A load prediction method of size model cooperation

Abstract

The application discloses a load prediction method based on a size model, and the method comprises the following steps: extracting multi-dimensional target load data from collected historical original load data of a power grid; determining multi-dimensional features based on the multi-dimensional target load data; the multi-dimensional features comprise time sequence features corresponding to a time sequence characteristic dimension, fluctuation features corresponding to a fluctuation characteristic dimension, and regional features corresponding to a regional characteristic dimension; determining a load type of the power grid based on the multi-dimensional features; determining a target size model collaborative prediction strategy matched with the load type from a plurality of size model collaborative prediction strategies, and determining a current load prediction result of the power grid based on the target size model collaborative prediction strategy. The method can improve the flexibility of power load prediction.

Description

Technical Field

[0001] This application relates to the field of data analysis technology, and in particular to a load forecasting method that combines large and small models. Background Technology

[0002] The stable and efficient operation of power systems is a core element of energy supply security. Load forecasting, as a crucial foundation for power dispatching decisions and grid resource allocation, is gaining increasing value with the intelligent development of the power industry. Currently, the power industry has ever-increasing demands for the real-time performance and adaptability of load forecasting. Accurate load forecasting has become a key support for grid planning, energy consumption, and dispatch optimization, making the research and optimization of related forecasting methods an important direction for power system technology upgrades. However, existing power load forecasting methods generally have low forecasting flexibility.

[0003] Therefore, improving the flexibility of power load forecasting has become an urgent problem to be solved. Summary of the Invention

[0004] This application provides a load forecasting method that combines large and small models, which can improve the flexibility of power load forecasting.

[0005] In a first aspect, embodiments of this application provide a load forecasting method that combines large and small models, the method comprising:

[0006] Extract multi-dimensional target load data from the collected historical raw load data of the power grid;

[0007] Based on multi-dimensional target load data, multi-dimensional features are determined; the multi-dimensional features include time-series features corresponding to the time-series characteristic dimension, fluctuation features corresponding to the fluctuation characteristic dimension, and regional features corresponding to the regional characteristic dimension.

[0008] Based on multi-dimensional characteristics, determine the load type of the power grid;

[0009] From multiple size model collaborative forecasting strategies, a target size model collaborative forecasting strategy that matches the load type is determined, and based on the target size model collaborative forecasting strategy, the current load forecast result of the power grid is determined.

[0010] In one embodiment, the multi-dimensional target load data includes load power data and regional load distribution data. Based on the multi-dimensional target load data, multi-dimensional features are determined, including: determining the time-series trend fitting degree based on the variation period and trend fitting degree of the load power data in different time dimensions, and using the time-series trend fitting degree as the time-series feature corresponding to the time-series characteristic dimension; determining the load volatility based on the change rate and numerical fluctuation range of the load power data within a preset time interval, and using the load volatility as the fluctuation feature corresponding to the fluctuation characteristic dimension; determining the load correlation between each power supply area corresponding to the power grid and the load data per unit area of ​​each power supply area based on the regional load distribution data, and determining the regional load correlation based on the load correlation and load data, and using the regional load correlation as the regional feature corresponding to the regional characteristic dimension; using the time-series feature, fluctuation feature, and regional feature as multi-dimensional features.

[0011] In one embodiment, the load type of the power grid is determined based on multi-dimensional features, including: determining the load type of the power grid as a stable load when the time-series trend fitting degree corresponding to the time-series feature is greater than or equal to a preset time-series trend fitting degree threshold, the load volatility corresponding to the fluctuation feature is less than or equal to a preset load volatility threshold, and the regional load correlation corresponding to the regional feature is greater than or equal to a preset regional correlation threshold; determining the load type of the power grid as a fluctuating load when the time-series trend fitting degree corresponding to the time-series feature is within a preset time-series trend fitting degree range, the load volatility corresponding to the fluctuation feature is within a preset load volatility range, and the regional load correlation corresponding to the regional feature is within a preset regional correlation range; and determining the load type of the power grid as a sudden load when the time-series trend fitting degree corresponding to the time-series feature is less than a preset time-series trend fitting degree threshold, the load volatility corresponding to the fluctuation feature is greater than a preset load volatility threshold, and the regional load correlation corresponding to the regional feature is less than a preset regional correlation threshold.

[0012] In one embodiment, the load type is a stable load. Based on a target-size model collaborative prediction strategy, the current load prediction result of the power grid is determined, including: inputting multi-dimensional target load data and the corresponding time-series characteristics into a small model to obtain an initial load prediction result, and determining the predicted load characteristics corresponding to the initial load prediction result; the small model is a lightweight time-series prediction model trained based on historical load time-series data of the target area in the power grid; using a large model, the predicted load characteristics are verified based on the actual load characteristics corresponding to the multi-dimensional target load data to obtain a verification result; the large model is a power time-series prediction model trained based on a training set including multi-source power data; the large model has more model parameters than the small model; when the verification result indicates that the difference between the actual load characteristics and the predicted load characteristics is greater than a preset difference threshold, the model parameters of the small model are adjusted using the large model to obtain an adjusted small model; the multi-dimensional target load data and the corresponding time-series characteristics are input into the adjusted small model to obtain the current load prediction result of the power grid.

[0013] In one implementation, the load type is a sudden load. Based on the target size model collaborative prediction strategy, the current load prediction result of the power grid is determined, including: inputting multi-dimensional target load data into a small model to obtain an initial load prediction result; the small model is a lightweight time-series prediction model trained based on historical load time-series data of the target area in the power grid; using a large model, the initial load prediction result is corrected based on the sudden load data associated with the historical sudden loads of the power grid to obtain the current load prediction result of the power grid; the large model is a power time-series prediction model trained based on a training set including multi-source power data; the large model has more model parameters than the small model.

[0014] In one implementation, the load type is fluctuating load. Based on a target-size model collaborative prediction strategy, the current load prediction result of the power grid is determined, including: inputting multi-dimensional target load data into a large model to obtain an initial load prediction result; the large model is a power time series prediction model trained based on a training set including multi-source power data; using a small model, extracting fluctuation characteristics corresponding to the fluctuation characteristics from the currently collected actual load data; the small model is a lightweight time series prediction model trained based on historical load time series data of the target area in the power grid; the small model has fewer model parameters than the small model; feeding the fluctuation characteristics back to the large model so that the large model can correct the initial load prediction result based on the fluctuation characteristics to obtain the current load prediction result of the power grid.

[0015] In one embodiment, the method further includes: dynamically monitoring each dimension of the multi-dimensional features and determining the change magnitude of each dimension based on the monitoring results; collecting multiple sets of historical raw load data at preset time intervals and determining the comprehensive load characteristics corresponding to each set of historical raw load data; determining the load characteristic evolution trend index based on multiple comprehensive load characteristics using a preset load characteristic evolution trend calculation model, and determining the corresponding trend judgment result based on the load characteristic evolution trend index; generating load characteristic evolution analysis data based on the change magnitude of each dimension feature, the load characteristic evolution trend index, and the trend judgment result; and, when it is determined based on the load characteristic evolution analysis data that the current load type has changed, switching the size model coordinated prediction strategy from a first size model coordinated prediction strategy that matches the current load type to a second size model coordinated prediction strategy that matches the changed load type, and determining the new load prediction result of the power grid based on the second size model coordinated prediction strategy.

