Multi-industry power load prediction method, device, equipment, medium and program product

Abstract

The invention discloses a multi-industry power load prediction method, device and equipment, a medium and a program product, and the method comprises the steps: obtaining the power load data of at least one data source, and determining a target industry field to which the power load data belongs based on the data characteristics of the power load data; obtaining a load probability prediction model corresponding to the target industry field, wherein the load probability prediction model at least comprises a conformal quantile regression model; and determining a load prediction interval of the power load data based on the load probability prediction model. According to the method, the target industry field is accurately identified by analyzing the data features of the power load data, and then the model is selected in a targeted manner, so that prediction deviation caused by industry difference of a traditional model is avoided, and prediction accuracy is improved; the prediction uncertainty is converted into the load prediction interval through the load probability prediction model, the problem that power supply and demand are difficult to accurately match due to the fact that a traditional prediction method cannot quantify the uncertainty is solved, and the prediction reliability is improved.

Description

Technical Field

[0001] This invention relates to the field of load forecasting technology, and in particular to a method, apparatus, equipment, medium, and program product for forecasting power load across multiple industries. Background Technology

[0002] Against the backdrop of accelerated power market reform, accurate power load forecasting has become a core technology for ensuring the safe operation of the power system. Traditional power load forecasting methods often rely on point forecasting techniques, outputting a single forecast value based on historical load data. While these methods can meet basic forecasting needs under normal circumstances, their prediction errors increase significantly when faced with extreme weather, major events, or other scenarios of drastic load fluctuations. Furthermore, traditional methods cannot quantify forecast uncertainty, making it difficult for grid dispatch to accurately match supply and demand. Therefore, providing a power load forecasting method that can effectively quantify forecast uncertainty has become an urgent problem to be solved in the field of load forecasting technology. Summary of the Invention

[0003] This invention provides a method, apparatus, equipment, medium, and program product for predicting power load across multiple industries. This invention avoids the prediction bias caused by industry differences in traditional general models, thereby improving prediction accuracy. It also solves the problem of difficulty in accurately matching power supply and demand caused by the inability to quantify uncertainties in traditional prediction methods, thus improving the reliability of power load prediction.

[0004] One aspect of this invention provides a multi-industry power load forecasting method, comprising:

[0005] Obtain power load data from at least one data source, and determine the target industry sector to which the power load data belongs based on the data characteristics of the power load data;

[0006] Obtain the load probability prediction model corresponding to the target industry sector. The load probability prediction model should include at least a conformal quantile regression model.

[0007] The load forecast interval for power load data is determined based on the load probability prediction model.

[0008] In one aspect of this invention, a multi-industry power load forecasting device is provided, comprising:

[0009] The domain determination module is used to acquire power load data from at least one data source and determine the target industry domain to which the power load data belongs based on the data characteristics of the power load data.

[0010] The model acquisition module is used to acquire the load probability prediction model corresponding to the target industry sector. The load probability prediction model includes at least a conformal quantile regression model.

[0011] The interval determination module is used to determine the load prediction interval of power load data based on the load probability prediction model.

[0012] In another aspect of the present invention, an apparatus is provided, comprising:

[0013] At least one processor; and

[0014] Memory that is communicatively connected to at least one processor;

[0015] The memory stores a computer program that can be executed by at least one processor, which enables the at least one processor to execute the multi-industry power load forecasting method of any embodiment of the present invention.

[0016] In another aspect of the present invention, a computer-readable storage medium is provided, comprising: computer instructions, which enable a processor to execute the multi-industry power load forecasting method of any embodiment of the present invention when executed.

[0017] In another aspect of the present invention, a computer program product is provided, comprising: a computer program, wherein the computer program is executed by a processor using the multi-industry power load forecasting method of any embodiment of the present invention.

[0018] This invention involves acquiring power load data from at least one data source, determining the target industry sector to which the power load data belongs based on its data characteristics, obtaining a load probability prediction model corresponding to the target industry sector (the load probability prediction model includes at least a conformal quantile regression model), and determining the load prediction interval for the power load data based on the load probability prediction model. By analyzing the data characteristics of power load data, this invention can accurately identify the target industry sector to which the power load data belongs, and thus select a targeted load probability prediction model. This avoids prediction biases caused by industry differences in traditional general models, significantly improving the accuracy of power load prediction. Furthermore, by transforming the uncertainty of power load prediction into a quantifiable power load prediction interval through the load probability prediction model, this invention solves the problem of inaccurate matching of power supply and demand caused by the inability to quantify uncertainty in traditional prediction methods, thereby improving the reliability of power load prediction.

[0019] It should be understood that the description in this section is not intended to identify key or essential features of the embodiments of the present invention, nor is it intended to limit the scope of the invention. Other features of the invention will become readily apparent from the following description. Attached Figure Description

[0020] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0021] Figure 1 This is a flowchart of a multi-industry power load forecasting method provided by an embodiment of the present invention;

[0022] Figure 2 This is a flowchart of a multi-industry power load forecasting method based on conformal prediction provided by an embodiment of the present invention;

[0023] Figure 3 A schematic diagram of the structure of a multi-industry power load forecasting device provided in an embodiment of the present invention;

[0024] Figure 4 This is a block diagram of an apparatus for performing a multi-industry power load forecasting method, provided in an embodiment of the present invention. Detailed Implementation

[0025] To enable those skilled in the art to better understand the present invention, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of the present invention.

[0026] It should be noted that the terms "first," "second," etc., in the specification, claims, and accompanying drawings of this invention are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of the invention described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover a non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0027] Figure 1This invention provides a flowchart of a multi-industry power load forecasting method, applicable to scenarios involving uncertainty quantification forecasting of power load. This method can be executed by a multi-industry power load forecasting device, which can be implemented in hardware and / or software and configured within an equipment. Figure 1 As shown, the method includes:

[0028] S110. Obtain power load data from at least one data source, and determine the target industry sector to which the power load data belongs based on the data characteristics of the power load data.

[0029] The data source refers to the source that provides power load data. For example, the data source may include an online system that collects power load data in real time, such as the power monitoring system of a power company, or a database that stores power load data, such as the database of a third-party data service provider.

