Power grid load prediction method and device and related product

Abstract

The invention relates to the technical field of power systems, and discloses a power grid load prediction method and device and a related product. The method comprises the following steps: acquiring a first single-region load prediction result reported by a plurality of sub-regions in a to-be-predicted region; obtaining an index value set corresponding to each sub-region; according to each index value set, calculating to obtain weight data of each sub-region; and according to each first single-region load prediction result and each piece of weight data, calculating to obtain an overall load prediction result of the to-be-predicted region. According to the invention, the index value can reflect the characteristics of load prediction quality, influence degree, difficulty and the like of each sub-region. Therefore, the weight data of each sub-region is calculated according to the first single-region load prediction result and the index value set, and the proportion of the weight data in the overall load prediction can be reasonably distributed. And finally, combining the load prediction result of the first single area with the weight, so that the overall load prediction result of the to-be-predicted area can be calculated more accurately and reliably.

Description

Technical Field

[0001] The present disclosure relates to the technical field of power systems, and in particular, to a power grid load forecasting method, apparatus, and related products. Background Art

[0002] In the field of power systems, power load forecasting is an important basis for scientific decision-making by power dispatching centers. Whether it is the overall load forecasting of provincial and municipal regions under the jurisdiction of the network-level dispatching center or the overall load forecasting of municipal regions under the jurisdiction of the provincial dispatching center, the results directly affect important links such as energy distribution strategies and supply plan formulation. Therefore, it is of great significance to research an efficient power grid load forecasting method.

[0003] Currently, for the problem of forecasting the overall load of multiple subordinate regions, the cumulative method is often used, that is, directly adding the load forecasting results of each subordinate region. However, this method does not consider the characteristics of each subordinate region itself, ignores the importance of the load forecasting results of individual regions, and makes it difficult for the final forecasting result to accurately reflect the actual load situation. Therefore, it is urgent to solve this technical problem. Summary of the Invention

[0004] In view of the above situation, embodiments of the present disclosure provide a power grid load forecasting method, apparatus, and related products, aiming to solve the above problems or at least partially solve the above problems.

[0005] In a first aspect, embodiments of the present disclosure provide a power grid load forecasting method, the method comprising:

[0006] Obtaining first single-region load forecasting results reported by a plurality of sub-regions within a region to be forecast; the first single-region load forecasting results are output by the sub-regions using their trained first load forecasting models;

[0007] Obtaining a set of index values respectively corresponding to each of the sub-regions; the set of index values includes index values of at least one of the following indexes: prediction result credibility index, sub-region load influence degree index, and sub-region load forecasting difficulty index;

[0008] Calculating weight data for each sub-region according to each set of index values;

[0009] Calculating an overall load forecasting result of the region to be forecast according to each of the first single-region load forecasting results and each of the weight data.

[0010] In a second aspect, embodiments of the present disclosure further provide a power grid load forecasting apparatus, the apparatus comprising:

[0011] A first acquisition module, configured to acquire first single-region load prediction results reported by multiple sub-regions within a region to be predicted; the first single-region load prediction results are output by the sub-regions using their trained first load prediction models.

[0012] A second acquisition module, configured to acquire a set of index values corresponding to each of the sub-regions; the set of index values includes index values of at least one of the following indexes: prediction result credibility index, sub-region load influence degree index, and sub-region load prediction difficulty index.

[0013] A calculation module, configured to calculate weight data for each sub-region according to each set of index values.

[0014] A prediction model, configured to calculate an overall load prediction result of the region to be predicted according to each of the first single-region load prediction results and each of the weight data.

[0015] In a third aspect, an embodiment of the present disclosure further provides an electronic device, including: a processor; and a memory arranged to store computer-executable instructions, and the executable instructions, when executed, cause the processor to execute the steps of the above grid load prediction method.

[0016] By means of the above technical solutions, the grid load prediction method, device, and related products provided by the embodiments of the present disclosure first acquire first single-region load prediction results reported by multiple sub-regions within a region to be predicted, and at least one of the prediction result credibility index value, sub-region load influence degree index value, and sub-region load prediction difficulty index value corresponding to each sub-region. These index values can reflect multiple characteristics of each sub-region, such as load prediction quality, influence degree, and load prediction difficulty. Then, weight data for each sub-region is calculated based on the first single-region load prediction results and the set of index values, so that the proportion of the first single-region load prediction results of each sub-region in the overall load prediction is reasonably allocated, reflecting the importance differences of different sub-regions. Finally, by combining each first single-region load prediction result and the weight data, a more accurate and reliable overall load prediction result can be calculated for the region to be predicted.

[0017] The above description is only an overview of the technical solutions of the present disclosure. In order to be able to understand the technical means of the present disclosure more clearly, it can be implemented according to the content of the description. And in order to make the above and other purposes, features, and advantages of the present disclosure more obvious and understandable, the specific embodiments of the present disclosure are specifically given below. Description of the Drawings

[0018] The drawings described herein are used to provide a further understanding of the present disclosure, and constitute a part of the present disclosure. The illustrative embodiments and descriptions thereof of the present disclosure are used to explain the present disclosure and do not constitute an improper limitation of the present disclosure. In the drawings:

[0019] Figure 1 shows a schematic flow chart of the power grid load forecasting method provided by an embodiment of the present disclosure;

[0020] Figure 2 shows a flow chart of constructing an evaluation index system provided by an embodiment of the present disclosure;

[0021] Figure 3 shows the model architecture of the i-Transformer provided by an embodiment of the present disclosure;

[0022] Figure 4 shows a schematic structural diagram of the power grid load forecasting device provided by an embodiment of the present disclosure;

[0023] Figure 5 shows a schematic structural diagram of an electronic device provided by an embodiment of the present disclosure. Detailed implementation manners

[0024] To make the objectives, technical solutions, and advantages of the present disclosure clearer, the technical solutions of the present disclosure will be clearly and completely described below in conjunction with specific embodiments of the present disclosure and the corresponding drawings. Apparently, the described embodiments are only a part of the embodiments of the present disclosure, rather than all of the embodiments. Based on the embodiments in the present disclosure, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present disclosure.