[0016] Secondly, this application provides a load forecasting device that combines large and small models, the device comprising:

[0017] The data extraction module is used to extract multi-dimensional target load data from the collected historical raw load data of the power grid;

[0018] The feature determination module is used to determine multi-dimensional features based on multi-dimensional target load data. The multi-dimensional features include time-series features corresponding to the time-series characteristic dimension, fluctuation features corresponding to the fluctuation characteristic dimension, and regional features corresponding to the regional characteristic dimension.

[0019] The load type determination module is used to determine the load type of the power grid based on the characteristics of each dimension in the multi-dimensional features.

[0020] The load forecasting module is used to determine the target size model collaborative forecasting strategy that matches the load type from multiple size model collaborative forecasting strategies, and to determine the current load forecasting result of the power grid based on the target size model collaborative forecasting strategy.

[0021] Thirdly, this application provides a computer device, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the steps in the method provided in the first aspect.

[0022] Fourthly, this application also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the method provided in the first aspect.

[0023] Fifthly, this application also provides a computer program product, including a computer program that, when executed by a processor, implements the steps in the method provided in the first aspect.

[0024] The aforementioned load forecasting method using a combination of large and small models extracts multi-dimensional target load data from historical raw load data of the power grid. Based on this multi-dimensional target load data, it determines multi-dimensional features, including time-series features corresponding to the time-series characteristic dimension, fluctuation features corresponding to the fluctuation characteristic dimension, and regional features corresponding to the regional characteristic dimension. Based on each dimension of these multi-dimensional features, it determines the load type of the power grid. From multiple large and small model collaborative forecasting strategies, it determines a target-size model collaborative forecasting strategy that matches the load type, and based on this strategy, it determines the current load forecasting result for the power grid. This method, by determining multi-dimensional features from historical load data of the power grid, identifying the load type based on these features, and determining the current load forecasting result based on a target-size model collaborative forecasting strategy that matches the load type, allows for precise matching of the technical characteristics of different large and small model collaborative forecasting strategies with the forecasting scenarios of various loads. This enables a targeted combination of model capabilities and load forecasting needs, allowing the forecasting process for various loads to be completed using suitable model combinations, thereby improving the flexibility of power load forecasting. Attached Figure Description

[0025] To more clearly illustrate the technical solutions in the embodiments of this application or related technologies, the drawings used in the description of the embodiments of this application or related technologies will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort.

[0026] Figure 1 This is a flowchart illustrating a load forecasting method that combines large and small models, as provided in an embodiment of this application.

[0027] Figure 2 This is a flowchart illustrating another load forecasting method using a combination of large and small models provided in this application embodiment;

[0028] Figure 3 This is a schematic diagram of the structure of a load forecasting device that combines large and small models, as provided in an embodiment of this application.

[0029] Figure 4 This is a schematic diagram of the structure of a computer device provided in an embodiment of this application. Detailed Implementation

[0030] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0031] Please see Figure 1 , Figure 1 This is a flowchart illustrating a load forecasting method using a combination of large and small models provided in an embodiment of this application. This method can be executed by a computer device. Optionally, the computer device can be a terminal device or a server. The terminal devices mentioned herein can include, but are not limited to, various personal computers, laptops, smartphones, tablets, IoT devices, and portable wearable devices. IoT devices can include smart TVs, smart air conditioners, smart in-vehicle devices, projection devices, etc., and portable wearable devices can include smartwatches, smart bracelets, head-mounted devices, etc. The server mentioned herein can be an independent physical server, a server cluster or distributed system composed of multiple physical servers, or a cloud server providing cloud computing services, etc., without limitation. Figure 1 As shown, this load forecasting method combining large and small models may include, but is not limited to, the following steps:

[0032] S101. Extract multi-dimensional target load data from the collected historical raw load data of the power grid.

[0033] The historical load data of the power grid includes, but is not limited to, historical load data collected by power data acquisition terminals deployed at the dedicated transformers of various production enterprises within the power grid, at the power grid inlet and outlet points, and at the distribution terminals of core production equipment within the power grid. Optionally, each power data acquisition terminal may collect historical load data at 30-second intervals.

[0034] In one optional implementation, before step S101, the computer device can acquire historical load data collected by each power data acquisition terminal; and integrate multiple sets of historical load data to obtain the historical raw load data of the power grid.

[0035] In some embodiments, historical raw load data may include, but is not limited to, load time series data, regional load data, etc.

[0036] In some embodiments, multi-dimensional target load data may include, but is not limited to, load power data, load change rate data, and regional load distribution data.

[0037] S102. Based on multi-dimensional target load data, determine multi-dimensional features; multi-dimensional features include time-series features corresponding to the time-series characteristic dimension, fluctuation features corresponding to the fluctuation characteristic dimension, and regional features corresponding to the regional characteristic dimension.

[0038] In one optional implementation, the computer device determines multi-dimensional features based on multi-dimensional target load data, which may involve: performing data cleaning and normalization on at least one of the multi-dimensional target load data to obtain processed multi-dimensional target load data; and determining multi-dimensional features based on the processed multi-dimensional target load data; wherein, data cleaning may include abnormal data removal and missing data completion.

[0039] S103. Determine the load type of the power grid based on multi-dimensional characteristics.

[0040] In some embodiments, the computer device pre-stores multiple load types and corresponding judgment conditions for each load type. After determining the multi-dimensional characteristics, the computer device can match the multi-dimensional characteristics with the judgment conditions corresponding to each load type, and take the load type corresponding to the successfully matched judgment conditions as the load type of the power grid.

[0041] For example, suppose there are multiple load types, including stable loads, burst loads, and fluctuating loads; the judgment condition for stable loads is judgment condition 1, the judgment condition for burst loads is judgment condition 2, and the judgment condition for fluctuating loads is judgment condition 3. After determining the multi-dimensional characteristics, the computer equipment can match the multi-dimensional characteristics with judgment conditions 1, 2, and 3 respectively. If the multi-dimensional characteristics match judgment condition 1, the load type of the power grid can be determined to be a stable load; if the multi-dimensional characteristics match judgment condition 2, the load type of the power grid can be determined to be a burst load; and if the multi-dimensional characteristics match judgment condition 3, the load type of the power grid can be determined to be a fluctuating load.

[0042] S104. From multiple size model collaborative prediction strategies, determine the target size model collaborative prediction strategy that matches the load type, and based on the target size model collaborative prediction strategy, determine the current load prediction result of the power grid.

[0043] In some embodiments, the computer device determines a target size model collaborative forecasting strategy that matches the load type from multiple size model collaborative forecasting strategies. This can be done by: determining a target size model collaborative forecasting strategy that matches the load type from multiple size model collaborative forecasting strategies based on a correspondence relationship; wherein the correspondence relationship includes the correspondence relationship between multiple load types and multiple size model collaborative forecasting strategies.

[0044] Optionally, the correspondence can be a table stored locally on the computer device (denoted as the correspondence table), or a table stored in a database that is accessible to the computer device (denoted as the correspondence table), etc., without limitation here. The correspondence table includes the correspondence between various load types and collaborative forecasting strategies for various model sizes.