[0030] Data characteristics can be understood as information used to describe the attributes of electricity load data, reflecting the distribution patterns of electricity load data. Data characteristics include, but are not limited to, dimensions such as temporal fluctuation characteristics, industry-driven characteristics, and climate-sensitive characteristics. Electricity load data exhibits different characteristics across different industry sectors. For example, in the industrial sector, the temporal fluctuation characteristics of electricity load data show periodic fluctuations closely coupled with production plans; industry-driven characteristics show a strong correlation with industrial production plans; and climate-sensitive characteristics show a relatively stable baseline level with small fluctuations. In the commercial sector, electricity load data shows significant daily and weekly fluctuations in temporal fluctuation characteristics, with significant differences in load levels between weekdays, holidays, and weekends. Industry-driven characteristics show a positive correlation with indicators such as business climate indices and passenger flow; and climate-sensitive characteristics show obvious seasonal fluctuations, with significant increases in electricity demand during the high-temperature periods of summer and the low-temperature periods of winter. In the green energy sector, the temporal fluctuation characteristics of electricity load data show non-stationary random fluctuations dominated by natural conditions; industry-driven characteristics show a functional relationship with renewable energy generation capacity and energy storage system operation strategies; and climate-sensitive characteristics show a high dependence on meteorological factors.

[0031] Electricity load data refers to a dataset reflecting electricity usage. For example, electricity load data may include actual power consumption values, meteorological data associated with those values, time-related data, or industry-specific characteristics. It may also include future meteorological data. The forecast timeframe can be determined based on the future meteorological data included in the electricity load data. For instance, if the electricity load data includes meteorological data for the next N days, the forecast target is the electricity load for those N days. In the electricity load data collection phase, full-volume electricity load data covering the overall electricity consumption of the power system can be directly extracted from the data source. Alternatively, differentiated collection strategies can be customized based on the core electricity consumption characteristics of different industry sectors. By adapting collection strategies to various industry scenarios, targeted collection of electricity load data that accurately reflects the electricity consumption patterns of the corresponding industry sectors can be achieved.

[0032] The target industry sector refers to the specific industry to which the power load data belongs, determined based on the data characteristics of the power load data. The target industry sector may include the industrial sector, the commercial sector, or the green energy sector, etc.

[0033] Specifically, power load data is obtained from data sources such as power monitoring systems or databases of third-party data service providers through open application programming interfaces (APIs) or by constructing structured query statements to query power load data. The distribution patterns of the power load data are statistically analyzed to determine its characteristics. Then, rule-based matching methods are used to match these characteristics with preset industry characteristic rules to determine the target industry sector to which the power load data belongs. Alternatively, a machine learning classification model can be used. The data characteristics are input into a pre-trained machine learning classification model, such as a decision tree model or a support vector machine model. The machine learning classification model outputs the target industry sector to which the power load data belongs. Preset industry characteristic rules refer to sector-specific criteria built based on the characteristics of electricity usage in an industry sector. For example, preset industry characteristic rules might include classifying power load data as belonging to the industrial sector if it has stable peaks on weekdays and its fluctuations are strongly correlated with production plans, or classifying it as belonging to the commercial sector if it exhibits significant diurnal rhythms and a sharp drop in load on weekends and holidays.

[0034] S120. Obtain the load probability prediction model corresponding to the target industry sector. The load probability prediction model shall include at least a conformal quantile regression model.

[0035] A load probability forecasting model can be understood as a tool for predicting future electricity load. The output of a load probability forecasting model can include predicted electricity load values ​​or predicted ranges, with the predicted range used to quantify the uncertainty of the predicted electricity load values. Load probability forecasting models include at least a conformal quantile regression model, and may also include conformal prediction or quantile regression models.

[0036] Specifically, the load probability prediction model is pre-stored in a model repository or database. In the model repository or database, there is a one-to-one correspondence between the load probability prediction model and the domain identifier or domain name of the target industry sector. After determining the target industry sector to which the power load data belongs, the domain identifier or domain name of the target industry sector is obtained. Based on the one-to-one correspondence between the domain identifier or domain name and the load probability prediction model, the load probability prediction model corresponding to the target industry sector is extracted from the model repository or database.

[0037] S130. Determine the load prediction range of power load data based on the load probability prediction model.

[0038] The load forecast interval refers to the range of future electricity load values, reflecting the uncertainty of electricity load forecasting. The width of the load forecast interval reflects the degree of uncertainty; a narrower interval indicates higher forecast certainty, while a wider interval indicates greater uncertainty. The load forecast interval can include an upper and lower limit. It's worth noting that the upper and lower limits are not limited to single values; they can also be a set of forecast load values ​​corresponding to multiple different times. From a data representation perspective, this set can be fitted as a continuous curve reflecting the dynamic changes of load over time.

[0039] Specifically, the power load data is input into the load probability prediction model. The load probability prediction model processes the power load data based on the learned algorithm logic and outputs the load prediction range of the power load data. This load prediction range can reasonably reflect the fluctuation range of power load in a specific future period.

[0040] Furthermore, in this embodiment of the invention, before determining the load prediction range of the power load data based on the load probability prediction model, the power load data may be preprocessed. The data preprocessing steps may include noise removal, missing value filling, or standardization. This embodiment of the invention does not limit the data preprocessing steps.

[0041] This invention involves acquiring power load data from at least one data source, determining the target industry sector to which the power load data belongs based on its data characteristics, obtaining a load probability prediction model corresponding to the target industry sector (the load probability prediction model includes at least a conformal quantile regression model), and determining the load prediction interval for the power load data based on the load probability prediction model. By analyzing the data characteristics of power load data, this invention can accurately identify the target industry sector to which the power load data belongs, thereby enabling the targeted selection of a load probability prediction model. This avoids prediction biases caused by industry differences in traditional general models, significantly improving the accuracy of power load prediction and reducing prediction errors. Furthermore, by transforming the uncertainty of power load prediction into a quantifiable power load prediction interval through the load probability prediction model, this invention solves the problem of inaccurate matching of power supply and demand caused by the inability to quantify uncertainty in traditional prediction methods, thus improving the reliability of power load prediction.

[0042] Furthermore, this embodiment of the invention further refines step S110, specifically detailing how to determine the target industry sector, including:

[0043] S1101. The time fluctuation characteristics, industry-driven characteristics, and climate sensitivity characteristics of statistical power load data are used as data features.

[0044] Among them, time fluctuation characteristics refer to the regularity of power load data changes over time, and the time scale can include hours, days, weeks, months or years, etc.

[0045] Industry-driven characteristics refer to the features exhibited by power load data due to production and operation activities, which may include production processes, production scale, or market demand.

[0046] Climate sensitivity refers to the characteristics of electricity load data as a result of climate conditions, which may include temperature, humidity, or wind speed.