[0025] It should be noted that similar reference numerals and letters denote similar items in the following drawings. Therefore, once an item is defined in one drawing, it does not need to be further defined and explained in subsequent drawings.

[0026] It should be noted that the terms "first", "second", etc. in the specification and claims of the present disclosure and the above-mentioned drawings are used to distinguish similar objects, and do not necessarily need to be used to describe a specific order or sequence. It should be understood that such use can be interchanged under appropriate circumstances so that the embodiments of the present disclosure described herein can be implemented in an order other than those illustrated or described herein. In addition, the term "including" and its variants should be interpreted as open-ended terms meaning "including but not limited to".

[0027] As introduced above, for the problem of predicting the overall load of a whole containing multiple subordinate regions, the prior art often adopts the addition method, that is, directly adding the load forecasting results of each subordinate region. However, this method does not consider the characteristics of each subordinate region itself, ignores the importance of the load forecasting result of a single region, and makes the final forecasting result difficult to accurately reflect the actual load situation. Based on this, the present invention proposes a power grid load forecasting method, device, and related products, and the present disclosure will be described in detail below through specific embodiments.

[0028] For ease of understanding of this embodiment, first, a power grid load forecasting method disclosed in the embodiments of the present disclosure will be introduced in detail. The execution subject of the power grid load forecasting method provided in the embodiments of the present disclosure is generally a computer device with certain computing capabilities, and such a computer device includes, for example: a terminal device, a server, or other processing devices. The terminal device may be a user equipment (UE), a mobile device, a user terminal, a terminal, etc. In some possible implementation manners, the power grid load forecasting method may be implemented by a processor invoking computer-readable instructions stored in a memory.

[0029] Figure 1 The flowchart of the power grid load forecasting method provided in the embodiments of the present disclosure is shown. From Figure 1 it can be seen that the embodiments of the present disclosure at least include steps S101 - S104:

[0030] S101: Obtain the first single-region load forecasting results reported by multiple sub-regions within the region to be forecast; the first single-region load forecasting results are output by the sub-regions using their trained first load forecasting models.

[0031] In this embodiment, there are no strict limitations on the attributes such as the level of the sub-region. For example, a sub-region may be a district in a city, a city, or a provincial administrative unit.

[0032] The first single-region load forecasting result refers to the result obtained by each sub-region within the region to be forecast using its specific first load forecasting model to forecast the power load situation within a specific time range. The first single-region load forecasting result includes the power load forecasting values of the corresponding sub-region at one or more future moments or time periods. Exemplarily, the first single-region load forecasting result includes 24 load forecasting values within a future day (forecast once every 1 hour).

[0033] In this embodiment, there are no limitations on the means for obtaining the first single-region load forecasting results. During implementation, for example, the execution subject of this embodiment may remotely obtain the first single-region load forecasting results of each sub-region through a network interface from a central database or a distributed data node.

[0034] S102: Obtain the set of index values corresponding to each of the sub-regions; the set of index values includes the index values of at least one of the following indexes: the forecasting result credibility index, the sub-region load influence degree index, and the sub-region load forecasting difficulty index.

[0035] Here, the prediction result credibility index is used to measure the reliability of the first single-region load prediction result of the sub-region. The sub-region load influence index is used to measure the degree of influence of the sub-region load on the region to be predicted. The sub-region load prediction difficulty index is used to measure the difficulty of predicting the load of the sub-region.

[0036] In implementation, for example, power field experts can be organized. Based on their understanding of the prediction model, data quality, and prediction process, combined with historical experience, they score each index of each sub-region to obtain the corresponding index value set and send it to the execution entity of this embodiment. Exemplarily, the index value set corresponding to a sub-region includes: 0.7, 0.5, 0.2. Among them, 0.7 is the prediction result credibility index value, 0.5 is the sub-region load influence index value, and 0.2 is the sub-region load prediction difficulty index value.

[0037] S103: Calculate the weight data of each sub-region according to each of the index value sets.

[0038] In implementation, for example, if each index value set includes three index values, the index values can be first standardized; then the subjective weighting method, objective weighting method, or combined weighting method can be used to determine the weights of these three indexes respectively; the standardized index values are multiplied by the corresponding index weights and then added together to obtain the comprehensive score of each sub-region; the comprehensive scores of each sub-region are normalized to obtain the weight data of each sub-region.

[0039] S104: Calculate the overall load prediction result of the region to be predicted according to each of the first single-region load prediction results and each of the weight data.

[0040] In specific implementation, the weight data is used to perform weighted average calculation on each first single-region load prediction result to obtain the overall load prediction result.

[0041] It can be seen that in the embodiment of the present disclosure, first, the first single-region load prediction results reported by multiple sub-regions within the region to be predicted are obtained, and at least one of the prediction result credibility index value, sub-region load influence index value, and sub-region load prediction difficulty index value corresponding to each sub-region. These index values can reflect various characteristics of each sub-region such as the load prediction quality, influence degree, and load prediction difficulty. Then, according to the first single-region load prediction results and the index value sets, the weight data of each sub-region is calculated, so that the proportion of the first single-region load prediction results of each sub-region in the overall load prediction is reasonably allocated, reflecting the importance differences of different sub-regions. Finally, combined with each first single-region load prediction result and the weight data, a more accurate and reliable overall load prediction result can be calculated for the region to be predicted.

