider the nonlinear impact of time factors such as holidays and weekday switching on load changes, resulting in insufficient adaptability of the models to sudden or structural load changes. Summary of the Invention

[0004] To overcome the above deficiencies, the present invention provides a method and system for predicting power load in weak power grids based on an improved Informer model with a time-series embedding mechanism, which overcomes or at least partially solves the above technical problems.

[0005] This invention is implemented as follows:

[0006] This invention provides a method for predicting power load in weak power grids based on an improved Informer model using a time-series embedding mechanism, comprising:

[0007] S1. For the micro-balanced power grid of the weak power grid, collect the load measurement point operation parameters of the micro-balanced power grid and the meteorological operation parameters of the micro-balanced power grid, and perform time alignment processing on the load measurement point operation parameters of the micro-balanced power grid and the meteorological operation parameters of the micro-balanced power grid to obtain a multi-source time series dataset.

[0008] S2. Based on multi-source time series datasets, perform differential calculations on the load time series to obtain load change parameters that characterize the degree of load change;

[0009] S3. Using an encoder-decoder structure, a time-series embedding module that switches between holidays and weekdays is introduced to construct a load forecasting model. After time-location encoding of the multi-source time-series dataset and the load time series, the model is input into the load forecasting model, and the load forecasting results of the micro-balanced power grid are output.

[0010] S4. Construct the impact coefficient of holidays based on multi-source time-series datasets. and the impact coefficient of weekday switching This is used to characterize the degree of influence of different time types on load changes;

[0011] S5. Based on the impact coefficient of holidays and the impact coefficient of weekday switching Weighted calculations are performed to construct a comprehensive time influence coefficient. And the overall time influence coefficient Conduct assessments and corrections;

[0012] S6. Impact coefficient on overall time The results are output for analysis of the load operation status of the power grid in the micro-balance zone.

[0013] In a preferred embodiment, S1 includes:

[0014] S11. The load measurement point operating parameters of the micro-balanced power grid include the instantaneous active power P of the power grid load, the instantaneous reactive power Q of the power grid load, and the cumulative energy value E.

[0015] S12. By setting up a data acquisition point at the outgoing end of the load measurement point in the micro-balance zone power grid, and installing current transformers and voltage transformers at the data acquisition point, connecting the current transformers and voltage transformers to the smart energy meter, the voltage and current are collected in real time through the energy meter, and the instantaneous active power P of the power grid load is calculated using the formula.

[0016] S13. Based on the current transformer and voltage transformer at the load measurement point outgoing terminal position in the micro-balanced power grid, the voltage and current are measured in real time, and the instantaneous reactive power Q of the power grid load is calculated by formula.

[0017] S14. Record the cumulative electrical energy value of the load measuring point in real time through the electricity meter to obtain the cumulative electrical energy value E of the power grid load;

[0018] S15. Perform time alignment processing on the instantaneous active power P, instantaneous reactive power Q, and cumulative energy value E at the load measurement points of the micro-balanced power grid, including:

[0019] A uniform sampling time interval is determined, a standard time series is constructed, the original time series of each parameter is interpolated and mapped onto the standard time series, outlier identification and removal are performed on the mapped time series, and the processed time series is standardized.

[0020] In a preferred embodiment, S1 further includes:

[0021] S16. The meteorological operating parameters of the micro-equilibrium zone include the temperature T and precipitation R of the micro-equilibrium zone;

[0022] S17. By selecting a fixed meteorological observation point within the micro-equilibrium zone and installing a temperature sensor at the meteorological observation point, the ambient temperature is measured in real time to obtain the temperature T of the micro-equilibrium zone.

[0023] S18. Rainfall in the micro-equilibrium zone is collected in real time by deploying rain gauges in the micro-equilibrium zone;

[0024] S19. Perform time resampling processing on the temperature T of the micro-equilibrium zone collected by the temperature sensor, map the temperature data to a standard time series, and when the temperature data of multiple meteorological observation points correspond to the same time point, perform statistical processing on the multiple temperature data to obtain the temperature data of the micro-equilibrium zone corresponding to the same time point.

[0025] The rainfall R collected by the rain gauge in the micro-equilibrium area is processed by time segmentation. The rainfall is accumulated or redistributed over time according to the standard time series so that each standard time period corresponds to a set of rainfall data. This yields the micro-equilibrium area rainfall data that is consistent with the standard time series, thereby constructing a multi-source time series dataset.

[0026] In a preferred embodiment, S2 includes:

[0027] S21. Extract the load time series corresponding to the load measurement points based on the multi-source time series dataset. The load time series includes load data arranged in a uniform time series.

[0028] S22. Based on the sampling time interval of the multi-source time series dataset, determine the differential time scale of the load time series so that the load changes between corresponding adjacent time points can be calculated differentially.

[0029] S23. Perform differential calculations on the load time series according to the differential time scale to obtain the load change between adjacent time points, where the load change is used to reflect the magnitude and direction of load change over time.

[0030] S24. Construct load change characteristics based on load change amounts, correlate these characteristics with corresponding time points to form a load change characteristic sequence, and construct load change parameters characterizing the degree of load change based on this sequence. These parameters are then used for subsequent time characteristic analysis and load forecasting calculations.

[0031] In a preferred embodiment, S3 includes:

[0032] S31. Construct an encoding structure for load time series feature extraction, including an input mapping unit and an attention calculation module set thereafter. Introduce a sparse attention mechanism in the attention calculation module, construct a decoding structure connected to the encoding structure, and introduce a time-series embedding module for switching between holidays and weekdays into the encoding and decoding structures. Fuse time-related features with load features to construct a load prediction model.

[0033] S32. The load forecasting model is trained and tested using a multi-source time series dataset. The trained load forecasting model is then used as a test and evaluation model for the multi-source time series dataset. At the same time, the intermediate layer output of the equipment operation load forecasting model is used as a feature vector to identify feature information. The trained load forecasting model is then used as a data operation forecast.

[0034] In a preferred embodiment, S4 includes:

[0035] S41. Based on the instantaneous active power P, instantaneous reactive power Q, cumulative electricity E, temperature T and rainfall R of the power grid load in the multi-source time series dataset, obtain the holiday impact coefficient and the weekday switching impact coefficient in the following ways.