[0045] In this embodiment, multi-dimensional features are determined from historical load data of the power grid, and the load type of the power grid is determined based on these features. The current load forecast result of the power grid is determined based on a target-size model collaborative forecasting strategy that matches the load type. This allows the technical characteristics of collaborative forecasting strategies of different sizes of models to be accurately matched with the forecasting scenarios of various loads. As a result, the model capabilities and load forecasting needs can be combined in a targeted manner, and the forecasting process of various loads can be completed by relying on the appropriate model combination. In this way, the flexibility of power load forecasting can be improved.

[0046] In one alternative implementation, Figure 1 In step S102 of the load forecasting method with coordinated large and small models, the multi-dimensional target load data includes load power data and regional load distribution data. The computer equipment determines multi-dimensional features based on the multi-dimensional target load data in the following ways: determining the time-series trend fitting degree based on the variation cycle and trend fitting degree of the load power data in different time dimensions, and using the time-series trend fitting degree as the time-series feature corresponding to the time-series characteristic dimension; determining the load volatility based on the rate of change and numerical fluctuation range of the load power data within a preset time interval, and using the load volatility as the fluctuation feature corresponding to the fluctuation characteristic dimension; determining the load correlation between each power supply area corresponding to the power grid and the load data per unit area of ​​each power supply area based on the regional load distribution data, and determining the regional load correlation based on the load correlation and load data, and using the regional load correlation as the regional feature corresponding to the regional characteristic dimension; using the time-series feature, fluctuation feature, and regional feature as multi-dimensional features.

[0047] Trend fit refers to the degree to which load power data fits the baseline trend line of a single time dimension. It is the basic data for a single time and a single dimension.

[0048] In some embodiments, the computer device determines the time series trend fitting degree based on the fitting degree of the variation period and trend of the load power data in different time dimensions. This can be achieved by: performing data cleaning and normalization on the load power data to obtain the target load power data; and determining the time series trend fitting degree based on the fitting degree of the variation period and trend of the target load power data in different time dimensions.

[0049] Optionally, the computer equipment performs data cleaning and normalization on the load power data to obtain the target load power data. This can be achieved by: removing outliers and filling in missing data in the load power data to obtain processed load power data; and then normalizing the processed load power data to obtain the target load power data. In this way, by removing outliers and filling in missing data, a continuous and uninterrupted load power data sequence can be obtained. By normalizing the load power data sequence, the dimensional differences between data from different production zones can be eliminated, thereby improving the accuracy and completeness of the data, or in other words, improving data quality, thus providing a high-quality data foundation for subsequent determination of time-series characteristics.

[0050] Optionally, abnormal data may be data generated due to at least one of the following: power supply failure of the power data acquisition terminal, electromagnetic interference of production equipment, or data transmission interruption.

[0051] Optionally, missing data filling can be performed by using computer equipment to fill in the missing data using interpolation.

[0052] In some embodiments, the computer device determines the load fluctuation rate based on the rate of change and numerical fluctuation range of load power data within a preset time interval. This can be achieved by: performing data cleaning and normalization on the load power data to obtain target load power data; and determining the load fluctuation rate based on the rate of change and numerical fluctuation range of the target load power data within the preset time interval.

[0053] Optionally, the computer equipment can perform data cleaning and normalization on the load power data to obtain the target load power data. The relevant process can be found in the previous description and will not be repeated here.

[0054] Optionally, the rate of change of the target load power data within a preset time interval can be determined by computer equipment in the following way: acquiring the endpoint target load power data, the starting point target load power data, and the interval duration of the preset time interval within the preset time interval; determining the difference between the endpoint target load power data and the starting point target load power data; and using the quotient of the difference and the interval duration as the rate of change of the target load power data within the preset time interval.

[0055] Optionally, the rate of change of the target load power data within a preset time interval can also be determined by computer equipment in the following ways: smoothing the target load power data within the preset time interval; differentiating the processed target load power data within the preset time interval to obtain the instantaneous rate at each moment within the preset time interval; and determining the rate of change of the target load power data within the preset time interval based on the instantaneous rate at each moment. Specifically, the computer equipment determining the rate of change of the target load power data within the preset time interval based on the instantaneous rate at each moment can be done by using the average instantaneous rate at each moment as the rate of change of the target load power data within the preset time interval, or by using the endpoint values ​​of the instantaneous rates at each moment (such as the maximum instantaneous rate or the minimum instantaneous rate), etc., without limitation here.

[0056] In some embodiments, the computer device determines the load correlation between each power supply area corresponding to the power grid and the load data per unit area of ​​each power supply area based on the regional load distribution data. This can be achieved by: performing data cleaning and normalization on the regional load distribution data to obtain the target regional load distribution data; and determining the load correlation between each power supply area corresponding to the power grid and the load data per unit area of ​​each power supply area based on the target regional load distribution data.

[0057] Load correlation is used to measure the degree of interdependence and similarity of load behavior between different power supply zones in terms of time, space, or logic. Optionally, load correlation can be determined by computer equipment using the correlation coefficient method based on time series data. Determining load correlation using the correlation coefficient method based on time series data can involve: determining the similarity of load curves between two power supply zones within the same time period; and using this similarity as the load correlation between the two power supply zones.

[0058] Load data per unit area, also known as load density, describes the concentration of electricity consumption in a power supply area. A higher value per unit area indicates greater power consumption per square kilometer in that power supply area, and vice versa. Optionally, the load data per unit area can be determined by computer equipment in the following way: obtaining the total load and area of ​​the power supply area; and using the quotient of the total load and area as the load data per unit area.

[0059] Optionally, the process of using computer equipment to clean and normalize regional load distribution data to obtain target regional load distribution data can be referred to the aforementioned process of using computer equipment to clean and normalize load power data to obtain target load power data, and will not be repeated here.

[0060] This implementation method can accurately determine multi-dimensional characteristics, thereby providing a data foundation for subsequent determination of the load type of the power grid.

[0061] In one alternative implementation, Figure 1 Step S103 of the load forecasting method using the combined large and small models, i.e., the way the computer equipment determines the load type of the power grid based on the multi-dimensional features, can be as follows: If the time-series trend fitting degree corresponding to the time-series feature is greater than or equal to a preset time-series trend fitting degree threshold, the load volatility corresponding to the fluctuation feature is less than or equal to a preset load volatility threshold, and the regional load correlation corresponding to the regional feature is greater than or equal to a preset regional correlation threshold, the load type of the power grid is determined to be a stable load. If the time-series trend fitting degree corresponding to the time-series feature is within a preset time-series trend fitting degree range, the load volatility corresponding to the fluctuation feature is within a preset load volatility range, and the regional load correlation corresponding to the regional feature is within a preset regional correlation range, the load type of the power grid is determined to be a fluctuating load. If the time-series trend fitting degree corresponding to the time-series feature is less than a preset time-series trend fitting degree threshold, the load volatility corresponding to the fluctuation feature is greater than a preset load volatility threshold, and the regional load correlation corresponding to the regional feature is less than a preset regional correlation threshold, the load type of the power grid is determined to be a sudden load.