[0042] Further, to better illustrate the process of the above grid load forecasting method, as a refinement and extension of the above embodiments, the embodiments of the present invention provide several embodiments, but are not limited thereto, as follows.

[0043] For the foregoing set of index values, the embodiments of the present disclosure also provide a method for determining index values, by constructing an index system, so as to realize the quantification of the prediction result credibility index, the sub-region load influence degree index, and the sub-region load prediction difficulty index. Figure 2 The flowchart showing the construction of the evaluation index system provided by the embodiments of the present disclosure is shown.

[0044] See Figure 2 As shown, the construction process of the evaluation system is divided into four main steps. The first is the purpose and requirement setting, where it is necessary to clarify the evaluation purpose, the use of the system, the evaluation scope and key points, and at the same time collect requirement information, including the expectations of the regional dispatching center and the background information data of the power industry. The second step is to determine the index content, covering the first-level index, that is, the highest-level index that needs to comprehensively cover the evaluation content; the second-level index, which is the decomposition of the first-level index and is the specific measurement dimension; the third-level index, which is the underlying specific index, and it is required to be directly quantifiable or calculable. The third step is to establish an evaluation model. First, assign weights to the indexes at all levels according to the importance of the indexes, and then perform data collection and processing, that is, preprocess according to the information of each province and city, and then construct an evaluation model based on the index system, and establish a scoring standard and a comparison benchmark. The last step is the trial evaluation and continuous monitoring. First, conduct a trial evaluation to detect the effectiveness, then revise and improve it. During the subsequent implementation and evaluation, regularly monitor the prediction fusion effect, and continuously monitor the evaluation results in the form of a report. Based on this, in some embodiments of the present disclosure, the index value of the prediction result credibility index is calculated according to the following method: using a preset index weight determination algorithm, process at least one of the obtained first historical prediction accuracy, real-time prediction accuracy, and prediction accuracy under extreme weather to calculate the historical prediction performance index value; using the index weight determination algorithm, process at least one of the obtained prediction model update frequency, prediction result continuity data, and abnormal prediction ratio to calculate the prediction model stability index value; using the index weight determination algorithm, process at least one of the obtained data integrity rate, data timeliness data, and data consistency data to calculate the data quality level index value of the data required for prediction; using the index weight determination algorithm, process at least one of the historical prediction performance index value, the prediction model stability index value, and the data quality level index value to calculate the index value of the prediction result credibility index.

[0045] In this embodiment, the preset algorithm for determining the index weight may be, for example, the entropy weight method, the standard deviation coefficient method, the CRITIC method, etc., and the embodiments of the present disclosure are not limited thereto. During implementation, the credibility index value of the prediction result is obtained by using the preset algorithm for determining the index weight to calculate the weight of at least one of the historical prediction performance index value, the prediction model stability index value, and the data quality level index value, and performing weighted averaging on each index value. Here, the historical prediction performance index is used to measure the accuracy of the load prediction of the prediction model in the past. The prediction model stability index is used to measure the stability of the prediction model. The data quality level index is used to measure the quality of the data input into the prediction model.

[0046] For the historical prediction performance index, during implementation, the weight of at least one of the first historical prediction accuracy rate, the real-time prediction accuracy rate, and the prediction accuracy rate under extreme weather is calculated by using the preset algorithm for determining the index weight, and the weighted average of each accuracy rate is obtained. Among them, the first historical prediction accuracy rate is used to measure the accuracy of the load prediction performed by the prediction model in the past. During implementation, for example, multiple sets of data sent by the prediction system can be received, each set of data including the load prediction value and the actual load value at a certain time point before the day. Then, the accuracy rate corresponding to each set of data is calculated through formula 1 "accuracy rate = 1 - |predicted value - actual value| / actual value × 100%", and finally the average is taken to obtain the first historical prediction accuracy rate. The real-time prediction accuracy rate is used to measure the accuracy of the real-time load prediction of the prediction model. During implementation, for example, the load prediction value and the actual value at the latest time point sent by the prediction system can be received in real time, and the real-time prediction accuracy rate is calculated through the aforementioned formula 1. The prediction accuracy rate under extreme weather is used to measure the prediction ability of the prediction model under extreme weather (such as heavy rain, high temperature, heavy snow, etc.). During implementation, multiple sets of data sent by the prediction system can be received, and each time the data is the load prediction value and the actual value of the prediction model when extreme weather occurs. The accuracy rate corresponding to each set of data is calculated using formula 1 and the average is taken to obtain the prediction accuracy rate under extreme weather.

[0047] For the stability index of the prediction model, during implementation, a preset algorithm for determining index weights is used to calculate the weights of at least one of the obtained prediction model update frequency, prediction result continuity, and abnormal prediction ratio, and the weighted average of each index value is obtained. Among them, the prediction model update frequency refers to the number of times the prediction model is updated within a period of time (such as one day or one week). The prediction result continuity data is used to measure the smoothness of the change between the load prediction results at adjacent time points and can reflect the fluctuation of the prediction results of the prediction model. During implementation, for example, the continuity can be quantified by calculating statistics such as the standard deviation of the differences between adjacent prediction results within a period of time. The abnormal prediction ratio refers to the proportion of abnormal prediction results among all prediction results within a period of time. During implementation, for example, the judgment criteria for abnormal predictions can be customized first (such as a deviation from the actual value exceeding a certain threshold), and then the number of abnormal prediction results within a period of time is counted and divided by the total number of prediction results to obtain the abnormal prediction ratio.