[0047] Specifically, electricity load data is obtained from at least one data source. The actual power consumption value in the electricity load data is used as the dependent variable, and the time series, production and operation activities, and climate conditions in the electricity load data are used as independent variables. In different industry sectors, the same independent variable will have different degrees of impact on the actual power consumption value. Curves of the actual power consumption value in the electricity load data changing with time series, production and operation activities, and climate conditions are plotted respectively. The distribution patterns of electricity load data under time series, production and operation activities, and climate conditions are identified through the curves. The identified distribution patterns are used as the time fluctuation characteristics, industry-driven characteristics, and climate-sensitive characteristics of electricity load data, respectively.

[0048] S1102. Obtain domain template features, match data features with domain template features, and determine the target industry domain to which the power load data belongs; the domain template features include at least: industrial domain template features, commercial domain template features, and green energy domain template features.

[0049] Domain template features refer to a pre-defined set of standardized features used to describe the electricity load characteristics of different industry sectors. These domain template features serve as a benchmark for industry sector classification and can be obtained through statistical analysis based on electricity usage and / or industry characteristics. Domain template features include at least industrial, commercial, and green energy domain template features. Industrial template features refer to the standard features of electricity load data summarized for the industrial sector; commercial template features refer to the standard features of electricity load data summarized for the commercial sector; and green energy domain template features refer to the standard features of electricity load data summarized for the green energy sector. The industrial, commercial, and green energy domain template features may respectively include time-varying standard features, industry-driven standard features, and climate-sensitive standard features of electricity load data.

[0050] Specifically, predefined domain template features can be stored in internal system configuration files or external databases. These predefined domain template features can be obtained by calling application programming interfaces or executing database queries. The domain template features include at least industrial, commercial, and green energy domain template features. The data features of the statistically obtained power load data are compared with the obtained domain template features. By using similarity calculation algorithms, such as cosine similarity or Euclidean distance, the similarity between the power load data features and each domain template feature is quantitatively evaluated. The industry category corresponding to the domain template feature with the highest similarity is determined as the target industry domain. Alternatively, based on pre-set industry feature rules, it is checked whether the power load data features meet the rule conditions of a specific domain. If the power load data features completely match the rules of a certain industry domain or the matching degree reaches a preset threshold, then that industry domain is determined as the target industry domain to which the power load data belongs.

[0051] Furthermore, this embodiment of the invention optimizes the multi-industry power load forecasting method. Specifically, it optimizes the specific steps for training the load probability forecasting model, including:

[0052] A1. Obtain power load data from at least one data source, and randomly divide the power load data into a training set, a calibration set, and a test set, with no overlap between the training set, the calibration set, and the test set.

[0053] Specifically, power load data from at least one data source can be obtained. The proportions of training set, calibration set, and test set in the power load data can be set. Data of the corresponding proportion can be extracted from the power load data according to the set proportion using a random sampling method, and used as the training set, calibration set, and test set. Alternatively, the power load data can be arranged according to the original acquisition time, and data of the set proportion can be extracted according to the chronological order as the training set, calibration set, and test set.

[0054] A2. Obtain the preset false coverage rate of the load probability prediction model, take half of the preset false coverage rate as the low quantile of the load probability prediction model, and take the complementary value of the half as the high quantile of the load probability prediction model.

[0055] The preset error coverage rate can be understood as a pre-set probability threshold that allows the predicted power load value to fall outside the prediction interval. It is used to determine the boundary of the load prediction interval. For example, half of the preset error coverage rate is taken as the lower quantile to determine the lower boundary of the prediction interval, and the complementary value of half of the preset error coverage rate is taken as the higher quantile to determine the upper boundary of the prediction interval.

[0056] Specifically, the preset error coverage rate of the load probability prediction model can be obtained by scanning the configuration file of the load probability prediction model, or the preset error coverage rate of the training load probability prediction model can be manually entered through the visual interactive interface. Half of the preset error coverage rate is determined as the low quantile, which is used to calculate the lower limit of the load prediction interval. The value of 1 minus the low quantile is determined as the high quantile, which is used to calculate the upper limit of the prediction interval.

[0057] A3. Obtain the loss function of the load probability prediction model, input the training set into the load probability prediction model, minimize the loss function through the stochastic gradient descent algorithm, combine the low quantile and high quantile, and output the first prediction interval of the training dataset. The first prediction interval includes at least the upper limit and the lower limit of the first prediction interval.

[0058] The first prediction interval refers to the load prediction range output by the load probability prediction model during the training phase. It is learned by training the load probability prediction model using the training set and includes the upper limit and the lower limit of the first prediction interval. The upper limit of the first prediction interval is the maximum value in the first prediction interval, which corresponds to the prediction result of the high quantile. The lower limit of the first prediction interval is the minimum value in the first prediction interval, which corresponds to the prediction result of the low quantile.

[0059] The loss function includes at least the following:

[0060]

[0061] Among them, y iLet i be the actual load value at the i-th time point. The predicted load value is the actual load value at the i-th time point. The predicted load value Compared with the actual load value y i The absolute difference between them, C i The predicted interval for the actual load value at the i-th time point is... The indicator function is of the form: τ is the target quantile of the quantile regression loss function. Let quantile regression loss function be the form of quantile regression loss function.

[0062] The absolute difference value can be understood as an absolute value used to quantify the difference between the actual load value and the predicted load value at the i-th time point. The absolute difference value can intuitively reflect the magnitude of the prediction error at a single time point. The actual load value refers to the actual power consumption value in the power load data at the i-th time point, which can be used as a benchmark reference data for predicting future load values. The predicted load value is the estimate of future power load by the load probability prediction model based on the actual load value, and is the output result of the load probability prediction model.

[0063] The forecast interval can be understood as an interval output by the load probability forecasting model that includes the possible range of actual load values, used to quantify the uncertainty of the forecast.

[0064] An indicator function can be understood as a binary function used to determine whether the predicted load value at the i-th time point is within the corresponding prediction interval. The indicator function intuitively identifies the inclusion relationship between the predicted value and the prediction interval by taking the value 1 or 0.

[0065] The target quantile can be understood as a quantile point parameter that is set in advance in quantile regression. The target quantile determines the focus of the quantile regression model fitting, and different values ​​of the target quantile correspond to different boundaries of the prediction interval.

[0066] The quantile regression loss function can be understood as a function used to optimize the parameters of a quantile regression model. The quantile regression loss function uses different calculation methods depending on the sign of the error u, and is used to make the load probability prediction model fit the predicted value corresponding to the specified quantile.