[0062] In some embodiments, at least one of the preset time-series trend fit threshold, preset load volatility threshold, and preset regional correlation threshold may be determined based on expert experience, based on the results of multiple experiments, or defined manually, etc., without limitation here.

[0063] In some embodiments, at least one of the preset time-series trend fit interval, preset load volatility interval, and preset regional correlation interval may be determined based on expert experience, or based on the results of multiple experiments, or may be manually defined, etc., without limitation here.

[0064] Using this implementation method, the load type of the power grid can be quickly and accurately determined based on multi-dimensional characteristics.

[0065] In one alternative implementation, Figure 1In step S104 of the load forecasting method using a combined large and small model, the load type is a stable load. The computer equipment determines the current load forecast result of the power grid based on the target large and small model combined forecasting strategy in the following ways: inputting multi-dimensional target load data and the corresponding time-series characteristics into the small model to obtain the initial load forecast result, and determining the predicted load characteristics corresponding to the initial load forecast result; the small model is a lightweight time-series forecasting model trained based on historical load time-series data of the target area in the power grid; using the large model, the predicted load characteristics are verified based on the actual load characteristics corresponding to the multi-dimensional target load data to obtain the verification result; the large model is a power time-series forecasting model trained based on a training set including multi-source power data; the large model has more model parameters than the small model; when the verification result indicates that the difference between the actual load characteristics and the predicted load characteristics is greater than a preset difference threshold, the model parameters of the small model are adjusted using the large model to obtain the adjusted small model; the multi-dimensional target load data and the corresponding time-series characteristics are input into the adjusted small model to obtain the current load forecast result of the power grid.

[0066] Among them, load characteristics refer to the pattern, form and statistical features of load changes over time.

[0067] Among them, the small model can be built based on long short-term memory network or lightweight Transformer architecture. The small model has the technical characteristics of low latency and fast inference, and can quickly capture the stable load change pattern during production periods. The large model can be a large power time series prediction model based on Transformer architecture. The large model has the technical characteristics of long time series dependency capture, multi-source feature fusion and cross-scenario generalized inference, and can accurately handle the complex change characteristics of sudden load.

[0068] The training set used to train the large model may include, but is not limited to, large-scale multi-source power data such as load time series data, meteorological data, electricity price data, and power grid topology data. The training tasks corresponding to the large model may include, but are not limited to, masked time series prediction, load type classification, and cross-regional feature alignment. After completing the above training tasks, the computer equipment can also fine-tune the model parameters for the load prediction and pre-characteristic matching task using the load characteristic data of the target area until the convergence condition is met, thus obtaining the large model.

[0069] In this implementation method, when the load type of the power grid is a stable load, a small model is used to carry out load forecasting, and a large model is used to verify the load characteristics of the small model throughout the forecasting process. If the verification finds that there is a mismatch between the load characteristics and the small model's prediction model, the model parameters of the small model are adjusted in a targeted manner through the large model, and the load forecast is completed based on the adjusted small model. In this way, the occurrence of forecasting errors can be effectively avoided, and the accuracy of the stable load forecasting results can be guaranteed as much as possible.

[0070] In one alternative implementation, Figure 1 In step S104 of the load forecasting method using a combined large and small model, the load type is a sudden load. The computer equipment determines the current load forecast result of the power grid based on the target large and small model combined forecasting strategy in the following ways: inputting multi-dimensional target load data into the small model to obtain the initial load forecast result; the small model is a lightweight time-series forecasting model trained based on historical load time-series data of the target area in the power grid; using the large model, the initial load forecast result is corrected based on the sudden load data associated with the historical sudden loads of the power grid to obtain the current load forecast result of the power grid; the large model is a power time-series forecasting model trained based on a training set including multi-source power data; the large model has more model parameters than the small model.

[0071] The descriptions of the small and large models can be found in the previous text and will not be repeated here.

[0072] In some embodiments, the computer device uses a large model to correct the initial load forecast result based on the burst load data associated with the historical burst loads of the power grid, thereby obtaining the current load forecast result of the power grid. This can be achieved by: determining a burst correction coefficient based on the burst load data associated with the historical burst loads of the power grid; determining a comprehensive load characteristic index corresponding to multi-dimensional features; and determining a benchmark comprehensive load characteristic index; and correcting the initial load forecast result based on the burst correction coefficient, the comprehensive load characteristic index, and the benchmark comprehensive load characteristic index, thereby obtaining the current load forecast result of the power grid.

[0073] Optionally, the computer equipment determines the sudden load correction coefficient based on the sudden load data associated with the historical sudden loads of the power grid. This can be achieved by: determining the correlation between the change range of the comprehensive load characteristic index corresponding to the sudden load and the change range of the actual load value based on the sudden load data associated with the historical sudden loads of the power grid; and fitting the sudden load correction coefficient based on the correlation.

[0074] Optionally, the multi-dimensional features may include time-series features, fluctuation features, and regional features. The computer equipment determines the comprehensive load characteristic index corresponding to the multi-dimensional features by: normalizing the time-series features, fluctuation features, and regional features respectively to obtain the processed time-series features, processed fluctuation features, and processed regional features; and weighting and fusing the processed time-series features, processed fluctuation features, and processed regional features to obtain the comprehensive load characteristic index. The comprehensive load characteristic index can also be called the comprehensive load feature. The process of the computer equipment determining the comprehensive load characteristic index can be expressed as the following formula (1).

[0075] (1)

[0076] In formula (1), F represents the comprehensive load characteristic index; F t This represents the processed temporal characteristics; F v This represents the fluctuation characteristics after processing; F r This represents the processed region characteristics; These represent the weights corresponding to the processed time-series features, the processed fluctuation features, and the processed regional features, respectively. It can be determined by fitting multiple regression analysis based on the characteristic contribution of each feature in the historical load operation data of the power grid to the load type determination.

[0077] Optionally, the benchmark comprehensive load characteristic index can be the comprehensive load characteristic index when the load is in a stable state.

[0078] Optionally, the computer equipment can correct the initial load forecast result based on the burst correction coefficient, the comprehensive load characteristic index and the benchmark comprehensive load characteristic index to obtain the current load forecast result of the power grid. This can be done by inputting the burst correction coefficient, the comprehensive load characteristic index, the benchmark comprehensive load characteristic index and the initial load forecast result into the calculation model for correcting the initial load forecast result as shown in the following formula (2) to obtain the current load forecast result of the power grid.

[0079] (2)

[0080] In formula (2), P c This represents the current load forecast result for the power grid; P p α represents the initial load forecast result; α represents the sudden correction coefficient; F represents the comprehensive load characteristic index, which can be determined by the aforementioned formula (1); F0 represents the baseline comprehensive load characteristic index.

[0081] In this implementation method, when the load type is a sudden load, load forecasting is carried out in parallel using a small model and a large model. The small model, with its low latency and fast inference, can quickly capture the sudden characteristics of the load and generate preliminary load forecast results based on these characteristics. The large model, by retrieving historical load data from the industrial park's power grid and combining it with historical data on sudden loads generated by the start-up and shutdown of similar core equipment, accurately corrects the preliminary load forecast results generated by the small model. This not only improves the speed of load forecasting but also enhances its accuracy.