[0048] For the data quality level index, during implementation, a preset algorithm for determining index weights is used to calculate the weights of at least one of the obtained data completeness rate, data timeliness data, and data consistency data, and the weighted average of each index value is obtained. Among them, for the data completeness rate (DCR), during implementation, it can be represented by the ratio of the total number of obtained valid data points to the total number of data points that should be collected. Here, the evaluation criteria for valid data points can be set according to actual needs and will not be elaborated here. For the data timeliness data (DT), during implementation, for example, it can be calculated according to the following formula:

[0049]

[0050] where Δt is the actual data delay time and τ is the maximum delay threshold allowed by the system. The DT value range is [0, 1], and the larger the value, the better the data timeliness.

[0051] For the data consistency data (DC), during implementation, for example, it can be calculated according to the following formula:

[0052] DC = DC format ×DC logic

[0053]

[0054] where n is the number of samples, N gormat is the number of data points with inconsistent formats, N structure is the number of data points with inconsistent structures, N sampling is the number of data points with inconsistent samplings. xi is the current data point, μ i is the historical data mean, σ i is the historical data standard deviation.

[0055] In some embodiments, the index value of the sub-region load influence degree index is calculated according to the following method: using a preset index weight determination algorithm, processing at least one of the obtained maximum load ratio, average load ratio, and peak load contribution rate to calculate the load scale ratio index value; using the index weight determination algorithm, processing at least one of the obtained peak shaving capacity ratio, new energy installed capacity ratio, and demand response ability data to calculate the regulation ability index value; using the index weight determination algorithm, processing at least one of the load scale ratio index value and the regulation ability index value to calculate the index value of the sub-region load influence degree index.

[0056] During implementation, the index value of the sub-region load influence degree index is obtained by using a preset index weight determination algorithm to calculate the weights of at least one of the load scale ratio index value and the regulation ability index value, and using each weight to perform weighted averaging on each index value. Here, the load scale ratio index refers to the ratio of the load scale of a certain sub-region to the total load scale of the region to be predicted. The regulation ability index is a quantitative index used to measure the ability of a sub-region to adjust and balance load changes, power output fluctuations, etc. in the power system.

[0057] For the load scale ratio index value, during implementation, a preset index weight determination algorithm is used to calculate the weights of at least one of the maximum load ratio, average load ratio, and peak load contribution rate, and using each weight to perform weighted averaging on each index value. Among them, the maximum load ratio refers to the ratio of the maximum load of a certain sub-region to the total maximum load of the region to be predicted within a specific time period. The average load ratio refers to the ratio of the average load of the sub-region to the total average load of the region to be predicted within a specific time period. The peak load contribution rate refers to the ratio of the peak load amount of the sub-region to the total peak load amount of the region to be predicted. During implementation, the determination criteria for peak load (such as the load exceeding a certain threshold and lasting for a short time) can be determined first, and based on this criterion, the peak load of each sub-region can be found, and then the peak load contribution rate of each sub-region can be calculated through the formula "peak load contribution rate = sub-region peak load amount / total peak load amount × 100%".

[0058] For the regulation ability index value, during implementation, an index weight determination algorithm is used to calculate the weights of at least one of the peak shaving capacity ratio, new energy installed capacity ratio, and demand response ability data, and using the weights to perform weighted averaging on each index value.

[0059] Among them, the peak shaving capacity ratio refers to the ratio of the capacity available for peak shaving in the sub-region power grid to the capacity available for peak shaving in the area to be predicted. The new energy installed capacity ratio refers to the ratio of the new energy installed capacity in the sub-region to the new energy installed capacity in the area to be predicted. The demand response ability data refers to the ratio of the adjustable load of users in the sub-region to the total adjustable load in the area to be predicted, reflecting the potential of the sub-region to assist the power grid regulation through demand-side management.

[0060] In some embodiments, the index value of the sub-region load prediction difficulty index is calculated according to the following method:

[0061] Using a preset index weight determination algorithm, at least one of the daily load volatility, weekly load volatility, and seasonal fluctuation intensity obtained is processed to calculate the load fluctuation characteristic index value;

[0062] Using the index weight determination algorithm, at least one of the temperature sensitivity, humidity sensitivity, and holiday sensitivity obtained is processed to calculate the external factor sensitivity index value;

[0063] Using the index weight determination algorithm, at least one of the industrial electricity consumption ratio, number of key users, and number of user types obtained is processed to calculate the electricity consumption structure complexity index value;

[0064] Using the index weight determination algorithm, at least one of the load fluctuation characteristic index value, the external factor sensitivity index value, and the electricity consumption structure complexity index value is processed to obtain the index value of the sub-region load prediction difficulty index.

[0065] During implementation, the sub-region load prediction difficulty index value is calculated by using a preset index weight determination algorithm to calculate the weights of at least one of the load fluctuation characteristic index value, the external factor sensitivity index value, and the electricity consumption structure complexity index value, and using each weight to perform weighted averaging on each index value. Here, the load fluctuation characteristic index is used to measure the severity of the sub-region load changing over time. The external factor sensitivity index is used to measure the degree to which the sub-region load is affected by various external factors. The electricity consumption structure complexity index is used to measure the composition of different types of electricity users in the sub-region and the complexity of their electricity consumption characteristics.

[0066] For the load fluctuation characteristic index value, during implementation, a preset index weight determination algorithm is used to calculate the weights of at least one of the daily load volatility, weekly load volatility, and seasonal fluctuation intensity, and using each weight to perform weighted averaging on each index value.