[0036] S42. Based on load data at adjacent time points, calculate the instantaneous active power change of the power grid load. Instantaneous reactive power change and the rate of change of accumulated electricity value ;

[0037] ;

[0038] ;

[0039] ;

[0040] In the three formulas, t represents the data acquisition time point. This represents the interval between adjacent data collection time points;

[0041] S43. Construct the load change characteristic value L;

[0042] ;

[0043] In the formula , and Represented as weighting coefficients;

[0044] S44, Constructing a temperature correction factor and rainfall correction factor ;

[0045] In the formula, Expressed as meteorological sensitivity coefficient, Represented as historical average temperature;

[0046] In the formula, Expressed as meteorological sensitivity coefficient, This is expressed as the historical average rainfall.

[0047] S45. Constructing the characteristic values ​​of load changes after meteorological correction. ;

[0048] ;

[0049] S46. Based on the time attribute, divide the time points into:

[0050] The set of holiday time points H;

[0051] The set of non-holiday time points N;

[0052] Calculate the average load change of the holiday time period set Average load variation at non-holiday time points ;

[0053] ;

[0054] ;

[0055] Finally, based on the average load change of the holiday time period set Average load variation at non-holiday time points The impact coefficient of holidays can be obtained using the following formula. ;

[0056] ;

[0057] S47. Preset the impact threshold Y of holidays and set the impact coefficient of holidays. The comparison was made with the impact threshold Y of holidays, including:

[0058] when When the value is greater than Y, it indicates that the power grid load fluctuation in the micro-balance zone is abnormal during holidays. It is necessary to enhance the adjustment of the holiday impact coefficient according to the impact intensity of holidays. The enhancement adjustment includes amplifying the holiday impact coefficient according to a preset ratio, where the amplification ratio is 10%-30%.

[0059] when When ≤Y, it indicates that the power grid load fluctuation in the micro-balance zone is normal during holidays.

[0060] In a preferred embodiment, S4 further includes:

[0061] S48. Determine the set of weekday switching times S and the set of non-switching times W within the week;

[0062] Based on the set S of weekday switching times, construct the mean of load change characteristics within the set of weekday switching times. Based on the set of non-switching time points W, the average load change characteristics within the set of non-working day switching time points are constructed. ;

[0063] ;

[0064] ;

[0065] Based on the average load change characteristics within the set of weekday switching time points Mean of load change characteristics within the set of non-working day switching time points The impact coefficient of weekday switching can be calculated using the following formula. ;

[0066] ;

[0067] S49. Preset the threshold K for the impact of weekday switching, and set the impact coefficient for weekday switching. A comparison was made with the impact threshold K of weekday switching, including:

[0068] when When the value is greater than K, it indicates that the weekday switching time factor has an abnormal impact on the load change of the micro-balance zone power grid, and the weekday switching impact coefficient needs to be increased by 31%-50%.

[0069] when When K is ≤K, it indicates that the impact of weekday switching time factors on the load changes of the micro-balance zone is normal.

[0070] In a preferred embodiment, S5 includes:

[0071] S51, Based on the impact coefficient of holidays Impact coefficient of switching between weekdays By correlating and weighting the data, the comprehensive time impact coefficient can be obtained using the following formula. ;

[0072] ;

[0073] In the formula, Represented as weighting coefficients;

[0074] S52. Preset the comprehensive time influence threshold M, and set the comprehensive time influence coefficient. Compare with the overall time impact threshold M, including:

[0075] when When the value is greater than M, it indicates that the combined effect of holiday time factors and weekday switching time factors has an abnormal impact on the load change of the power grid in the micro-balance area. The comprehensive time influence coefficient needs to be amplified and adjusted by 10%-30% to enhance the comprehensive time factor's ability to represent load changes.

[0076] when When M ≤ M, it indicates that the combined effect of holiday time factors and weekday switching time factors on the load change of the micro-balance zone is normal.

[0077] In a preferred embodiment, S6 includes:

[0078] S61, Incorporate the time impact coefficient Output is based on a unified time series and a comprehensive time influence coefficient. The magnitude of the value and its variation over time are used to analyze the time sensitivity, fluctuation degree and stability of the power grid load operation status in the micro-balance zone.

[0079] In a preferred embodiment, a weak power grid load forecasting system based on an improved Informer model using a time-series embedding mechanism includes:

[0080] The data acquisition module is used to collect the load measurement point operation parameters and meteorological operation parameters of the micro-balanced area of ​​the power grid in the weak power grid.

[0081] The data processing module is used to perform time alignment processing on the load measurement point operating parameters of the micro-balanced power grid and the meteorological operating parameters of the micro-balanced area to obtain a multi-source time series dataset.

[0082] The module for constructing load change parameters is used to obtain load change parameters that characterize the degree of load change based on multi-source time series datasets and by performing differential calculations on load time series.

[0083] The model building module is used for encoder-decoder structure. It introduces a time-series embedding module that switches between holidays and weekdays to build a load forecasting model. After time-location encoding of multi-source time-series datasets and load time series, it is input into the load forecasting model for analysis and outputs the load forecasting results of the micro-balanced power grid.

[0084] The calculation module continues to construct the impact coefficient of holidays based on multi-source time-series datasets. and the impact coefficient of weekday switching ;

[0085] The comprehensive impact module utilizes the impact coefficient of holidays. and the impact coefficient of weekday switching Correlation, constructing a comprehensive time influence coefficient And evaluate and correct;

[0086] The output module is used to calculate the overall time impact coefficient. Output

[0087] This invention provides a method and system for predicting power load in weak power grids based on an improved Informer model using a time-series embedding mechanism. Its beneficial effects include:

[0088] 1. By performing time location encoding on the load time series in multi-source time series datasets, and combining an encoder-decoder structure with a time series embedding module that switches between holidays and weekdays, a load forecasting model is constructed. This enables the model to explicitly retain the sequential relationship of load data in the time dimension when performing load forecasting, thereby enhancing the modeling ability of long-term load change patterns and improving the accuracy and stability of load forecasting for weak power grids. By performing differential calculations on the load time series, load change parameters characterizing the degree of load change are constructed, so that the load forecasting method not only focuses on the absolute value change of load, but also can characterize the magnitude and trend of load change. This is beneficial for identifying fluctuation characteristics such as rapid load rise or fall, and improving the ability to perceive abnormal load changes in weak power grids.