[0082] In one alternative implementation, Figure 1 In step S104 of the load forecasting method using a combined large and small model, the load type is fluctuating load. The computer equipment determines the current load forecast result of the power grid based on the target large and small model collaborative forecasting strategy in the following ways: Multi-dimensional target load data is input into the large model to obtain the initial load forecast result; the large model is a power time-series forecasting model trained on a training set including multi-source power data; the small model extracts the fluctuation characteristics corresponding to the fluctuation features from the currently collected actual load data; the small model is a lightweight time-series forecasting model trained on historical load time-series data of the target area in the power grid; the small model has fewer model parameters than the large model; the fluctuation characteristics are fed back to the large model so that the large model can correct the initial load forecast result based on the fluctuation characteristics to obtain the current load forecast result of the power grid.

[0083] The relevant explanations of the large model and the small model can be found in the previous descriptions, and will not be repeated here.

[0084] Using this implementation method, when the load type is fluctuating load, the initial load forecast result can be determined using a large model, and the fluctuation characteristics corresponding to the fluctuation characteristics can be generated using a small model. Then, the fluctuation characteristics are fed back to the large model so that the large model can correct the initial forecast result based on the fluctuation characteristics. In this way, the accuracy of load forecasting under fluctuating load can be improved.

[0085] In one alternative implementation, Figure 1In the load forecasting method using a combination of large and small models, the computer equipment can dynamically monitor each dimension of the multi-dimensional characteristics and determine the magnitude of change of each dimension based on the monitoring results. Multiple sets of historical raw load data are collected at preset time intervals, and the comprehensive load characteristics corresponding to each set of historical raw load data are determined. Based on multiple comprehensive load characteristics, a preset load characteristic evolution trend calculation model is used to determine the load characteristic evolution trend index, and the corresponding trend judgment result is determined based on the load characteristic evolution trend index. Load characteristic evolution analysis data is generated based on the magnitude of change of each dimension of characteristics, the load characteristic evolution trend index, and the trend judgment result. When the load characteristic evolution analysis data indicates a change in the current load type, the large and small model coordinated forecasting strategy is switched from a first large and small model coordinated forecasting strategy matching the current load type to a second large and small model coordinated forecasting strategy matching the changed load type. Based on the second large and small model coordinated forecasting strategy, a new load forecast result for the power grid is determined.

[0086] In some embodiments, the preset time interval may be two minutes.

[0087] In some embodiments, the computer device determines the comprehensive load characteristics corresponding to each set of historical load data by: determining the multi-dimensional characteristics corresponding to each set of historical load data; and obtaining the comprehensive load characteristics corresponding to each set of historical load data from the multi-dimensional characteristics corresponding to each set of historical load data.

[0088] In some embodiments, the preset load characteristic evolution trend calculation model can be shown in the following formula (3).

[0089] (3)

[0090] In formula (3), T represents the load characteristic evolution trend index; n represents the preset number of time-series calculation nodes; F i This represents the overall load characteristics of the i-th time-series node; F i-1 λi represents the comprehensive load characteristics of the (i-1)th time-series node; λi represents the time weight of the ith time-series node.

[0091] In some embodiments, the computer device determines the corresponding trend judgment result based on the load characteristic evolution trend index, which may be as follows: when the load characteristic evolution trend index T is greater than or equal to a first preset trend index threshold and the load volatility is within a preset volatility range, the corresponding trend judgment result is determined to be that the load characteristics are evolving towards a fluctuating load; when T is greater than or equal to the first preset trend index threshold and the load volatility exceeds a preset volatility range, the corresponding trend judgment result is determined to be that the load characteristics are evolving towards a sudden load; when T is less than or equal to a second preset trend index threshold, the corresponding trend judgment result is determined to be that the load characteristics are evolving towards a stable load; when the absolute value of T is less than the first preset trend index threshold, the corresponding trend judgment result is determined to be that the load characteristics have no significant evolution.

[0092] Optionally, the first preset trend index threshold and the second preset trend index threshold can be obtained by statistical analysis of the distribution characteristics of the load characteristic evolution trend index in historical load type change events by computer equipment. For example, the first preset trend index threshold is 0.15, and the second preset trend index threshold is -0.15.

[0093] In some embodiments, the computer device generates load characteristic evolution analysis data based on the change amplitude of each dimension of characteristics, the load characteristic evolution trend index, and the trend determination result. This can be achieved by integrating the change amplitude of each dimension of characteristics, the load characteristic evolution trend index, and the trend determination result to obtain load characteristic evolution analysis data.

[0094] In some embodiments, the computer device may also determine whether the trend determination results included in the load characteristic evolution analysis data match the current load type. If yes, it determines that the current load type has not changed; otherwise, it determines that the current load type has changed.

[0095] For example, assuming the current load type is a stable load and the trend determination result indicates no significant change in load characteristics, the computer equipment can determine that the trend determination result matches the current load type. In this case, the computer equipment can determine that the current load type will not change. However, assuming the current load type is a stable load and the trend determination result indicates that the load characteristics are evolving towards a bursty load, the computer equipment can determine that the trend determination result does not match the current load type. In this case, the computer equipment can determine that the current load type is about to change, and the changed load type will be a bursty load.

[0096] Following the example above, assuming the computer equipment determines that the changed load type is a burst load, the big-small model co-prediction strategy can be switched to a big-small model co-prediction strategy that matches the burst load.

[0097] In some embodiments, the computer device may also maintain a first size model collaborative forecasting strategy that matches the current load type, and determine the current load forecast result of the power grid based on the first size model collaborative forecasting strategy, provided that the current load type does not change.

[0098] By adopting this implementation method, when it is determined that the current load type has changed, the large and small model collaborative prediction strategy is immediately switched to a collaborative prediction strategy that matches the changed load type. In this way, real-time linkage between load type change and prediction strategy switching can be achieved, thereby effectively avoiding prediction inaccuracy caused by load type change, and thus ensuring the real-time performance and accuracy of power grid load prediction.

[0099] The following is an overall description of the load forecasting method using a combination of large and small models provided in the embodiments of this application. Please refer to... Figure 2 , Figure 2 This is a flowchart illustrating another load forecasting method using a combination of large and small models provided in this application, as illustrated in the embodiments. Figure 2 As shown, this load forecasting method combining large and small models may include, but is not limited to, the following steps:

[0100] S201. Extract multi-dimensional target load data from the collected historical raw load data of the power grid; the multi-dimensional target load data includes load power data and regional load distribution data.

[0101] In an optional implementation, the relevant description of step S201 can be found in the description of step S101 above, and will not be repeated here.

[0102] S202. Perform data cleaning and normalization on the load power data and regional load distribution data respectively to obtain the processed load power data and processed regional load distribution data.

[0103] In one optional implementation, data cleaning may include outlier removal and missing data imputation. Optionally, missing data imputation may be performed by a computer device using interpolation to fill in the missing data.

[0104] S203. Based on the processed load power data and the processed regional load distribution data, determine multi-dimensional features; the multi-dimensional features include time-series features, fluctuation features, and regional features.