[0067] Among them, the daily load volatility is used to measure the fluctuation of the power load in a sub-region within a day. During implementation, for example, it can be calculated as the ratio of the difference between the daily maximum load and the daily minimum load in the sub-region to the daily average load. The weekly load volatility is used to measure the power load fluctuation in a sub-region within a week. During implementation, for example, the weekly load volatility of a sub-region can be calculated using the following formula:

[0068]

[0069] Among them, L i represents the average load of the sub-region on the i-th day, and represents the average daily load of the sub-region within this week.

[0070] The seasonal fluctuation intensity is used to measure the degree of change in the sub-region load between different seasons, and can reflect the fluctuation characteristics of the power load affected by seasonal factors. During implementation, for example, the seasonal fluctuation intensity of a sub-region can be calculated using the following formula:

[0071]

[0072] Among them, SI represents the seasonal fluctuation intensity, and L i represents the average load of the sub-region within the i-th season, n represents the number of seasons, and represents the annual average load of the sub-region.

[0073] For the external factor sensitivity index value, during implementation, the weight determination algorithm for the index is used to calculate the weight of at least one of the temperature sensitivity, humidity sensitivity, and holiday sensitivity, and the obtained weights are used to perform a weighted average on each index value.

[0074] Among them, the temperature sensitivity is used to measure the degree of influence of temperature factors on the power load in the sub-region. During implementation, for example, it can be represented by the ratio of the change in power load caused by a unit temperature change to the average load. The humidity sensitivity is used to measure the degree of influence of humidity factors on the power load in the sub-region. During implementation, for example, it can be represented by the ratio of the change in power load caused by a unit humidity change to the average load. The holiday sensitivity is used to measure the degree of influence of holidays on the power load in the sub-region. During implementation, for example, the holiday sensitivity of the sub-region can be calculated using the following formula:

[0075]

[0076] Among them, L 节假日 represents the average load of the sub-region during the holiday period, and L 非节假日 represents the average load of the sub-region during the non-holiday period with a time length similar to that before and after the holiday.

[0077] For the index value of the complexity of the electricity consumption structure, during implementation, an index weight determination algorithm is used to calculate the weight of at least one of the industrial electricity consumption ratio, the number of key users, and the number of user types, and the weight of each index value is obtained by weighted averaging using each weight.

[0078] Among them, the industrial electricity consumption ratio is used to measure the important position and scale of industrial electricity consumption in the entire electricity consumption structure within a sub-region, and can be represented by the proportion of electricity consumption in the industrial field in the total electricity consumption. The number of key users refers to the number of key users in the power system that have an important impact on the stability and reliability of power supply, have a relatively large electricity consumption scale, or have special electricity consumption characteristics. Key users include, but are not limited to, large industrial enterprises, important public service institutions, etc. The number of user types is used to measure the richness and difference degree of user types in the power system, and considers the composition of various users with different industries, natures, and electricity consumption demands in electricity consumption. The more user types and the greater the difference, the more complex the electricity consumption structure, and the difficulty of system operation and management may also increase accordingly.

[0079] In some embodiments, the method further includes:

[0080] Obtain the load forecasting influencing factor data of each sub-region, and preprocess each load forecasting influencing factor data to obtain the corresponding processed influencing factor data;

[0081] For any target sub-region, use the corresponding second load forecasting model to predict the corresponding target processed influencing factor data to obtain the corresponding second single-region load forecasting result;

[0082] The calculating the overall load forecasting result of the region to be forecasted according to each of the first single-region load forecasting results and each of the weight data includes:

[0083] For any sub-region, use the corresponding weight data to perform weighted averaging on the corresponding first single-region load forecasting result and the second single-region load forecasting result to obtain the corresponding integrated load forecasting result;

[0084] Add up each of the integrated load forecasting results to obtain the overall load forecasting result.

[0085] In this embodiment, the load prediction influencing factor data of each sub-region includes, but is not limited to, at least one of the following: power grid operation data (such as load data, industrial electricity consumption data, distributed energy generation situation data, peak-valley difference, average load, maximum load, minimum load, load curve trend data, etc.), climate data (such as temperature, humidity, precipitation, wind speed, etc.), social factor data (such as holidays, population quantity, real-time electricity price, electricity consumption behavior preference, industrial structure, etc.). For the acquisition means of each load prediction influencing factor data, for example, it can be obtained through a Supervisory Control And Data Acquisition (SCADA) system, an open-source data platform, etc. This embodiment does not make any limitations in this regard. After obtaining the load prediction influencing factor data of each sub-region, for the convenience of subsequent efficient data processing, during implementation, the load prediction influencing factor data can be preprocessed first. The preprocessing means includes, but is not limited to, at least one of the following: data cleaning (such as outlier processing, missing value imputation), feature engineering (such as feature selection, feature encoding), etc.

[0086] Before predicting each processed influencing factor data, a corresponding second load prediction model is trained for each sub-region first.

[0087] During implementation, an initial model can be selected first. The initial model can be, for example, an ARIMA (Autoregressive Integrated Moving Average Model), Holt-Winters, LSTM (Long Short-Term Memory), GRU (Gated Recurrent Unit), Transformer, or other models. Then, the initial model is trained using a pre-constructed data set to obtain each second load prediction model.

[0088] In some embodiments, the second load prediction model corresponding to the target sub-region is trained based on the i-transformer model.

[0089] The i-Transformer model is a generative artificial intelligence model. It reverses the module order in the Transformer architecture. After inputting data, it can map the entire time series of the same feature variable into a high-dimensional vector, obtaining the main body of the time series described by the high-dimensional feature vector, which can independently reflect the changes in the historical sequence.