[0089] 2. Based on multi-source time-series datasets, holiday impact coefficients and weekday switching impact coefficients are constructed, and a comprehensive time impact coefficient is further formed. This enables a quantitative description of the impact of load changes under different time types, avoiding the uncertainty caused by relying solely on empirical rules or simple classification. The load forecasting process is separated from the calculation and evaluation process of the time impact coefficient. The load forecasting model focuses on predicting load results, while the time impact coefficient is used to analyze and explain the load change mechanism. This helps to reduce model complexity and improve the deployability and maintainability of the method in the actual operation of weak power grids. Attached Figure Description

[0090] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the embodiments will be briefly introduced below. It should be understood that the following drawings only show some embodiments of the present invention and should not be regarded as a limitation of the scope. For those skilled in the art, other related drawings can be obtained from these drawings without creative effort.

[0091] Figure 1 This is a flowchart of the method of the present invention;

[0092] Figure 2 This is a system block diagram of the present invention. Detailed Implementation

[0093] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0094] Example 1, referring to Figure 1 This invention provides a technical solution: a method for predicting power load in weak power grids based on an improved Informer model using a time-series embedding mechanism, comprising:

[0095] S1. For the micro-balanced power grid of the weak power grid, collect the load measurement point operation parameters of the micro-balanced power grid and the meteorological operation parameters of the micro-balanced power grid, and perform time alignment processing on the load measurement point operation parameters of the micro-balanced power grid and the meteorological operation parameters of the micro-balanced power grid to obtain a multi-source time series dataset.

[0096] S2. Based on multi-source time series datasets, perform differential calculations on the load time series to obtain load change parameters that characterize the degree of load change;

[0097] S3. Using an encoder-decoder structure, a time-series embedding module that switches between holidays and weekdays is introduced to construct a load forecasting model. After time-location encoding of the multi-source time-series dataset and the load time series, the model is input into the load forecasting model, and the load forecasting results of the micro-balanced power grid are output.

[0098] S4. Construct the impact coefficient of holidays based on multi-source time-series datasets. and the impact coefficient of weekday switching ;

[0099] S5. Based on the impact coefficient of holidays and the impact coefficient of weekday switching Weighted calculations are performed to construct a comprehensive time influence coefficient. And the overall time influence coefficient Conduct assessments and corrections;

[0100] S6. Impact coefficient on overall time Output the results.

[0101] In this embodiment, by performing time location encoding on the load time series in the multi-source time series dataset and constructing a load prediction model using an encoder-decoder structure, the model can explicitly retain the sequential relationship of load data in the time dimension when making load predictions. This enhances the modeling ability of long-term load change patterns and improves the accuracy and stability of load prediction for weak power grids. By performing differential calculations on the load time series, load change parameters characterizing the degree of load change are constructed. This allows the load prediction method to not only focus on the absolute value change of load but also to depict the magnitude and trend of load changes. This is beneficial for identifying fluctuation characteristics such as rapid load increases or decreases and improves the ability to perceive abnormal load changes in weak power grids.

[0102] Based on multi-source time-series datasets, holiday impact coefficients and weekday switching impact coefficients are constructed, and a comprehensive time impact coefficient is further formed. This enables a quantitative description of the degree of load change impact under different time types, avoiding the uncertainty caused by relying solely on empirical rules or simple classification. The load forecasting process is separated from the calculation and evaluation process of the time impact coefficient. The load forecasting model focuses on predicting load results, while the time impact coefficient is used to analyze and explain the load change mechanism. This helps to reduce model complexity and improve the deployability and maintainability of the method in the actual operation of weak power grids.

[0103] Example 2 is an explanation of Example 1; please refer to it. Figure 1 Specifically, S1 includes:

[0104] S11. The load measurement point operating parameters of the micro-balanced power grid include the instantaneous active power P of the power grid load, the instantaneous reactive power Q of the power grid load, and the cumulative energy value E.

[0105] S12. By setting up a data acquisition point at the outgoing end of the load measurement point in the micro-balance zone power grid, and installing current transformers and voltage transformers at the data acquisition point, connecting the current transformers and voltage transformers to the smart energy meter, the voltage and current are collected in real time through the energy meter, and the instantaneous active power P of the power grid load is calculated using the formula.

[0106] S13. Based on the current transformer and voltage transformer at the load measurement point outgoing terminal position in the micro-balanced power grid, the voltage and current are measured in real time, and the instantaneous reactive power Q of the power grid load is calculated by formula.

[0107] S14. Record the cumulative electrical energy value of the load measuring point in real time through the electricity meter to obtain the cumulative electrical energy value E of the power grid load;

[0108] S15. Perform time alignment processing on the instantaneous active power P, instantaneous reactive power Q, and cumulative energy value E at the load measurement points of the micro-balanced power grid, including:

[0109] A uniform sampling time interval is determined, a standard time series is constructed, the original time series of each parameter is interpolated and mapped onto the standard time series, outlier identification and removal are performed on the mapped time series, and the processed time series is standardized.

[0110] In this embodiment, by collecting the instantaneous active power, instantaneous reactive power, and cumulative energy value of the power grid load, a multi-dimensional load measurement point operation parameter system covering power state and energy change is constructed. This system can more comprehensively reflect the operating characteristics of the power grid load in the micro-balance zone, avoiding information loss caused by relying solely on a single load indicator. By setting up collection points at the load measurement point outgoing terminals and using current transformers and voltage transformers in conjunction with smart energy meters to collect voltage and current in real time, online calculation of instantaneous active power and instantaneous reactive power is achieved, effectively ensuring the accuracy and timeliness of the load measurement point operation parameter acquisition. By performing time alignment processing on the instantaneous active power, instantaneous reactive power, and cumulative energy value under a unified sampling time interval, the original data with different sampling frequencies or time offsets are mapped to a standard time series, avoiding modeling errors caused by inconsistencies in the time dimension of multi-source load data.