[0067] Specifically, the loss function of the load probability prediction model is obtained, the training set data is converted into a format recognizable by the load probability prediction model, and input into the load probability prediction model. At each iteration, the gradient of the loss function with respect to the parameters is calculated, and the parameters of the load probability prediction model are adjusted in the opposite direction of the gradient to reduce the loss value. When the loss function converges, the first prediction interval for each time point in the training set is output by combining the low quantile and the high quantile. The first prediction interval includes an upper limit and a lower limit, wherein the upper limit is determined by the high quantile and the lower limit is determined by the low quantile.

[0068] A4. When the rate of change of the loss function converges, the training of the load probability prediction model is considered complete.

[0069] Specifically, continuously monitor the rate of change of the loss function during the training of the load probability prediction model. The ratio of the difference between the loss values ​​of two adjacent iterations to the previous loss value is taken as the rate of change of the loss function. If the rate of change of the loss function is stabilized below a preset threshold, the loss function can be considered to have reached the convergence condition. When the rate of change of the loss function converges, the training of the load probability prediction model is stopped.

[0070] Furthermore, in this embodiment of the invention, when the training of the load probability prediction model reaches the required number of iterations, the training of the load probability prediction model can also be considered complete.

[0071] A5. Calibrate the load probability prediction model based on the calibration set, and validate the load probability prediction model based on the test set.

[0072] Specifically, after training the load probability prediction model using the training set, the calibration set and test set are input into the load probability prediction model, and the quasi-set and test set are used to calibrate and verify the load probability prediction model.

[0073] Furthermore, this embodiment of the invention optimizes the multi-industry power load forecasting method. Specifically, it optimizes the specific steps for calibrating the load probability forecasting model, including:

[0074] B1. Obtain a calibration set of power load data, input the calibration set into the trained load probability prediction model, and output the second prediction interval of the calibration set. The second prediction interval includes at least the upper limit and the lower limit of the second prediction interval.

[0075] The second prediction interval refers to the load prediction range output by the load probability prediction model during the calibration phase. It is obtained by calibrating the load probability prediction model using a calibration set and includes the upper limit and the lower limit of the second prediction interval. The upper limit of the second prediction interval is the maximum value in the second prediction interval, which is used to define the upper limit boundary of the possible values ​​of the predicted load value. The lower limit of the second prediction interval is the minimum value in the second prediction interval, which is used to define the lower limit boundary of the possible values ​​of the predicted load value.

[0076] Specifically, a pre-divided calibration set is extracted from the power load data. After the calibration set is converted into a format that can be processed by the load probability prediction model, it is input into the load probability prediction model. The load probability prediction model processes each power load data in the calibration set and outputs the predicted values ​​corresponding to the high quantile and low quantile, respectively, forming the second prediction interval. The upper limit of the second prediction interval is the predicted value corresponding to the high quantile, and the lower limit of the second prediction interval is the predicted value corresponding to the low quantile.

[0077] B2. Obtain the predicted load value corresponding to the calibration set, and the first false coverage rate where the predicted load value does not fall within the second prediction interval.

[0078] The first false coverage rate refers to the proportion of predicted load values ​​corresponding to actual load values ​​in the calibration set that do not fall within the second prediction interval. It is the ratio of the number of samples of predicted load values ​​outside the second prediction interval to the total number of power loads in the calibration set, and is used to measure the reliability of the prediction interval of the load probability prediction model.

[0079] Specifically, the actual load value at each time point is extracted from the original power load data of the calibration set. The predicted load value output by the load probability prediction model after processing each actual load value is recorded. The predicted load value is compared with the second prediction interval. If the predicted load value is greater than the upper limit of the second prediction interval or less than the lower limit of the second prediction interval, it is determined that the predicted load value does not fall within the second prediction interval. The number of predicted load values ​​that do not fall within the interval is counted. The proportion of the number of predicted load values ​​that fall within the second prediction interval to the total power load data of the calibration set is calculated. This proportion is used as the first false coverage rate.

[0080] B3. If the first false coverage rate is not equal to the preset false coverage rate, then compare the second prediction interval with the first prediction interval to obtain the boundary offset between the second prediction interval and the first prediction interval; wherein, the upper limit offset of the boundary offset is the difference between the upper limit of the second prediction interval and the upper limit of the first prediction interval, and the lower limit offset of the boundary offset is the difference between the lower limit of the second prediction interval and the lower limit of the first prediction interval.

[0081] The boundary offset can be understood as a parameter used to quantify the difference between the boundaries of the second and first prediction intervals, and can also reflect the degree of offset of the prediction intervals. The boundary offset includes an upper limit offset and a lower limit offset. The upper limit offset is the difference between the upper limit of the second and first prediction intervals. If the upper limit offset is positive, it means that the upper limit of the second prediction interval in the calibration set is higher than the upper limit of the first prediction interval in the training set; if it is negative, the upper limit of the second prediction interval is lower than the upper limit of the first prediction interval in the training set. The lower limit offset is the difference between the lower limit of the second and first prediction intervals. If the lower limit offset is positive, it means that the lower limit of the second prediction interval in the calibration set is higher than the lower limit of the first prediction interval in the training set; if it is negative, the lower limit of the second prediction interval is lower than the lower limit of the first prediction interval in the training set.

[0082] Specifically, the following steps are taken: obtaining the first false coverage rate, the preset false coverage rate, the first prediction interval output by the load probability prediction model during the training phase, and the second prediction interval output by the load probability prediction model during the calibration phase. The first prediction interval includes at least the upper limit and the lower limit of the first prediction interval, and the second prediction interval includes at least the upper limit and the lower limit of the second prediction interval. The first false coverage rate is compared with the preset false coverage rate. If the first false coverage rate is not equal to the preset false coverage rate, the boundary offset between the second prediction interval and the first prediction interval is calculated. The difference between the upper limit of the second prediction interval and the upper limit of the first prediction interval is used as the upper limit offset of the boundary offset, and the difference between the lower limit of the second prediction interval and the lower limit of the first prediction interval is used as the lower limit offset of the boundary offset.

[0083] Furthermore, this embodiment of the invention optimizes the multi-industry power load forecasting method. Specifically, it optimizes the steps for validating the load probability forecasting model, including:

[0084] C1. Obtain a test set of power load data, input the test set into the trained load probability prediction model, and output the third prediction interval of the test set. The third prediction interval includes at least the upper limit and the lower limit of the third prediction interval.