[0105] In one optional implementation, the computer device determines multi-dimensional features based on the processed load power data and the processed regional load distribution data. This can be achieved by: determining the time-series trend fitting degree based on the variation period and trend fitting degree of the processed load power data in different time dimensions, and using the time-series trend fitting degree as the time-series feature corresponding to the time-series characteristic dimension; determining the load volatility rate based on the change rate and numerical fluctuation range of the processed load power data within a preset time interval, and using the load volatility rate as the fluctuation feature corresponding to the fluctuation characteristic dimension; determining the load correlation degree between each power supply area corresponding to the power grid and the load data per unit area of ​​each power supply area based on the processed regional load distribution data, and determining the regional load correlation degree based on the load correlation degree and load data, and using the regional load correlation degree as the regional feature corresponding to the regional characteristic dimension; and using the time-series feature, fluctuation feature, and regional feature as multi-dimensional features.

[0106] S204. Based on multi-dimensional characteristics, determine the load type of the power grid; the load type includes any one of the following: stable load, fluctuating load, or sudden load.

[0107] In one optional implementation, the computer device determines the load type of the power grid based on multi-dimensional features. This can be achieved by: determining the power grid load type as a stable load when the time-series trend fitting degree corresponding to the time-series features is greater than or equal to a preset time-series trend fitting degree threshold, the load volatility corresponding to the fluctuation features is less than or equal to a preset load volatility threshold, and the regional load correlation corresponding to the regional features is greater than or equal to a preset regional correlation threshold; determining the power grid load type as a fluctuating load when the time-series trend fitting degree corresponding to the time-series features is within a preset time-series trend fitting degree range, the load volatility corresponding to the fluctuation features is within a preset load volatility range, and the regional load correlation corresponding to the regional features is within a preset regional correlation range; and determining the power grid load type as a sudden load when the time-series trend fitting degree corresponding to the time-series features is less than a preset time-series trend fitting degree threshold, the load volatility corresponding to the fluctuation features is greater than a preset load volatility threshold, and the regional load correlation corresponding to the regional features is less than a preset regional correlation threshold.

[0108] S205. From multiple size model collaborative prediction strategies, determine the target size model collaborative prediction strategy that matches the load type, and based on the target size model collaborative prediction strategy, determine the current load prediction result of the power grid.

[0109] S206. Dynamically monitor each dimension of the multi-dimensional features and determine the magnitude of change of each dimension based on the monitoring results.

[0110] S207. Collect multiple sets of historical raw load data at preset time intervals, and determine the comprehensive load characteristics corresponding to each set of historical raw load data.

[0111] S208. Based on multiple comprehensive load characteristics, a preset load characteristic evolution trend calculation model is adopted to determine the load characteristic evolution trend index, and based on the load characteristic evolution trend index, the corresponding trend judgment result is determined.

[0112] S209. Based on the change amplitude of each dimension's characteristics, the load characteristic evolution trend index, and the trend determination results, generate load characteristic evolution analysis data.

[0113] S210. Based on load characteristic evolution analysis data, if it is determined that the current load type has changed, the size model coordinated prediction strategy is switched from the first size model coordinated prediction strategy that matches the current load type to the second size model coordinated prediction strategy that matches the changed load type. Based on the second size model coordinated prediction strategy, the new load prediction result of the power grid is determined.

[0114] S211. Output load forecasting related information, which includes new load forecasting results, load types, and load characteristic evolution analysis data.

[0115] In some embodiments, the computer device outputs load forecasting information by outputting the load forecasting information to the power grid dispatching decision terminal, so that the power grid dispatching decision terminal displays the load forecasting information on the user interface.

[0116] In some embodiments, the computer equipment can also systematically integrate various data in load forecasting information according to production zones, load types, production periods, etc., to obtain an integrated data report of load forecasting and characteristic analysis; the integrated data report is then output to the power grid dispatching decision terminal. Using this embodiment, the target entity can formulate resource allocation plans for the park's power grid, load protection strategies for the start-up and shutdown of core equipment, and emergency control measures for sudden loads based on the integrated data report. This effectively ensures the safe and stable operation of the park's power grid and provides reliable power assurance for the orderly start-up and shutdown and efficient production of the park's production equipment.

[0117] In this embodiment, through the entire process of data preparation, characteristic classification, collaborative prediction, dynamic adaptation, and result output, it can not only accurately classify three types of loads—stable, sudden, and fluctuating—and execute differentiated large and small model division of labor and collaborative prediction strategies, but also automatically switch prediction strategies according to changes in load type. At the same time, relevant data and integrated data reports are synchronously output to the park power grid dispatching and decision-making terminal, which can provide accurate data support for park power grid resource allocation and sudden load protection. Thus, it can not only ensure the safe and stable operation of the park power grid, but also provide reliable power guarantee for the orderly start-up and shutdown and efficient production of park production equipment.

[0118] It should be understood that although the steps in the flowcharts of the embodiments described above are shown sequentially according to the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless explicitly stated herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least some steps in the flowcharts of the embodiments described above may include multiple steps or multiple stages. These steps or stages are not necessarily completed at the same time, but can be executed at different times. The execution order of these steps or stages is not necessarily sequential, but can be performed alternately or in turn with other steps or at least some of the steps or stages of other steps.

[0119] Based on the same inventive concept, this application also provides a load forecasting apparatus for implementing the load forecasting method involving both large and small models as described above. The solution provided by this apparatus is similar to the implementation described in the above method; therefore, the specific limitations in one or more embodiments of the load forecasting apparatus involving both large and small models provided below can be found in the limitations of the load forecasting method involving both large and small models described above, and will not be repeated here.

[0120] Please see Figure 3 , Figure 3 This is a schematic diagram of another load forecasting device with coordinated large and small models provided in an embodiment of this application. Figure 3 As shown, the load forecasting device that coordinates the size model may include, but is not limited to:

[0121] The data extraction module 301 is used to extract multi-dimensional target load data from the collected historical raw load data of the power grid;

[0122] The feature determination module 302 is used to determine multi-dimensional features based on multi-dimensional target load data; the multi-dimensional features include time-series features corresponding to the time-series characteristic dimension, fluctuation features corresponding to the fluctuation characteristic dimension, and regional features corresponding to the regional characteristic dimension.

[0123] The load type determination module 303 is used to determine the load type of the power grid based on the characteristics of each dimension in the multi-dimensional features.

[0124] The load forecasting module 304 is used to determine the target size model collaborative forecasting strategy that matches the load type from multiple size model collaborative forecasting strategies, and to determine the current load forecasting result of the power grid based on the target size model collaborative forecasting strategy.

[0125] In one embodiment, the multi-dimensional target load data includes load power data and regional load distribution data. When determining multi-dimensional features based on the multi-dimensional target load data, the feature determination module 302 specifically performs the following: Based on the variation period and trend fitting degree of the load power data in different time dimensions, determine the time-series trend fitting degree, and use the time-series trend fitting degree as the time-series feature corresponding to the time-series characteristic dimension; Based on the change rate and numerical fluctuation range of the load power data within a preset time interval, determine the load volatility, and use the load volatility as the fluctuation feature corresponding to the fluctuation characteristic dimension; Based on the regional load distribution data, determine the load correlation between each power supply area corresponding to the power grid and the load data per unit area of ​​each power supply area, and based on the load correlation and load data, determine the regional load correlation, and use the regional load correlation as the regional feature corresponding to the regional characteristic dimension; The time-series feature, fluctuation feature, and regional feature are used as multi-dimensional features.