[0111] Example 3 is an explanation of Example 1; please refer to the provided text. Figure 1 Specifically, S1 also includes:

[0112] S16. The meteorological operating parameters of the micro-equilibrium zone include the temperature T and precipitation R of the micro-equilibrium zone;

[0113] S17. By selecting a fixed meteorological observation point within the micro-equilibrium zone and installing a temperature sensor at the meteorological observation point, the ambient temperature is measured in real time to obtain the temperature T of the micro-equilibrium zone.

[0114] S18. Rainfall in the micro-equilibrium zone is collected in real time by deploying rain gauges in the micro-equilibrium zone;

[0115] S19. Perform time resampling processing on the temperature T of the micro-equilibrium zone collected by the temperature sensor, map the temperature data to a standard time series, and when the temperature data of multiple meteorological observation points correspond to the same time point, perform statistical processing on the multiple temperature data to obtain the temperature data of the micro-equilibrium zone corresponding to the same time point.

[0116] The rainfall R collected by the rain gauge in the micro-equilibrium area is processed by time segmentation. The rainfall is accumulated or redistributed over time according to the standard time series so that each standard time period corresponds to a set of rainfall data. This yields the micro-equilibrium area rainfall data that is consistent with the standard time series, thereby constructing a multi-source time series dataset.

[0117] In this embodiment, by collecting temperature and rainfall data from the micro-balance zone as meteorological operating parameters, external environmental factors that can directly affect electricity consumption behavior and load changes are introduced. This allows the multi-source time-series dataset to not only contain the power grid's own operating information but also reflect the potential impact of meteorological condition changes on the load, improving the comprehensiveness of load forecasting and load change analysis. By selecting fixed meteorological observation points within the micro-balance zone and installing temperature sensors and rain gauges for real-time data collection, meteorological operating parameters of the micro-balance zone can be continuously and stably acquired, avoiding data fluctuations caused by inconsistent meteorological data sources or differences in collection methods, and improving the reliability of meteorological data.

[0118] When performing time resampling on temperature data, if the temperature data from multiple meteorological observation points correspond to the same time point, the overall temperature value of the micro-equilibrium area can be obtained by statistically processing the multiple temperature data. This effectively reduces the impact of local meteorological fluctuations on load analysis, making the temperature parameters more reflective of the overall meteorological characteristics of the micro-equilibrium area. By performing time segmentation processing on the rainfall data collected by the rain gauge and performing time accumulation or time redistribution according to the standard time series, the rainfall data and the operating parameters of the load measurement points are kept consistent in the time dimension, avoiding data misalignment caused by different sampling periods.

[0119] Example 4 is an explanation of Example 1; please refer to the provided text. Figure 1 Specifically, S2 includes:

[0120] S21. Extract the load time series corresponding to the load measurement points based on the multi-source time series dataset. The load time series includes load data arranged in a uniform time series.

[0121] S22. Based on the sampling time interval of the multi-source time series dataset, determine the differential time scale of the load time series so that the load changes between corresponding adjacent time points can be calculated differentially.

[0122] S23. Perform differential calculations on the load time series according to the differential time scale to obtain the load change between adjacent time points, where the load change is used to reflect the magnitude and direction of load change over time.

[0123] S24. Construct load change characteristics based on load change amounts, correlate these characteristics with corresponding time points to form a load change characteristic sequence, and construct load change parameters characterizing the degree of load change based on this sequence. These parameters are then used for subsequent time characteristic analysis and load forecasting calculations.

[0124] In this embodiment, the load time series corresponding to the load measurement points is extracted from the multi-source time series dataset, and the load data is arranged according to a unified time series to ensure the continuity and consistency of the load data in the time dimension. The differential time scale is determined according to the sampling time interval of the multi-source time series dataset, so that the differential calculation strictly corresponds to the load change between adjacent time points. It can reflect the load change process over time in a fine-grained manner and avoid the distortion of load change information due to time scale mismatch. By performing differential calculation on the load time series, the load change between adjacent time points is obtained, so that the magnitude and direction of load change are clearly quantified, which is helpful to distinguish different operating states such as load increase, load decrease and load stability.

[0125] By associating load changes with corresponding time points, a load change characteristic sequence is formed, so that the load change characteristics not only reflect numerical changes, but also retain their temporal sequence information.

[0126] Example 5 is an explanation of Example 1; please refer to it. Figure 1 Specifically, S3 includes:

[0127] S31. Construct an encoding structure for load time series feature extraction, including an input mapping unit and an attention calculation module set thereafter. Introduce a sparse attention mechanism in the attention calculation module, construct a decoding structure connected to the encoding structure, and introduce a time-series embedding module for switching between holidays and weekdays into the encoding and decoding structures. Fuse time-related features with load features to construct a load prediction model.

[0128] S32. The load forecasting model is trained and tested using a multi-source time series dataset. The trained load forecasting model is then used as a test and evaluation model for the multi-source time series dataset. At the same time, the intermediate layer output of the equipment operation load forecasting model is used as a feature vector to identify feature information. The trained load forecasting model is then used as a data operation forecast.

[0129] In this embodiment, an encoding structure for load time series feature extraction is constructed, and an input mapping unit is set in the encoding structure. This enables multi-source time series data to be uniformly mapped to the model feature space, reducing the differences in the dimensions and distribution of data from different sources and improving the stability and consistency of load feature extraction. A sparse attention mechanism is introduced into the attention calculation module in the encoding structure, enabling the model to focus on key time points that have a significant impact on load changes, reducing the ineffective modeling of redundant time information. This reduces the computational complexity of the model while ensuring prediction accuracy, making it suitable for long-term load forecasting tasks in weak power grid scenarios.

[0130] By constructing a decoding structure connected to the encoding structure, the model can fully learn the global temporal dependencies of historical load characteristics during the encoding stage and gradually predict future load change trends during the decoding stage. This improves the model's comprehensive modeling ability for long-term load change patterns and short-term fluctuation characteristics. By introducing a time feature embedding module into the encoding and decoding structures, time-related features are fused with load features, enabling the model to explicitly perceive time location, periodic changes, and time pattern differences. This enhances the model's ability to express load changes caused by time factors such as holidays and workday switching.