[0085] The third prediction interval refers to the load prediction range output by the load probability prediction model during the testing phase. It is obtained by calibrating the load probability prediction model using the test set and includes the upper limit and lower limit of the third prediction interval. The upper limit of the third prediction interval is the maximum value in the third prediction interval, which is used to define the upper limit boundary of the possible values ​​of the predicted load value. The lower limit of the third prediction interval is the minimum value in the third prediction interval, which is used to define the lower limit boundary of the possible values ​​of the predicted load value.

[0086] Specifically, a test set of power load data is obtained and input into a pre-trained load probability prediction model. After processing, the load probability prediction model outputs a third prediction interval for the test set, which includes at least the upper limit and the lower limit of the third prediction interval.

[0087] C2. Obtain the upper limit offset and lower limit offset corresponding to the calibration set, adjust the upper limit of the third prediction interval according to the upper limit offset, and use the adjusted upper limit of the third prediction interval as the final upper limit of the third prediction interval.

[0088] The final third forecast interval refers to the third forecast interval after adjustments for the upper and lower limit offsets. The final third forecast interval accurately reflects the range within which the forecast load value may fall. The final third forecast interval may include an upper limit and a lower limit. The upper limit is the upper value of the third forecast interval after adjustment according to the upper limit offset, and the lower limit is the lower value of the third forecast interval after adjustment according to the lower limit offset.

[0089] Specifically, the upper and lower offsets of the calibration set are obtained from the corresponding data storage location of the load probability prediction model. The upper limit of the third prediction interval is adjusted according to the upper limit offset, and the adjusted value is used as the upper limit of the final third prediction interval.

[0090] C3. Adjust the lower limit of the third prediction interval according to the lower limit offset, and use the adjusted lower limit of the third prediction interval as the final lower limit of the third prediction interval.

[0091] Specifically, the upper and lower limit offsets of the calibration set are obtained from the corresponding data storage location of the load probability prediction model. The lower limit of the third prediction interval is adjusted according to the lower limit offset, and the adjusted value is used as the lower limit of the final third prediction interval.

[0092] C4. Obtain the predicted load value corresponding to the test set, and the second false coverage rate where the predicted load value does not fall within the final third prediction interval.

[0093] The second false coverage rate refers to the proportion of predicted load values ​​corresponding to actual load values ​​in the test set that do not fall within the final third prediction interval. It is the ratio of the number of samples with predicted load values ​​outside the final third prediction interval to the total number of power load data in the test set, and is used to measure the reliability of the prediction interval of the load probability prediction model.

[0094] Specifically, the actual load value at each time point is extracted from the raw power load data of the test set. The predicted load value output by the load probability prediction model for each actual load value is recorded. The predicted load value is compared with the final third prediction interval. If the predicted load value is greater than the upper limit of the final third prediction interval or less than the lower limit of the final third prediction interval, it is determined that the predicted load value does not fall within the final third prediction interval. The number of predicted load values ​​that do not fall within the interval is counted. The proportion of the number of predicted load values ​​that fall within the final third prediction interval to the total number of power load data in the calibration set is calculated. This proportion is used as the second false coverage rate.

[0095] C5. If the second false coverage rate is equal to the preset false coverage rate, then the load probability prediction model is verified; otherwise, the load probability prediction model is retrained and calibrated.

[0096] Specifically, the second false coverage rate and the preset false coverage rate are obtained, and the first false coverage rate is compared with the preset false coverage rate. If the first false coverage rate is equal to the preset false coverage rate, it indicates that the load probability prediction model has passed the validation. If the first false coverage rate is not equal to the preset false coverage rate, it indicates that the load probability prediction model has not passed the validation and needs to be retrained and calibrated.

[0097] This invention provides another method for multi-industry power load forecasting. This invention is an optimization of the above-mentioned embodiments. Specifically, this invention optimizes the preprocessing steps of power load data, the steps of power load forecasting using a conformal prediction model, the steps of power load forecasting using a quantile regression model, and the steps of power load forecasting using a conformal quantile regression model. This method mainly includes three parts: data preprocessing, industry feature adaptation, and load probability prediction model training.

[0098] In the data preprocessing stage, firstly, the power load data is cleaned to remove noise and incomplete data records. Next, for missing values ​​in the power load data, appropriate interpolation methods are used to fill in the missing values. To ensure the stability and consistency of the power load data, all processed power load data are standardized using the standardization formula (1).

[0099]

[0100] Among them, X ′ The data is standardized, where X is the original data, μ is the mean, and σ is the standard deviation. Preprocessing ensures consistency and high quality across different industries, providing reliable input data for subsequent model training.

[0101] In the industry-specific adaptation phase, the load characteristics of the electricity load data are analyzed. These characteristics can include temporal fluctuations, industry-driven features, and climate sensitivity. Based on these characteristics, the industry sector to which the electricity load data belongs (e.g., industry, commerce, and green energy) can be determined. Specifically, the load characteristics of electricity load data vary across different industry sectors: In the industrial sector, temporal fluctuations are often highly correlated with production shifts, exhibiting regular fluctuations. Among the industry-driven features, production order volume and equipment operating efficiency are the core influencing factors, while climate sensitivity is relatively weak. In the commercial sector, temporal fluctuations show significant differences in load between weekdays, holidays, and weekends. Industry-driven features are closely related to the intensity of commercial activities, while climate sensitivity is moderate, with summer air conditioning use and winter heating causing a phased increase in load as temperature changes. In the green energy sector, temporal fluctuations are highly random due to natural conditions. Among the industry-driven features, the installed capacity of renewable energy and the charging and discharging strategies of energy storage devices are key factors, while climate sensitivity is most prominent, with meteorological factors such as sunshine duration, wind speed, and cloud cover directly determining the size and stability of the load.

[0102] By adapting features, this invention improves prediction accuracy and enables rapid optimization of the load probability prediction model based on load characteristics, thereby effectively addressing different power load prediction needs. The flexible adjustment of the load probability prediction model allows for wide application in different industries and dynamic power dispatching scenarios, enhancing the robustness and adaptability of the prediction.

[0103] During the model training phase, this embodiment of the invention employs three methods: conformal prediction, quantile regression, and conformal quantile regression. These methods effectively quantify the uncertainty of power load forecasting. By using these three methods, confidence intervals for the forecast results can be generated, providing decision-makers with a more comprehensive risk assessment and quantifying the uncertainty of power load forecasting.