[0126] In one embodiment, when the load type determination module 303 determines the load type of the power grid based on multi-dimensional features, it specifically determines the load type of the power grid as a stable load when the time-series trend fitting degree corresponding to the time-series feature is greater than or equal to a preset time-series trend fitting degree threshold, the load volatility corresponding to the fluctuation feature is less than or equal to a preset load volatility threshold, and the regional load correlation corresponding to the regional feature is greater than or equal to a preset regional correlation threshold; when the time-series trend fitting degree corresponding to the time-series feature is within a preset time-series trend fitting degree range, the load volatility corresponding to the fluctuation feature is within a preset load volatility range, and the regional load correlation corresponding to the regional feature is within a preset regional correlation range, the load type of the power grid is determined as a fluctuating load; when the time-series trend fitting degree corresponding to the time-series feature is less than a preset time-series trend fitting degree threshold, the load volatility corresponding to the fluctuation feature is greater than a preset load volatility threshold, and the regional load correlation corresponding to the regional feature is less than a preset regional correlation threshold, the load type of the power grid is determined as a sudden load.

[0127] In one embodiment, the load type is a stable load. When the load forecasting module 304 determines the current load forecast result of the power grid based on a target-size model collaborative forecasting strategy, it specifically performs the following steps: inputting multi-dimensional target load data and the corresponding time-series characteristics into a small model to obtain an initial load forecast result, and determining the predicted load characteristics corresponding to the initial load forecast result; the small model is a lightweight time-series forecasting model trained based on historical load time-series data of the target area in the power grid; using a large model, based on the actual load characteristics corresponding to the multi-dimensional target load data, the predicted load characteristics are verified to obtain a verification result; the large model is a power time-series forecasting model trained based on a training set including multi-source power data; the large model has more model parameters than the small model; when the verification result indicates that the difference between the actual load characteristics and the predicted load characteristics is greater than a preset difference threshold, the model parameters of the small model are adjusted using the large model to obtain an adjusted small model; the multi-dimensional target load data and the corresponding time-series characteristics are input into the adjusted small model to obtain the current load forecast result of the power grid.

[0128] In one embodiment, the load type is a sudden load; when the load forecasting module 304 determines the current load forecast result of the power grid based on the target size model collaborative forecasting strategy, it specifically performs the following steps: inputting multi-dimensional target load data into a small model to obtain an initial load forecast result; the small model is a lightweight time-series forecasting model trained based on historical load time-series data of the target area in the power grid; using a large model, the initial load forecast result is corrected based on the sudden load data associated with the historical sudden loads of the power grid to obtain the current load forecast result of the power grid; the large model is a power time-series forecasting model trained based on a training set including multi-source power data; the large model has more model parameters than the small model.

[0129] In one embodiment, the load type is a fluctuating load; when the load forecasting module 304 determines the current load forecast result of the power grid based on the target size model collaborative forecasting strategy, it specifically performs the following steps: inputting multi-dimensional target load data into a large model to obtain an initial load forecast result; the large model is a power time series forecasting model trained based on a training set including multi-source power data; using a small model, extracting the fluctuation characteristics corresponding to the fluctuation characteristics from the currently collected actual load data; the small model is a lightweight time series forecasting model trained based on historical load time series data of the target area in the power grid; the small model has fewer model parameters than the small model; feeding back the fluctuation characteristics to the large model so that the large model can correct the initial load forecast result based on the fluctuation characteristics to obtain the current load forecast result of the power grid.

[0130] In one embodiment, the method may further include a dynamic adaptation module, which is used to dynamically monitor each dimension of the multi-dimensional features and determine the change magnitude of each dimension based on the monitoring results; collect multiple sets of historical raw load data at preset time intervals and determine the comprehensive load characteristics corresponding to each set of historical raw load data; determine the load characteristic evolution trend index based on multiple comprehensive load characteristics using a preset load characteristic evolution trend calculation model, and determine the corresponding trend judgment result based on the load characteristic evolution trend index; generate load characteristic evolution analysis data based on the change magnitude of each dimension feature, the load characteristic evolution trend index, and the trend judgment result; and, when it is determined based on the load characteristic evolution analysis data that the current load type has changed, switch the size model coordinated prediction strategy from a first size model coordinated prediction strategy that matches the current load type to a second size model coordinated prediction strategy that matches the changed load type, and determine the new load prediction result of the power grid based on the second size model coordinated prediction strategy.

[0131] The various modules in the aforementioned load forecasting device that coordinates large and small models can be implemented entirely or partially through software, hardware, or a combination thereof. These modules can be embedded in the processor of the terminal device in hardware form or independent of it, or stored in the memory of the terminal device in software form, so that the processor can call and execute the corresponding operations of each module.

[0132] In one exemplary embodiment, a computer device is provided, which may be a terminal, and its internal structure diagram may be as follows: Figure 4As shown, the computer device includes a processor, memory, input / output interfaces, a communication interface, a display unit, and an input device. The processor, memory, and input / output interfaces are connected via a system bus, and the communication interface, display unit, and input device are also connected to the system bus via the input / output interfaces. The processor provides computational and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system and computer programs. The internal memory provides an environment for the operation of the operating system and computer programs stored in the non-volatile storage media. The input / output interfaces are used for exchanging information between the processor and external devices. The communication interface is used for wired or wireless communication with external terminals; wireless communication can be achieved through Wi-Fi, mobile cellular networks, Near Field Communication (NFC), or other technologies. When executed by the processor, the computer program implements a load forecasting method that combines large and small models. The display unit is used to form a visually visible image and can be a display screen, a projection device, or a virtual reality imaging device. The display screen can be an LCD screen or an e-ink screen. The input device of the computer device can be a touch layer covering the display screen, or buttons, trackballs, or touchpads set on the casing of the computer device, or external keyboards, touchpads, or mice, etc.

[0133] Those skilled in the art will understand that Figure 4 The structure shown is merely a block diagram of a portion of the structure related to the present application and does not constitute a limitation on the computer device to which the present application is applied. Specific computer devices may include more or fewer components than those shown in the figure, or combine certain components, or have different component arrangements.

[0134] In one exemplary embodiment, this application provides a computer device including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the steps in the above-described load forecasting method of size-model coordination.

[0135] In one exemplary embodiment, this application provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps in the above-described load forecasting method of size-model coordination.

[0136] In one exemplary embodiment, this application provides a computer program product including a computer program that, when executed by a processor, implements the steps in the above-described load forecasting method with coordinated size models.

[0137] It should be noted that the data involved in this application (including but not limited to acquired data, data used for analysis, and stored data) are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of the relevant data must comply with relevant regulations.

[0138] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium, and when executed, it can include the processes of the embodiments of the above methods. Any references to memory, databases, or other media used in the embodiments provided in this application can include at least one of non-volatile memory and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM can take many forms, such as Static Random Access Memory (SRAM) or Dynamic Random Access Memory (DRAM). The databases involved in the embodiments provided in this application may include at least one type of relational database and non-relational database. Non-relational databases may include, but are not limited to, blockchain-based distributed databases. The processors involved in the embodiments provided in this application may be general-purpose processors, central processing units, graphics processing units, digital signal processors, programmable logic devices, quantum computing-based data processing logic devices, artificial intelligence (AI) processors, etc., and are not limited to these.