[0131] Example 6 is an explanation of Example 1; please refer to the provided text. Figure 1 Specifically, S4 includes:

[0132] S41. Based on the instantaneous active power P, instantaneous reactive power Q, cumulative electricity E, temperature T and rainfall R of the power grid load in the multi-source time series dataset, obtain the holiday impact coefficient and the weekday switching impact coefficient in the following ways.

[0133] S42. Based on load data at adjacent time points, calculate the instantaneous active power change of the power grid load. Instantaneous reactive power change and the rate of change of accumulated electricity value ;

[0134] ;

[0135] ;

[0136] ;

[0137] In the three formulas, t represents the data acquisition time point. This represents the interval between adjacent data collection time points;

[0138] S43. Construct the load change characteristic value L;

[0139] ;

[0140] In the formula , and This is represented as a weighting coefficient, and the weighting coefficient is analyzed based on historical data. , and The proportion of the impact on the characteristic value L of load change is obtained. , and The value;

[0141] S44, Constructing a temperature correction factor and rainfall correction factor ;

[0142] In the formula, This is expressed as a meteorological sensitivity coefficient. Through statistical or regression analysis of historical sample sets, the sensitivity of load changes to temperature and rainfall changes is obtained, and this sensitivity is used as the meteorological sensitivity coefficient in the temperature correction factor. It is represented as the historical average temperature, obtained by combining historical temperature statistics;

[0143] In the formula, This is expressed as a meteorological sensitivity coefficient. By performing statistical or regression analysis on the historical sample set, the sensitivity of load changes to temperature and rainfall changes is obtained, and these sensitivities are used as meteorological sensitivity coefficients in the temperature correction factor. It is expressed as the historical average rainfall, obtained by combining historical rainfall statistics;

[0144] S45. Constructing the characteristic values ​​of load changes after meteorological correction. ;

[0145] ;

[0146] S46. Based on the time attribute, divide the time points into:

[0147] The set of holiday time points H;

[0148] The set of non-holiday time points N;

[0149] Calculate the average load change of the holiday time period set Average load variation at non-holiday time points ;

[0150] ;

[0151] ;

[0152] Finally, based on the average load change of the holiday time period set Average load variation at non-holiday time points The impact coefficient of holidays can be obtained using the following formula. ;

[0153] ;

[0154] S47. Preset the impact threshold Y of holidays, extract holiday impact coefficient samples corresponding to historical holiday time points from multi-source time series datasets, perform statistical analysis on holiday impact coefficient samples, obtain statistical characteristics of historical holiday impact coefficients, and determine the holiday impact threshold Y based on statistical characteristics;

[0155] And the impact coefficient of holidays The comparison was made with the impact threshold Y of holidays, including:

[0156] when When the value is greater than Y, it indicates that the power grid load fluctuation in the micro-balance zone is abnormal during holidays. It is necessary to enhance the adjustment of the holiday impact coefficient according to the impact intensity of holidays. The enhancement adjustment includes amplifying the holiday impact coefficient according to a preset ratio, where the amplification ratio is 10%-30%.

[0157] when When ≤Y, it indicates that the power grid load fluctuation in the micro-balance zone is normal during holidays.

[0158] In this embodiment, load change characteristic values ​​are constructed based on instantaneous active power change, instantaneous reactive power change, and the rate of change of accumulated electricity, and a holiday impact coefficient is further formed. This transforms the qualitative analysis of the impact of holidays on the load changes of the micro-balanced power grid into a calculable and comparable quantitative indicator, improving the objectivity and repeatability of holiday load analysis. By weighting and fusing the active power change, reactive power change, and rate of change of electricity in the power grid load, a unified load change characteristic value is constructed, avoiding the bias caused by a single load indicator in the assessment of the impact of holidays. This allows the holiday load fluctuation characteristics to more comprehensively reflect the actual operating status of the micro-balanced power grid. By constructing temperature correction factors and rainfall correction factors and applying meteorological correction to the load change characteristic value, the interference of meteorological factors on load changes is effectively eliminated or weakened, enabling the holiday impact coefficient to more accurately reflect the effect of the holiday time attribute itself on load changes and improving the reliability of holiday impact identification.

[0159] By dividing the time points into holiday time point sets and non-holiday time point sets, calculating the average load change for each set, and then constructing the holiday impact coefficient in the form of a ratio, the impact of a single abnormal time point on the assessment results can be reduced, and the stability and robustness of the holiday load impact judgment results can be improved.

[0160] Example 7 is an explanation of Example 1; please refer to it. Figure 1 Specifically, S4 also includes:

[0161] S48. Determine the set of weekday switching times S and the set of non-switching times W within the week;

[0162] Based on the set S of weekday switching times, construct the mean of load change characteristics within the set of weekday switching times. Based on the set of non-switching time points W, the average load change characteristics within the set of non-working day switching time points are constructed. ;

[0163] ;

[0164] ;

[0165] Based on the average load change characteristics within the set of weekday switching time points Mean of load change characteristics within the set of non-working day switching time points The impact coefficient of weekday switching can be calculated using the following formula. ;

[0166] ;

[0167] S49. Preset the impact threshold K of weekday switching, extract load change feature samples from the set of historical weekday switching time points in the multi-source time series dataset, calculate the corresponding weekday switching impact coefficient, form a sample set of historical weekday switching impact coefficients, perform statistical analysis on the sample set of historical weekday switching impact coefficients, and determine the impact threshold K of weekday switching based on its statistical characteristics.

[0168] The impact coefficient of switching weekdays A comparison was made with the impact threshold K of weekday switching, including:

[0169] when When the value is greater than K, it indicates that the weekday switching time factor has an abnormal impact on the load change of the micro-balance zone power grid, and the weekday switching impact coefficient needs to be increased by 31%-50%.

[0170] when When K is ≤K, it indicates that the impact of weekday switching time factors on the load changes of the micro-balance zone is normal.

[0171] In this embodiment, by clearly distinguishing between the set of weekday switching time points and the set of non-switching time points, and constructing corresponding average load change characteristics for each, the impact of weekday switching time factors such as Sunday and Monday on the load change of the micro-balance zone is explicitly characterized, avoiding the weakening or masking of weekday periodic changes in the overall time series. By statistically analyzing the average load change characteristics within the set of weekday switching time points and the set of non-switching time points, and constructing the weekday switching impact coefficient in the form of a ratio, the impact of individual abnormal sampling points on the evaluation results is effectively reduced, improving the stability and robustness of the weekday switching impact assessment.