[0104] Conformal forecasting is a non-parametric forecasting method that generates confidence intervals for load forecasts by calculating the forecast error between the predicted results and actual data. The core idea of ​​conformal forecasting is to estimate uncertainty by comparing power load data with the forecast results (predicted values) of a load point forecasting model, without making specific distribution assumptions. A flowchart of conformal forecasting is shown below. Figure 2 As shown in the diagram, firstly, the power load data is divided into a training set, a calibration set, and a test set. The training set is used to train a pre-defined probabilistic prediction model, the calibration set is used to calibrate the pre-defined point prediction model, and the test set is input into the load probabilistic prediction model to output predicted values. And determine the prediction error ∈ t The prediction error calculation formula includes at least formula (2):

[0105]

[0106] in, Represents the predicted value, y t Represents the actual load value in the power load data, ∈ t Represents the actual observed value y t Compared with the predicted value The absolute difference between them is the prediction error.

[0107] Then, the prediction error of all power load data is calculated using the conformal prediction method, and appropriate prediction error quantiles (usually high and low quantiles) are selected according to the preset error coverage rate (e.g., 5%) to construct a confidence interval (prediction interval). The low quantile is half the error coverage rate, and the high quantile is the complementary value of the half-value. The confidence interval is expressed by formula (3):

[0108]

[0109] in, Let represent the confidence interval at time t. q is the predicted value. α / 2 Let α be the quantile, and α represent the false coverage rate, which is the probability that the true value falls outside this interval. This is the upper bound of the prediction interval. This represents the lower bound of the prediction interval. The width of the confidence interval is determined by the magnitude of the prediction error; a larger error results in a wider prediction interval, and vice versa. The formula for the interval width is:

[0110]

[0111] Among them, Width t Width is the width of the prediction interval. t It is calculated from the quantiles of the predicted value and the prediction error, reflecting the uncertainty of the prediction.

[0112] Determine whether all data in the test set has been predicted. If all data has been predicted, output the prediction interval.

[0113] Quantile regression generates predicted values ​​at different quantiles through regression analysis, thus providing multiple confidence intervals for electricity load forecasting. The quantile regression method performs regression fitting on specified quantiles (such as 5% or 95%) to produce multiple prediction intervals. Its main goal is to obtain predicted values ​​at different confidence levels by training the regression model on different quantiles.

[0114] In quantile regression, the loss function is defined based on the quantile τ as follows:

[0115]

[0116] in, It is the quantile loss function, τ is the target quantile, and y i It is the actual load, X i It is the eigenvector, and β is the regression coefficient. This is an indicator function, representing the state when u < 0. otherwise,

[0117] By fitting multiple quantile regression models, the quantile regression method provides upper and lower bounds for the prediction at each time step, thus generating a confidence interval. For a 95% confidence level, we typically select the prediction results at the 5th and 95th percentiles as the upper and lower bounds of the prediction interval.

[0118]

[0119] in, Let represent the confidence interval at time t. This represents the regression results for the lower quantiles. This represents the regression results for the high quantiles. The interval width is determined by the difference between quantiles; a smaller difference indicates a narrower prediction interval, and a larger difference indicates a wider prediction interval.

[0120] Conformal quantile regression combines the advantages of conformal prediction and quantile regression. It first generates preliminary prediction intervals through quantile regression, and then optimizes them using conformal prediction to ensure that the model can better adapt to uncertainty and generate more accurate confidence intervals.

[0121] The loss function of conformal quantile regression combines elements of both methods and is defined as:

[0122]

[0123] Among them, y i Let i be the actual load value at the i-th time point. The predicted load value is the actual load value at the i-th time point. The predicted load value Compared with the actual load value y i The absolute difference between them, C i The predicted interval for the actual load value at the i-th time point is... The indicator function is of the form: τ is the target quantile of the quantile regression loss function. Let quantile regression loss function be the form of quantile regression loss function. In conformal quantile regression, the width of the confidence interval is determined by both the prediction error and the output of the quantile regression. Therefore, conformal quantile regression can provide a more refined prediction interval for each industry and power load.

[0124] Figure 3 This is a schematic diagram of a multi-industry power load forecasting device provided in an embodiment of the present invention. Figure 3 As shown, the device based on the multi-industry power load forecasting method specifically includes: a domain determination module 310, a model acquisition module 320, and an interval determination module 330.

[0125] Domain determination module 310 is used to acquire power load data from at least one data source and determine the target industry domain to which the power load data belongs based on the data characteristics of the power load data.

[0126] The model acquisition module 320 is used to acquire the load probability prediction model corresponding to the target industry sector. The load probability prediction model includes at least a conformal quantile regression model.

[0127] The interval determination module 330 is used to determine the load prediction interval of power load data based on the load probability prediction model.

[0128] Furthermore, the domain determination module 310 includes: a feature statistics unit, used to statistically analyze the temporal fluctuation characteristics, industry-driven characteristics, and climate-sensitive characteristics of power load data as data features;

[0129] The domain matching unit is used to obtain domain template features, match data features with domain template features, and determine the target industry domain to which the power load data belongs; the domain template features include at least: industrial domain template features, commercial domain template features and green energy domain template features.

[0130] Furthermore, the multi-industry power load forecasting device also includes: a model training module, used to acquire power load data from at least one data source, randomly dividing the power load data into a training set, a calibration set, and a test set, with no overlap between the training set, calibration set, and test set; acquiring a preset error coverage rate of the load probability forecasting model, using half of the preset error coverage rate as the lower quantile of the load probability forecasting model, and using the complementary value of the half as the higher quantile of the load probability forecasting model; acquiring the loss function of the load probability forecasting model, inputting the training set into the load probability forecasting model, minimizing the loss function through a stochastic gradient descent algorithm, and combining the lower and higher quantiles to output a first prediction interval for the training dataset, the first prediction interval including at least an upper limit and a lower limit, and the loss function including at least: Among them, y i Let i be the actual load value at the i-th time point. The predicted load value is the actual load value at the i-th time point. The predicted load value Compared with the actual load value y i The absolute difference between them, C i The predicted interval for the actual load value at the i-th time point is... The indicator function is of the form: τ is the target quantile of the quantile regression loss function. Let quantile regression loss function be the form of quantile regression loss function. When the rate of change of the loss function converges, the load probability prediction model is considered to have completed training. Then, the load probability prediction model is calibrated based on the calibration set and validated based on the test set.