[0139] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this application.

[0140] The embodiments described above are merely illustrative of several implementation methods of this application, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of this patent application. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of this application, and these all fall within the protection scope of this application. Therefore, the protection scope of this application should be determined by the appended claims.

Claims

1. A load forecasting method that combines large and small models, characterized in that, The method includes: Extract multi-dimensional target load data from the collected historical raw load data of the power grid; Based on the multi-dimensional target load data, multi-dimensional features are determined; the multi-dimensional features include time-series features corresponding to the time-series characteristic dimension, fluctuation features corresponding to the fluctuation characteristic dimension, and regional features corresponding to the regional characteristic dimension. Based on the aforementioned multi-dimensional characteristics, the load type of the power grid is determined; From multiple size model collaborative prediction strategies, a target size model collaborative prediction strategy that matches the load type is determined, and based on the target size model collaborative prediction strategy, the current load prediction result of the power grid is determined.

2. The method according to claim 1, characterized in that, The multi-dimensional target load data includes load power data and regional load distribution data; the determination of multi-dimensional features based on the multi-dimensional target load data includes: Based on the variation period and trend fitting degree of the load power data under different time dimensions, the time series trend fitting degree is determined, and the time series trend fitting degree is used as the time series feature corresponding to the time series characteristic dimension. Based on the rate of change and range of fluctuation of the load power data within a preset time interval, the load volatility is determined and the load volatility is used as the volatility feature corresponding to the volatility characteristic dimension. Based on the regional load distribution data, the load correlation between each power supply area corresponding to the power grid and the load data per unit area of ​​each power supply area are determined. Based on the load correlation and the load data, the regional load correlation is determined, and the regional load correlation is used as the regional feature corresponding to the regional characteristic dimension. The time-series features, the fluctuation features, and the regional features are used as multi-dimensional features.

3. The method according to claim 2, characterized in that, Determining the load type of the power grid based on the multi-dimensional features includes: If the time-series trend fitting degree corresponding to the time-series feature is greater than or equal to a preset time-series trend fitting degree threshold, the load volatility corresponding to the fluctuation feature is less than or equal to a preset load volatility threshold, and the regional load correlation corresponding to the regional feature is greater than or equal to a preset regional correlation threshold, then the load type of the power grid is determined to be a stable load. If the time-series trend fitting degree corresponding to the time-series feature is within a preset time-series trend fitting degree range, the load volatility corresponding to the fluctuation feature is within a preset load volatility range, and the regional load correlation corresponding to the regional feature is within a preset regional correlation range, then the load type of the power grid is determined to be a fluctuating load. If the time-series trend fitting degree corresponding to the time-series feature is less than the preset time-series trend fitting degree threshold, the load volatility corresponding to the fluctuation feature is greater than the preset load volatility threshold, and the regional load correlation corresponding to the regional feature is less than the preset regional correlation threshold, then the load type of the power grid is determined to be a sudden load.

4. The method according to claim 1, characterized in that, The load type is a stable load; the method of determining the current load forecast result of the power grid based on the target size model collaborative forecasting strategy includes: The multi-dimensional target load data and the corresponding time-series characteristics are input into a small model to obtain an initial load prediction result, and the predicted load characteristics corresponding to the initial load prediction result are determined; the small model is a lightweight time-series prediction model trained based on historical load time-series data of the target area in the power grid. Using a large model, the predicted load characteristics are verified based on the actual load characteristics corresponding to the multi-dimensional target load data, and the verification results are obtained; the large model is a power time series prediction model trained based on a training set including multi-source power data; the large model has more model parameters than the small model. If the verification result indicates that the difference between the actual load characteristics and the predicted load characteristics is greater than a preset difference threshold, the model parameters of the small model are adjusted using the large model to obtain the adjusted small model. The multi-dimensional target load data and the corresponding time-series characteristics are input into the adjusted small model to obtain the current load prediction result of the power grid.

5. The method according to claim 1, characterized in that, The load type is a sudden load; the method of determining the current load forecast result of the power grid based on the target size model collaborative forecasting strategy includes: The multi-dimensional target load data is input into a small model to obtain the initial load prediction result; the small model is a lightweight time-series prediction model trained based on the historical load time-series data of the target area in the power grid. Using a large model, the initial load forecast result is corrected based on the burst load data associated with the historical burst loads of the power grid to obtain the current load forecast result of the power grid; the large model is a power time series forecast model trained based on a training set including multi-source power data; the large model has more model parameters than the small model.

6. The method according to claim 1, characterized in that, The load type is fluctuating load; the method of determining the current load forecast result of the power grid based on the target size model collaborative forecasting strategy includes: The multi-dimensional target load data is input into the large model to obtain the initial load prediction result; the large model is a power time series prediction model trained based on a training set including multi-source power data. A small model is used to extract fluctuation characteristics corresponding to fluctuation features from the currently collected actual load data; the small model is a lightweight time series prediction model trained based on historical load time series data of the target area in the power grid; the model parameters of the small model are fewer than those of the model parameters of the small model. The fluctuation characteristics are fed back to the large model, so that the large model can correct the initial load forecast result based on the fluctuation characteristics to obtain the current load forecast result of the power grid.

7. The method according to any one of claims 1 to 6, characterized in that, The method further includes: The multi-dimensional features are dynamically monitored, and the magnitude of change of each of the multi-dimensional features is determined based on the monitoring results. Collect multiple sets of historical raw load data at preset time intervals, and determine the comprehensive load characteristics corresponding to each set of historical raw load data; Based on multiple comprehensive load characteristics, a preset load characteristic evolution trend calculation model is used to determine the load characteristic evolution trend index, and based on the load characteristic evolution trend index, the corresponding trend judgment result is determined. Based on the change magnitude of the characteristics of each dimension, the evolution trend of the load characteristics, and the trend determination results, load characteristic evolution analysis data is generated. Based on the load characteristic evolution analysis data, if it is determined that the current load type has changed, the size model coordinated prediction strategy is switched from a first size model coordinated prediction strategy that matches the current load type to a second size model coordinated prediction strategy that matches the changed load type, and the load prediction result of the power grid is determined based on the second size model coordinated prediction strategy.

8. A load forecasting device that coordinates large and small models, characterized in that, The device includes: The data extraction module is used to extract multi-dimensional target load data from the collected historical raw load data of the power grid; The feature determination module is used to determine multi-dimensional features based on the multi-dimensional target load data; the multi-dimensional features include time-series features corresponding to the time-series characteristic dimension, fluctuation features corresponding to the fluctuation characteristic dimension, and regional features corresponding to the regional characteristic dimension. The load type determination module is used to determine the load type of the power grid based on each dimension of the multi-dimensional features. The load forecasting module is used to determine a target size model collaborative forecasting strategy that matches the load type from multiple size model collaborative forecasting strategies, and to determine the current load forecasting result of the power grid based on the target size model collaborative forecasting strategy.

9. A computer device, comprising a memory and a processor, characterized in that, The memory stores a computer program; when the processor executes the computer program, it implements the steps of the method according to any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1 to 7.

Images