[0172] By comparing the weekday switching impact coefficient with a preset threshold, it is possible to quantitatively identify whether the weekday switching time factor has an abnormal impact on the load change of the micro-balance zone power grid, providing a reliable basis for identifying load mutations, surges or drops caused by changes in work rhythm.

[0173] Example 8 is an explanation of Example 1; please refer to it. Figure 1 Specifically, S5 includes:

[0174] S51, Based on the impact coefficient of holidays Impact coefficient of switching between weekdays By correlating and weighting the data, the comprehensive time impact coefficient can be obtained using the following formula. ;

[0175] ;

[0176] In the formula, This is represented as a weighting coefficient, and the weighting coefficient is analyzed based on historical data. Impact coefficient on overall time The influence ratio is used to obtain the weighting coefficient. value;

[0177] S52. Preset the comprehensive time impact threshold M. Based on the historical holiday time points and weekday switching time points in the multi-source time series data, construct a historical comprehensive time impact coefficient sample set according to the calculation method of the comprehensive time impact coefficient, perform statistical analysis on the historical comprehensive time impact coefficient sample set, and determine the comprehensive time impact threshold M based on its statistical characteristics.

[0178] The overall time influence coefficient will be combined Compare with the overall time impact threshold M, including:

[0179] when When the value is greater than M, it indicates that the combined effect of holiday time factors and weekday switching time factors has an abnormal impact on the load change of the power grid in the micro-balance area. The comprehensive time influence coefficient needs to be amplified and adjusted by 10%-30% to enhance the comprehensive time factor's ability to represent load changes.

[0180] when When M ≤ M, it indicates that the combined effect of holiday time factors and weekday switching time factors on the load change of the micro-balance zone is normal.

[0181] In this embodiment, a comprehensive time influence coefficient is constructed by weighting and fusing the impact coefficients of holidays and weekday switching. This allows the impact of multiple time factors on the load changes of the micro-balanced power grid to be uniformly expressed in the form of a single parameter, avoiding the complexity and redundancy caused by modeling different time factors separately. By setting weight coefficients, the contribution ratio of holiday time factors and weekday switching time factors in the comprehensive time influence coefficient can be adjusted. This allows for flexible configuration of the degree of influence of different time factors according to the operating characteristics of different micro-balanced power grids, improving the adaptability of time influence modeling.

[0182] By comparing the comprehensive time impact coefficient with a preset threshold, it is possible to identify whether there are abnormal load changes under the superposition of holiday time factors and weekday switching time factors, thus effectively characterizing the complex time impact effect that is difficult to reflect by a single time factor.

[0183] Example 9, this example is an explanation of Example 1, please refer to it. Figure 1 Specifically, S6 includes:

[0184] S61, Incorporate the time impact coefficient Output is based on a unified time series and a comprehensive time influence coefficient. The magnitude of the value and its variation over time are used to analyze the time sensitivity, fluctuation degree and stability of the power grid load operation status in the micro-balance zone.

[0185] In this embodiment, by outputting the comprehensive time influence coefficient according to a unified time series, the coefficient can be aligned and compared with the load measurement point operating parameters and meteorological operating parameters in the same time dimension. This facilitates the continuous tracking and analysis of the impact of time factors on load changes. Based on the changing characteristics of the comprehensive time influence coefficient over time, it is possible to analyze the degree of load response to time factors such as holidays and weekday switching in different time periods, thereby revealing the differences in the sensitivity of the power grid load operation in the micro-balance area to time factors. By analyzing the magnitude and fluctuation amplitude of the comprehensive time influence coefficient, the strength of load fluctuation in the micro-balance area under different time conditions can be indirectly reflected, transforming the degree of load fluctuation from qualitative judgment to quantitative analysis.

[0186] When the comprehensive time influence coefficient remains relatively stable over a continuous period, it indicates that the load operation status of the power grid in the micro-balance zone is relatively stable; when the comprehensive time influence coefficient fluctuates significantly, it reflects a decline in load operation stability, thus providing an objective basis for load operation status assessment.

[0187] Example 10: This example is an explanation of Example 1. Please refer to the provided text. Figure 2 Specifically, a weak power grid load forecasting system based on an improved Informer model using a time-series embedding mechanism includes:

[0188] The data acquisition module is used to collect the load measurement point operation parameters and meteorological operation parameters of the micro-balanced area of ​​the power grid in the weak power grid.

[0189] The data processing module is used to perform time alignment processing on the load measurement point operating parameters of the micro-balanced power grid and the meteorological operating parameters of the micro-balanced area to obtain a multi-source time series dataset.

[0190] The module for constructing load change parameters is used to obtain load change parameters that characterize the degree of load change based on multi-source time series datasets and by performing differential calculations on load time series.

[0191] The model building module is used for encoder-decoder structure to build load forecasting model. After time location encoding of multi-source time series datasets and load time series, it is input into the load forecasting model for analysis and outputs the load forecast results of micro-balanced power grid.

[0192] The calculation module continues to construct the impact coefficient of holidays based on multi-source time-series datasets. and the impact coefficient of weekday switching ;

[0193] The comprehensive impact module utilizes the impact coefficient of holidays. and the impact coefficient of weekday switching Correlation, constructing a comprehensive time influence coefficient And evaluate and correct;

[0194] The output module is used to calculate the overall time impact coefficient. Output.

[0195] In this embodiment, the data acquisition module collects the load measurement point operating parameters and meteorological operating parameters of the micro-balanced power grid in a unified manner, and performs time alignment processing through the data processing module to construct a multi-source time series dataset with a unified structure. This effectively solves the problem of inconsistent sampling periods and time misalignment between load data and meteorological data, providing high-quality basic data for subsequent load analysis and forecasting. The load change parameter construction module performs differential calculation on the load time series to extract load change parameters that reflect the magnitude and direction of load change, transforming the load change process from a raw numerical expression to a change feature expression, which is beneficial for capturing key change information such as load abrupt changes and rapid fluctuations.