[0131] Furthermore, the multi-industry power load forecasting device also includes: a model calibration module, used to acquire a calibration set of power load data, input the calibration set into a trained load probability forecasting model, and output a second prediction interval of the calibration set, wherein the second prediction interval includes at least an upper limit and a lower limit; acquire the predicted load value corresponding to the calibration set, and a first false coverage rate in which the predicted load value falls within the second prediction interval; if the first false coverage rate is not equal to a preset false coverage rate, then compare the second prediction interval with the first prediction interval to acquire the boundary offset between the second prediction interval and the first prediction interval; wherein the upper limit offset of the boundary offset is the difference between the upper limit of the second prediction interval and the upper limit of the first prediction interval, and the lower limit offset of the boundary offset is the difference between the lower limit of the second prediction interval and the lower limit of the first prediction interval.

[0132] Furthermore, the multi-industry power load forecasting device also includes: a model verification module, used to acquire a test set of power load data, input the test set into the trained load probability forecasting model, and output the third prediction interval of the test set, the third prediction interval including at least the upper limit and the lower limit of the third prediction interval; acquire the upper limit offset and lower limit offset corresponding to the calibration set, adjust the upper limit of the third prediction interval according to the upper limit offset, and use the adjusted upper limit of the third prediction interval as the final upper limit of the final third prediction interval; adjust the lower limit of the third prediction interval according to the lower limit offset, and use the adjusted lower limit of the third prediction interval as the final upper limit of the final third prediction interval; acquire the predicted load value corresponding to the test set, and the second false coverage rate of the predicted load value falling in the final third prediction interval; if the second false coverage rate is equal to the preset false coverage rate, the load probability forecasting model is determined to have passed verification; otherwise, the load probability forecasting model is retrained and calibrated.

[0133] The multi-industry power load forecasting device provided in the embodiments of the present invention can execute the multi-industry power load forecasting method provided in any embodiment of the present invention, and has the corresponding beneficial effects of executing the method.

[0134] This invention provides an apparatus for performing a multi-industry power load forecasting method, a computer-readable storage medium, and a computer program product.

[0135] Figure 4 A schematic diagram of an apparatus that can be used to implement the multi-industry power load forecasting method of any embodiment of the present invention is shown. The apparatus is intended to represent various forms of digital computers, such as laptop computers, desktop computers, workstations, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers. The apparatus may also represent various forms of mobile devices, such as personal digital processors, cellular phones, smartphones, wearable devices (e.g., helmets, glasses, watches, etc.), and other similar computing devices. The components shown in the embodiments of the present invention, their connections and relationships, and their functions are merely illustrative and are not intended to limit the implementation of the embodiments of the present invention described and / or claimed herein.

[0136] like Figure 4 As shown, the device includes at least one processor 11 and a memory, such as a read-only memory (ROM) 12 or a random access memory (RAM) 13, communicatively connected to the at least one processor 11. The memory stores computer programs executable by the at least one processor. The processor 11 can perform various appropriate actions and processes based on the computer program stored in the ROM 12 or loaded into the RAM 13 from the storage unit 18. The RAM 13 can also store various programs and data required for device operation. The processor 11, ROM 12, and RAM 13 are interconnected via a bus 14. An input / output (I / O) interface 15 is also connected to the bus 14.

[0137] Multiple components in the device are connected to I / O interface 15, including: input unit 16, such as a keyboard, mouse, etc.; output unit 17, such as various types of displays, speakers, etc.; storage unit 18, such as a disk, optical disk, etc.; and communication unit 19, such as a network card, modem, wireless transceiver, etc. The communication unit 19 allows the device to exchange information / data with other devices through computer networks such as the Internet and / or various telecommunications networks.

[0138] Processor 11 can be a variety of general-purpose and / or special-purpose processing components with processing and computing capabilities. Some examples of processor 11 include, but are not limited to, central processing units, graphics processing units, various special-purpose artificial intelligence computing chips, various processors running machine learning model algorithms, digital signal processors, and any suitable processor, controller, microcontroller, etc. Processor 11 performs the various methods and processes described above, such as multi-industry power load forecasting methods.

[0139] In some embodiments, the multi-industry power load forecasting method may be implemented as a computer program tangibly contained in a computer-readable storage medium, such as storage unit 18. In some embodiments, part or all of the computer program may be loaded into and / or installed on the device via ROM 12 and / or communication unit 19. When the computer program is loaded into RAM 13 and executed by processor 11, one or more steps of the multi-industry power load forecasting method may be performed. Alternatively, in other embodiments, processor 11 may be configured as the multi-industry power load forecasting method by any other suitable means (e.g., by means of firmware).

[0140] Various embodiments of the systems and technologies described above in these embodiments of the present invention can be implemented in digital electronic circuit systems, integrated circuit systems, field-programmable gate arrays, application-specific integrated circuits (ASICs), application-specific standard products (ASICs), systems-on-a-chip (SoCs), payload programmable logic devices, computer hardware, firmware, software, and / or combinations thereof. These various embodiments may include implementations in one or more computer programs that can be executed and / or interpreted on a programmable system including at least one programmable processor, which may be a dedicated or general-purpose programmable processor, capable of receiving data and instructions from a storage system, at least one input device, and at least one output device, and transmitting data and instructions to the storage system, the at least one input device, and the at least one output device.

[0141] Computer programs for implementing the methods of embodiments of the present invention may be written in any combination of one or more programming languages. These computer programs may be provided to a processor of a general-purpose computer, a special-purpose computer, or other programmable data processing apparatus, such that when executed by the processor, the computer programs cause the functions / operations specified in the flowcharts and / or block diagrams to be implemented. The computer programs may be executed entirely on a machine, partially on a machine, or as a standalone software package, partially on a machine and partially on a remote machine, or entirely on a remote machine or server.

[0142] In the context of embodiments of the present invention, a computer-readable storage medium may be a tangible medium that may contain or store a computer program for use by or in conjunction with an instruction execution system, apparatus, or device. A computer-readable storage medium may include, but is not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, apparatus, or devices, or any suitable combination thereof. Alternatively, a computer-readable storage medium may be a machine-readable signal medium. More specific examples of machine-readable storage media include electrical connections based on one or more wires, portable computer disks, hard disks, RAM, ROM, erasable programmable read-only memory (EPROM or flash memory), optical fibers, compact disc read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination thereof.

[0143] To provide interaction with a user, the systems and techniques described herein can be implemented on a device having: a display device (e.g., a cathode ray tube or liquid crystal display monitor) for displaying information to the user; and a keyboard and pointing device (e.g., a mouse or trackball) through which the user provides input to the device. Other types of devices can also be used to provide interaction with the user; for example, feedback provided to the user can be any form of sensory feedback (e.g., visual feedback, auditory feedback, or haptic feedback); and input from the user can be received in any form (including: sound input, voice input, or haptic input).