[0090] Figure 3 Shows the model architecture of the i-Transformer provided by the embodiments of the present disclosure. Figure 3It can be seen that the input data is first converted into vectors by the embedding layer and then enters the conversion module composed of a multi-head attention layer, an addition and normalization layer, and a feed-forward layer. This module can be repeated L times to gradually extract complex features. The multi-head attention layer calculates Q, K, and V vectors to capture element dependencies. The addition and normalization layer contains residual connections and layer normalization to assist training. The feed-forward layer performs non-linear transformations to enhance the expressive power. After that, it is converted by the projection layer, and finally the output result is obtained. Table 1 below shows the main parameters of the i-Transformer model.

[0091] Table 1 Main parameters of the i-Transformer model

[0092] Code Name Chinese Name Parameter Meaning Num of attention heads Number of attention heads Number of multi-attention heads Model dimension Model dimension Attention head dimension Hidden Layer Dimension Hidden layer dimension Hidden layer dimension size Feed-Forward Network Layers Feed-forward neural network Number of feed-forward neural network layers Num of Encoder Layers Number of encoder layers Number of encoder stacks in the model Num of Decoder Layers Number of decoder layers Number of decoder stacks in the model Loss Function Loss function Optimization objective during training Optimizer Optimizer Optimization algorithm for training the model Learning Rate Learning rate Compensation for weight update Batch Size Batch size Number of samples used for training Activation Function Activation function Output mapping inside the neuron

[0093] In this embodiment, the i-Transformer model is selected as the basic architecture of the second load prediction model. Through its unique sequence perspective and inverted structure design, the i-Transformer model can effectively solve problems such as noise interference in traditional models for multi-variable time series and difficulties in modeling lag relationships between variables, and significantly improve the load prediction accuracy. Its attention mechanism can accurately capture the dynamic associations between the load and external factors in multi-source heterogeneous data, and the feed-forward network strengthens the extraction of time-dimensional features. While reducing the dependence on long historical data, it enables fast prediction through parallelization, especially suitable for multi-region overall load prediction scenarios with transmission delays and high variable endogenous complexity.

[0094] In some embodiments, the second load prediction model corresponding to the target sub-region is generated according to the following method: Datasets under various power grid operation scenarios are obtained respectively; the power grid operation scenarios are divided according to at least one dimension such as whether it is a holiday or the season type; for any target power grid operation scenario, using the corresponding dataset, at least two models among ARIMA, Holt-Winters, LSTM, GRU, Transformer, and i-Transformer are trained respectively to obtain multiple candidate load prediction models; the prediction model with the best performance among the multiple candidate load prediction models is determined as the load prediction model under the target power grid operation scenario; according to the power grid operation scenario information of the target load prediction influencing factor data, the power grid operation scenario is matched, and the load prediction model corresponding to the matched power grid operation scenario is used as the second load prediction model corresponding to the target sub-region.

[0095] In this embodiment, the power grid operation scenarios are obtained by dividing according to at least one of the following dimensions: whether it is a holiday and the season type. Exemplarily, by dividing according to the two dimensions of whether it is a holiday (such as holidays and weekdays) and the season type (such as spring, summer, autumn, winter), the following 8 power grid operation scenarios can be obtained: spring holidays, spring weekdays, summer holidays, summer weekdays, autumn holidays, autumn weekdays, winter holidays, and winter weekdays.

[0096] During implementation, first obtain the data set for each power grid operation scenario. It can be understood that the types and acquisition methods of the original data corresponding to the feature data of the samples in the data set are similar to those of the aforementioned load prediction influencing factor data, and will not be elaborated here.

[0097] Then, for each power grid operation scenario, at least two of ARIMA, Holt-Winters, LSTM, GRU, Transformer, and i-Transformer can be trained using the corresponding data set. For example, ARIMA, Holt-Winters, LSTM, GRU, Transformer, and i-Transformer can be trained respectively to obtain 8 candidate load prediction models. The specific training method is prior art and will not be elaborated here.

[0098] Next, for any power grid operation scenario, determine the prediction model with the best performance among its corresponding multiple candidate load prediction models as the corresponding load prediction model. This embodiment does not limit the method for determining the optimal performance model. For example, the mean squared error of each candidate load prediction model on the corresponding validation set can be calculated respectively, and the model with the smallest mean squared error is the prediction model with the best performance.

[0099] Finally, according to the power grid operation scenario information of the target load prediction influencing factor data, match the power grid operation scenario, and use the load prediction model corresponding to the matched power grid operation scenario as the second load prediction model for the target sub-region. Exemplarily, assuming that the power grid operation scenario information corresponding to the target load prediction influencing factor data is "spring during the May 1st Labor Day holiday", then the prediction model with the optimal performance pre-trained for the "spring holidays" power grid operation scenario can be used as the second load prediction model for the target sub-region.

[0100] In this embodiment, the power grid operation scenarios are divided according to dimensions such as whether it is a holiday or a season type, and the corresponding data sets are obtained. Then, for each target power grid operation scenario, various models (such as ARIMA, LSTM, etc.) are trained using the corresponding data sets to obtain multiple candidate load forecasting models. In this way, the advantages of different models are fully utilized, and the model potential suitable for each scenario is explored. Then, the one with the best performance is selected from multiple candidate load forecasting models as the load forecasting model for the target power grid operation scenario, ensuring that each scenario has the most suitable model and improving the prediction accuracy of the model in a specific scenario. Finally, according to the power grid operation scenario information of the target load forecasting influencing factor data, the scenario is matched and the second load forecasting model corresponding to the target sub-region is determined, so that the forecasting model can accurately correspond to the actual scenario, improving the pertinence and applicability of the model. In summary, this embodiment can achieve accurate matching of the optimal load forecasting model according to different power grid operation scenarios, significantly improving the accuracy and reliability of load forecasting.