[0196] The model building module employs an encoder-decoder architecture combined with a time-series embedding module for holiday and weekday switching. Multi-source time-series datasets and load time series are processed with time-location encoding before being input into the load forecasting model. This allows the model to simultaneously learn load numerical characteristics and time-location information, enhancing its ability to model long-term series dependencies and improving the accuracy and stability of load forecasting. The calculation module constructs holiday impact coefficients and weekday switching impact coefficients based on the multi-source time-series datasets, quantifying the impact of key time factors such as holiday and weekday switching on load changes. This avoids the problem of weakened impact caused by time factors only participating in modeling implicitly.

[0197] The comprehensive impact module correlates and weights the impact coefficients of holidays and weekday switching to construct a comprehensive time impact coefficient, which is then evaluated and corrected. This allows the superimposed impact of multiple time factors to be uniformly represented by a single parameter, improving the ability to analyze load changes under complex time structures. The output module outputs the comprehensive time impact coefficient in a time-series format, enabling it to be used to analyze the sensitivity, fluctuation, and stability of the power grid load operation status in micro-balance areas to changes in time factors, providing an intuitive and quantifiable time impact indicator for load operation status analysis.

[0198] The threshold is set to facilitate comparison. The size of the threshold depends on the amount of sample data and the number of bases set by those skilled in the art for each set of sample data; as long as it does not affect the ratio between the parameter and the quantized value, it is acceptable.

[0199] The above formulas are all derived from software simulation using a large amount of data and are selected to be close to the actual values. The coefficients in the formulas are set by those skilled in the art according to the actual situation. The above description is only a preferred embodiment of the present invention, but the protection scope of the present invention is not limited thereto. Any equivalent substitutions or changes made by those skilled in the art within the technical scope disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the protection scope of the present invention.

Claims

1. A method for predicting power load in weak power grids based on an improved Informer model using a time-series embedding mechanism, characterized in that, include: S1. For the micro-balanced power grid of the weak power grid, collect the load measurement point operation parameters of the micro-balanced power grid and the meteorological operation parameters of the micro-balanced power grid, and perform time alignment processing on the load measurement point operation parameters of the micro-balanced power grid and the meteorological operation parameters of the micro-balanced power grid to obtain a multi-source time series dataset. S2. Based on multi-source time series datasets, perform differential calculations on the load time series to obtain load change parameters that characterize the degree of load change; S3. Using an encoder-decoder structure and introducing a time-series embedding module that switches between holidays and weekdays, a load forecasting model is constructed. After time-location encoding of the multi-source time-series dataset and the load time series, the model is input into the load forecasting model, and the load forecasting results of the micro-balanced power grid are output. S4. Construct the impact coefficient of holidays based on multi-source time-series datasets. and the impact coefficient of switching between weekdays ; S5. Based on the impact coefficient of holidays and the impact coefficient of switching between weekdays Weighted calculations are performed to construct a comprehensive time influence coefficient. And the overall time influence coefficient Conduct assessments and corrections; S6. Impact coefficient on overall time Output the results.

2. The method for predicting power load in weak power grids based on an improved Informer model using a time-series embedding mechanism as described in claim 1, characterized in that, S1 includes: S11. The load measurement point operating parameters of the micro-balanced power grid include the instantaneous active power P of the power grid load, the instantaneous reactive power Q of the power grid load, and the cumulative energy value E. S12. By setting up a data acquisition point at the outgoing end of the load measurement point in the micro-balance zone power grid, and installing current transformers and voltage transformers at the data acquisition point, connecting the current transformers and voltage transformers to the smart energy meter, the voltage and current are collected in real time through the energy meter, and the instantaneous active power P of the power grid load is calculated using the formula. S13. Based on the current transformer and voltage transformer at the load measurement point outgoing terminal position in the micro-balanced power grid, the voltage and current are measured in real time, and the instantaneous reactive power Q of the power grid load is calculated by formula. S14. Record the cumulative electrical energy value of the load measuring point in real time through the electricity meter to obtain the cumulative electrical energy value E of the power grid load; S15. Perform time alignment processing on the instantaneous active power P, instantaneous reactive power Q, and cumulative energy value E at the load measurement points of the micro-balanced power grid, including: A uniform sampling time interval is determined, a standard time series is constructed, the original time series of each parameter is interpolated and mapped onto the standard time series, outlier identification and removal are performed on the mapped time series, and the processed time series is standardized.

3. The method for predicting power load in weak power grids based on an improved Informer model using a time-series embedding mechanism, as described in claim 2, is characterized in that... S1 also includes: S16. The meteorological operating parameters of the micro-equilibrium zone include the temperature T and precipitation R of the micro-equilibrium zone; S17. By selecting a fixed meteorological observation point within the micro-equilibrium zone and installing a temperature sensor at the meteorological observation point, the ambient temperature is measured in real time to obtain the temperature T of the micro-equilibrium zone. S18. Rainfall in the micro-equilibrium zone is collected in real time by deploying rain gauges in the micro-equilibrium zone; S19. Perform time resampling processing on the temperature T of the micro-equilibrium zone collected by the temperature sensor, map the temperature data to a standard time series, and when the temperature data of multiple meteorological observation points correspond to the same time point, perform statistical processing on the multiple temperature data to obtain the temperature data of the micro-equilibrium zone corresponding to the same time point. The rainfall R collected by the rain gauge in the micro-equilibrium area is processed by time segmentation. The rainfall is accumulated or redistributed over time according to the standard time series so that each standard time period corresponds to a set of rainfall data. This yields the micro-equilibrium area rainfall data that is consistent with the standard time series, thereby constructing a multi-source time series dataset.

4. The method for predicting power load in weak power grids based on an improved Informer model with a time-series embedding mechanism as described in claim 3, characterized in that, S2 includes: S21. Extract the load time series corresponding to the load measurement points based on the multi-source time series dataset. The load time series includes load data arranged in a uniform time series. S22. Determine the differential time scale of the load time series based on the sampling time interval of the multi-source time series dataset; S23. Perform differential calculations on the load time series according to the differential time scale to obtain the load change between adjacent time points; S24. Construct load change characteristics based on load change amount, associate load change characteristics with corresponding time points to form a load change characteristic sequence, and construct load change parameters that characterize the degree of load change based on the load change characteristic sequence.