[0144] The systems and technologies described herein can be implemented in computing systems that include backend components (e.g., as data servers), or middleware components (e.g., application servers), or frontend components (e.g., user computers with graphical user interfaces or web browsers through which users can interact with implementations of the systems and technologies described herein), or any combination of such backend, middleware, or frontend components. The components of the system can be interconnected via digital data communication of any form or medium (e.g., communication networks). Examples of communication networks include local area networks (LANs), wide area networks (WANs), blockchain networks, and the Internet.

[0145] A computing system can include clients and servers. Clients and servers are generally located far apart and typically interact through communication networks. The client-server relationship is created by computer programs running on the respective computers and having a client-server relationship with each other. The server can be a cloud server, also known as a cloud computing server or cloud host, which is a host product within the cloud computing service system. This addresses the shortcomings of traditional physical hosts and virtual private servers, such as high management difficulty and weak business scalability.

[0146] It should be understood that the various forms of processes shown above can be used, with steps reordered, added, or deleted. For example, the steps described in this invention can be executed in parallel, sequentially, or in different orders, as long as the desired result of the technical solution of this invention can be achieved, and this is not limited herein.

[0147] The specific embodiments described above do not constitute a limitation on the scope of protection of the present invention. Those skilled in the art should understand that various modifications, combinations, sub-combinations, and substitutions can be made according to design requirements and other factors. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for predicting electricity load across multiple industries, characterized in that, The method includes: Obtain power load data from at least one data source, and determine the target industry sector to which the power load data belongs based on the data characteristics of the power load data; Obtain the load probability prediction model corresponding to the target industry sector, wherein the load probability prediction model includes at least a conformal quantile regression model; The load prediction range of the power load data is determined based on the load probability prediction model.

2. The method according to claim 1, characterized in that, The determination of the target industry sector to which the power load data belongs based on the data characteristics of the power load data includes: The temporal fluctuation characteristics, industry-driven characteristics, and climate sensitivity characteristics of the power load data are statistically analyzed as the data features. Obtain domain template features, match the data features with the domain template features, and determine the target industry domain to which the power load data belongs; The domain template features include at least: industrial domain template features, commercial domain template features, and green energy domain template features.

3. The method according to claim 1, characterized in that, Also includes: Obtain power load data from at least one data source, and randomly divide the power load data into a training set, a calibration set, and a test set, wherein there is no overlap between the training set, the calibration set, and the test set; Obtain the preset false coverage rate of the load probability prediction model, take half of the preset false coverage rate as the low quantile of the load probability prediction model, and take the complementary value of the half value as the high quantile of the load probability prediction model. Obtain the loss function of the load probability prediction model, input the training set into the load probability prediction model, minimize the loss function through the stochastic gradient descent algorithm, combine the low quantile and the high quantile, and output the first prediction interval of the training dataset. The first prediction interval includes at least the upper limit of the first prediction interval and the lower limit of the first prediction interval. When the rate of change of the loss function converges, the training of the load probability prediction model is considered complete. The load probability prediction model is calibrated based on the calibration set and validated based on the test set.

4. The method according to claim 3, characterized in that, The loss function includes at least: Among them, y i Let i be the actual load value at the i-th time point. The predicted load value is the actual load value at the i-th time point. The predicted load value Compared with the actual load value y i The absolute difference between them, C i The predicted interval for the actual load value at the i-th time point is... The indicator function is of the form: τ is the target quantile of the quantile regression loss function. Let quantile regression loss function be the form of quantile regression loss function.

5. The method according to claim 3, characterized in that, The calibration of the load probability prediction model based on the calibration set includes: Obtain a calibration set of the power load data, input the calibration set into the trained load probability prediction model, and output a second prediction interval of the calibration set, wherein the second prediction interval includes at least an upper limit and a lower limit of the second prediction interval. Obtain the predicted load value corresponding to the calibration set, and the first false coverage rate of the predicted load value not falling within the second prediction interval; If the first false coverage rate is not equal to the preset false coverage rate, then compare the second prediction interval with the first prediction interval to obtain the boundary offset between the second prediction interval and the first prediction interval; Wherein, the upper limit offset of the boundary offset is the difference between the upper limit of the second prediction interval and the upper limit of the first prediction interval, and the lower limit offset of the boundary offset is the difference between the lower limit of the second prediction interval and the lower limit of the first prediction interval.

6. The method according to claim 3, characterized in that, The validation of the load probability prediction model based on the test set includes: A test set of the power load data is obtained, the test set is input into the trained load probability prediction model, and a third prediction interval of the test set is output. The third prediction interval includes at least an upper limit and a lower limit of the third prediction interval. Obtain the upper limit offset and lower limit offset corresponding to the calibration set, adjust the upper limit of the third prediction interval according to the upper limit offset, and use the adjusted upper limit of the third prediction interval as the final upper limit of the third prediction interval. The lower limit of the third prediction interval is adjusted according to the lower limit offset, and the adjusted lower limit of the third prediction interval is used as the final lower limit of the third prediction interval. Obtain the predicted load value corresponding to the test set, and the second false coverage rate where the predicted load value does not fall within the final third prediction interval; If the second false coverage rate is equal to the preset false coverage rate, then the load probability prediction model is determined to have passed the verification; otherwise, the load probability prediction model is retrained and calibrated.

7. A multi-industry power load forecasting device, characterized in that, The device includes: The domain determination module is used to acquire power load data from at least one data source and determine the target industry domain to which the power load data belongs based on the data characteristics of the power load data. The model acquisition module is used to acquire the load probability prediction model corresponding to the target industry sector, and the load probability prediction model includes at least a conformal quantile regression model. The interval determination module is used to determine the load prediction interval of the power load data based on the load probability prediction model.

8. A device, characterized in that, The device includes: At least one processor; and A memory communicatively connected to the at least one processor; wherein, The memory stores a computer program that can be executed by the at least one processor, the computer program being executed by the at least one processor to enable the at least one processor to perform the multi-industry power load forecasting method according to any one of claims 1-6.

9. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer instructions that cause a processor to execute the scenario-based multi-industry power load forecasting method according to any one of claims 1-6.

10. A computer program product, characterized in that, The computer program product includes a computer program that, when executed by a processor, implements the multi-industry power load forecasting method according to any one of claims 1-6.

Images