[0101] After training the second load forecasting model for each sub-region, the processed influencing factor data can be input into the corresponding second load forecasting model respectively in the way of a sliding window, and the model outputs the second single-region load forecasting results.

[0102] Then, for any sub-region, the corresponding first single-region load forecasting result and the second single-region load forecasting result are weighted and averaged using the corresponding weight data to obtain the corresponding fused load forecasting result. Finally, the fused load forecasting results are added together to obtain the overall load forecasting result.

[0103] Exemplarily, the overall load forecasting result can be calculated according to the following formula:

[0104]

[0105] where Load final,t represents the overall load forecasting result of the area to be predicted at time t, n represents the total number of sub-regions, w i represents the weight data corresponding to the i-th province (sub-region), Load province,i,t represents the first single-region load forecasting result of the i-th province (sub-region) at time t, and Load main,i,t represents the second single-region load forecasting result of the i-th province (sub-region) at time t by the technical solution of this embodiment. If the index value used to calculate the weight data of the sub-region changes with time, then w i,t can be used to replace w i , and w i,t represents the weight data corresponding to the i-th province (sub-region) at time t.

[0106] The execution entity of this embodiment uses the second load prediction model deployed by itself to predict the processed data of each sub-region, and obtains the second single-region load prediction result. Then, for each sub-region, the single-region load prediction results from two sources are weighted and averaged according to the weight data to obtain the fused load prediction result. Finally, the fused results of each sub-region are added together to obtain the overall load prediction result. It can be seen that this embodiment can comprehensively consider the advantages of the sub-region's own model and the model of this embodiment, reduce the prediction error of a single model, and improve the accuracy, reliability, and scientific nature of the overall load prediction of the region to be predicted. Moreover, the first single-region load results of each sub-region contain more detailed regional characteristics, which can enhance the interpretability of the overall load prediction result to a certain extent.

[0107] In some embodiments, the method further includes: obtaining the second historical prediction accuracy of each of the first load prediction models within a preset time period, multiplying each of the second historical prediction accuracies by the corresponding weight data to obtain a plurality of products; calculating a dynamic adjustment coefficient according to the plurality of products; and for any sub-region, using the corresponding weight data to perform weighted averaging on the corresponding first single-region load prediction result and the second single-region load prediction result to obtain the corresponding fused load prediction result, including: for any sub-region, using the corresponding weight data and the dynamic adjustment coefficient to perform weighted averaging on the corresponding first single-region load prediction result and the second single-region load prediction result to obtain the fused load prediction result.

[0108] In this embodiment, the preset time period can be set according to actual needs, and this embodiment does not limit it. Exemplarily, the preset time period is the most recent 3 months. The calculation method of the second historical prediction accuracy is similar to that of the foregoing first historical prediction accuracy, and will not be elaborated here.

[0109] During implementation, each of the second historical prediction accuracies can be multiplied by the corresponding weight data to obtain a plurality of products, and then the products are added together to obtain the dynamic adjustment coefficient.

[0110] During implementation, a basic adjustment coefficient (which can be used to adjust the influence of the two single-region load prediction results) can also be introduced, multiplying the dynamic adjustment coefficient by the basic adjustment coefficient, and taking the product as the new dynamic adjustment coefficient. The value range of the basic adjustment coefficient can be 0.4 - 0.6.

[0111] During implementation, if the index value used to calculate the weight data of the sub-region and the time resolution of the dynamic adjustment coefficient are consistent with the time resolution of the single-region load prediction result, then w i,t can be used to replace w i, at this time, the second historical prediction accuracy rate can be the average value of all prediction accuracy rates at the same moment within a preset time period. Specifically, the dynamic adjustment coefficient can be calculated according to the following formula:

[0112]

[0113] where β is the basic adjustment coefficient, and accuracy i,t is the historical prediction accuracy rate of the i-th province (sub-region) at time t (such as the average value of the accuracy rates at the same moment in the recent 3 months), and w i,t is the weight data of the i-th province (sub-region) at time t.

[0114] After obtaining the dynamic adjustment coefficient, for any sub-region, the corresponding weight data and dynamic adjustment coefficient can be used to perform weighted averaging on the corresponding first single-region load prediction result and the second single-region load prediction result to obtain the fused load prediction result.

[0115] Specifically in implementation, the overall load prediction result of the area to be predicted at time t can be calculated according to the following formula:

[0116]

[0117] The overall load prediction result of the area to be predicted at time t can also be calculated according to the following formula:

[0118]

[0119] In this embodiment, by obtaining the second historical prediction accuracy rates of each first load prediction model over a period of time in history, the performance of the model in a specific time period is evaluated. Then, the accuracy rate is multiplied by the weight of the first load prediction result of the corresponding sub-region, and the dynamic adjustment coefficient is calculated based on the product. For the sub-region, the weight and the dynamic adjustment coefficient that plays a secondary adjustment function are used to perform weighted averaging on the first and second single-region load prediction results, fusing different prediction advantages and dynamically adjusting the proportion. Finally, the fused results of each sub-region are added to obtain the overall load prediction result. This embodiment can effectively improve the accuracy of the overall load prediction and the flexibility in coping with load changes, model performance changes, and time changes.