5. The method for predicting power load in weak power grids based on an improved Informer model using a time-series embedding mechanism as described in claim 4, characterized in that... S3 includes: S31. Construct an encoding structure for load time series feature extraction, including an input mapping unit and an attention calculation module set thereafter. Introduce a sparse attention mechanism in the attention calculation module, construct a decoding structure connected to the encoding structure, and introduce a time-series embedding module for switching between holidays and weekdays into the encoding and decoding structures. Fuse time-related features with load features to construct a load prediction model. S32. The load forecasting model is trained and tested using a multi-source time series dataset. The trained load forecasting model is then used as a test and evaluation model for the multi-source time series dataset. At the same time, the intermediate layer output of the equipment operation load forecasting model is used as a feature vector to identify feature information. The trained load forecasting model is then used as a data operation forecast.

6. The method for predicting power load in weak power grids based on an improved Informer model using a time-series embedding mechanism, as described in claim 5, is characterized in that... S4 includes: S41. Based on the instantaneous active power P, instantaneous reactive power Q, cumulative electricity E, temperature T and rainfall R of the power grid load in the multi-source time series dataset, obtain the holiday impact coefficient and the weekday switching impact coefficient in the following ways. S42. Based on load data at adjacent time points, calculate the instantaneous active power change of the power grid load. Instantaneous reactive power change and the rate of change of accumulated electricity ; S43. Construct the load change characteristic value L; S44, Constructing a temperature correction factor and rainfall correction factor ; S45. Constructing the characteristic values ​​of load changes after meteorological correction. ; S46. Based on the time attribute, divide the time points into: The set of holiday time points H; The set of non-holiday time points N; Calculate the average load change of the holiday time period set Average load variation at non-holiday time points ; Finally, based on the average load change of the holiday time period set Average load variation at non-holiday time points The impact coefficient of holidays can be obtained using the following formula. ; ； S47. Preset the impact threshold Y of holidays and set the impact coefficient of holidays. The comparison was made with the impact threshold Y of holidays, including: when When the value is greater than Y, it indicates that the power grid load fluctuation in the micro-balance zone is abnormal during holidays. It is necessary to enhance the adjustment of the holiday impact coefficient according to the impact intensity of holidays. The enhancement adjustment includes amplifying the holiday impact coefficient according to a preset ratio, where the amplification ratio is 10%-30%. when When ≤Y, it indicates that the power grid load fluctuation in the micro-balance zone is normal during holidays.

7. A method for predicting power load in weak power grids based on an improved Informer model using a time-series embedding mechanism, as described in claim 6, is characterized in that... S4 also includes: S48. Determine the set of weekday switching times S and the set of non-switching times W within the week; Based on the set S of weekday switching times, construct the mean of load change characteristics within the set of weekday switching times. Based on the set of non-switching time points W, the average load change characteristics within the set of non-working day switching time points are constructed. ; Based on the average load change characteristics within the set of weekday switching time points Mean of load change characteristics within the set of non-working day switching time points The impact coefficient of weekday switching can be calculated using the following formula. ; ； S49. Preset the threshold K for the impact of weekday switching, and set the impact coefficient for weekday switching. A comparison was made with the impact threshold K of weekday switching, including: when When the value is greater than K, it indicates that the weekday switching time factor has an abnormal impact on the load change of the micro-balance zone power grid, and the weekday switching impact coefficient needs to be increased by 31%-50%. when When K is ≤K, it indicates that the impact of weekday switching time factors on the load changes of the micro-balance zone is normal.

8. A method for predicting power load in weak power grids based on an improved Informer model using a time-series embedding mechanism, as described in claim 7, is characterized in that... S5 includes: S51, Based on the impact coefficient of holidays Impact coefficient of switching between weekdays By correlating and weighting the data, the comprehensive time impact coefficient can be obtained using the following formula. ; ； In the formula, Represented as weighting coefficients; S52. Preset the comprehensive time influence threshold M, and set the comprehensive time influence coefficient. Compare with the overall time impact threshold M, including: when When the value is greater than M, it indicates that the combined effect of holiday time factors and weekday switching time factors has an abnormal impact on the load change of the power grid in the micro-balance area. The comprehensive time influence coefficient needs to be amplified and adjusted by 10%-30% to enhance the comprehensive time factor's ability to represent load changes. when When M ≤ M, it indicates that the combined effect of holiday time factors and weekday switching time factors on the load change of the micro-balance zone is normal.

9. A method for predicting power load in a weak power grid based on an improved Informer model using a time-series embedding mechanism, as described in claim 8, is characterized in that... S6 includes: S61, Incorporate the time impact coefficient Output is based on a unified time series and a comprehensive time influence coefficient. The magnitude of the value and its variation over time are used to analyze the time sensitivity, fluctuation degree and stability of the power grid load operation status in the micro-balance zone.

10. A power load forecasting system for weak power grids based on an improved Informer model with a time-series embedding mechanism, applied to the power load forecasting method for weak power grids based on an improved Informer model with a time-series embedding mechanism as described in any one of claims 1-9, characterized in that, include: The data acquisition module is used to collect the load measurement point operation parameters and meteorological operation parameters of the micro-balanced area of ​​the power grid in the weak power grid. The data processing module is used to perform time alignment processing on the load measurement point operating parameters of the micro-balanced power grid and the meteorological operating parameters of the micro-balanced area to obtain a multi-source time series dataset. The module for constructing load change parameters is used to obtain load change parameters that characterize the degree of load change based on multi-source time series datasets and by performing differential calculations on load time series. The model building module is used for encoder-decoder structure to build load forecasting model. After time location encoding of multi-source time series datasets and load time series, it is input into the load forecasting model for analysis and outputs the load forecast results of micro-balanced power grid. The calculation module continues to construct the impact coefficient of holidays based on multi-source time-series datasets. and the impact coefficient of switching between weekdays ; The comprehensive impact module utilizes the impact coefficient of holidays. and the impact coefficient of switching between weekdays Correlation, constructing a comprehensive time influence coefficient And evaluate and correct; The output module is used to calculate the overall time impact coefficient. Output.

Images