[0120] This embodiment of the present disclosure also provides a power grid load prediction method, including the following steps:

[0121] Step S1: Obtain the first single-region load prediction results reported by multiple sub-regions within the area to be predicted; the first single-region load prediction result is output by the sub-region using its trained first load prediction model. Here, the area to be predicted is the main grid, and the sub-region is a provincial administrative unit.

[0122] Step S2: Obtain the set of index values corresponding to each sub-region; the set of index values includes the index values of the following indices: the prediction result credibility index, the sub-region load impact degree index, and the sub-region load prediction difficulty index.

[0123] During implementation, the index value of the prediction result credibility index is calculated according to the following method: Using the entropy weight method, process the obtained first historical prediction accuracy, real-time prediction accuracy, and prediction accuracy under extreme weather conditions to calculate the historical prediction performance index value; Using the entropy weight method, process the obtained prediction model update frequency, prediction result continuity data, and abnormal prediction ratio to calculate the prediction model stability index value; Using the entropy weight method, process the obtained data integrity rate, data timeliness data, and data consistency data to calculate the data quality level index value of the data required for prediction; Using the entropy weight method, process the historical prediction performance index value, prediction model stability index value, and data quality level index value to calculate the index value of the prediction result credibility index.

[0124] The index value of the sub-region load impact degree index is calculated according to the following method: Using the entropy weight method, process the obtained maximum load ratio, average load ratio, and peak load contribution rate to calculate the load scale ratio index value; Using the entropy weight method, process the obtained peak shaving capacity ratio, new energy installed capacity ratio, and demand response ability data to calculate the regulation ability index value; Using the entropy weight method, process the load scale ratio index value and the regulation ability index value to calculate the index value of the sub-region load impact degree index.

[0125] The index value of the sub-region load prediction difficulty index is calculated according to the following method: Using the entropy weight method, process the obtained daily load volatility, weekly load volatility, and seasonal fluctuation intensity to calculate the load fluctuation characteristic index value; Using the entropy weight method, process the obtained temperature sensitivity, humidity sensitivity, and holiday sensitivity to calculate the external factor sensitivity index value; Using the entropy weight method, process the obtained industrial electricity consumption ratio, number of key users, and number of user types to calculate the electricity consumption structure complexity index value; Using the entropy weight method, process the load fluctuation characteristic index value, external factor sensitivity index value, and electricity consumption structure complexity index value to obtain the index value of the sub-region load prediction difficulty index.

[0126] Step S3: Use the principal component analysis method to process each set of index values and calculate the weight data of each sub-region. During implementation, based on historical prediction deviations, statistical methods can also be used to determine the confidence interval of the predicted value. The calculation formulas for the upper and lower bounds of the prediction interval are as follows:

[0127]

[0128] Among them, X M-Average , X Mean , X L-Average are respectively the upper bound, lower bound and mean value of the interval of the final overall load prediction result, and x Mi , x Li , x mean-i are respectively the upper bound of the probability interval, lower bound of the interval and predicted mean value of the single-region load prediction result of each sub-region.

[0129] Step S4: Obtain the load prediction influencing factor data of each sub-region, and preprocess the load prediction influencing factor data of each sub-region to obtain the corresponding processed influencing factor data respectively. When specifically implemented, the preprocessing process includes three parts: data acquisition, data cleaning, and feature engineering. First is data acquisition, then data cleaning, specifically including outlier isolation forest algorithm detection, context-related analysis, missing value NaN detection, and LS-SVR (Least Squares Support Vector Regression) completion. Finally is feature engineering, including interval coding and label coding in feature encoding, Pearson coefficient and random forest method in feature selection, and Z-score normalization processing.

[0130] Step S5: For any target sub-region, use the corresponding second load prediction model to predict the corresponding target processed influencing factor data to obtain the corresponding second single-region load prediction result.

[0131] When implemented, the second load prediction model corresponding to the target sub-region is generated according to the following method: respectively obtain data sets under various power grid operation scenarios; the power grid operation scenarios are divided according to whether it is a holiday and the season type dimension; for any target power grid operation scenario, use the corresponding data set to train the ARIMA, Holt-Winters, LSTM, GRU, Transformer, and i-Transformer models respectively to obtain multiple candidate load prediction models; determine the prediction model with the best performance among the multiple candidate load prediction models as the load prediction model under the target power grid operation scenario; according to the power grid operation scenario information of the target load prediction influencing factor data, match the power grid operation scenario, and use the load prediction model corresponding to the matched power grid operation scenario as the second load prediction model corresponding to the target sub-region.

[0132] Step S6: Obtain the second historical prediction accuracy of each first load prediction model within 3 months, and multiply each second historical prediction accuracy by the corresponding weight data to obtain multiple products.

[0133] Step S7: Add the multiple products and multiply by the basic weight coefficient to obtain the dynamic adjustment coefficient.

[0134] Step S8: For any sub-region, use the corresponding weight data and dynamic adjustment coefficient to perform weighted averaging on the corresponding first single-region load prediction result and the second single-region load prediction result to obtain a fused load prediction result.

[0135] Step S9: Add up the fused load prediction results to obtain an overall load prediction result.

[0136] This embodiment proposes an innovative power load prediction technology with strong feature extraction ability and data integration ability. By constructing a comprehensive evaluation index system, quantitatively evaluating the power load characteristics of each province and city, using the principal component analysis method to determine weights, and combining the multi-source data fusion method, the results of the first load prediction model and the second load prediction model customized based on the power grid operation scenario are effectively integrated. This method not only adapts to the diversity of the economy, climate, and industrial structure of each province and city, but also solves problems such as data update frequency, transmission delay, and time scale processing, significantly improving the accuracy and reliability of load prediction. Through feature extraction and fusion, the accuracy and robustness of the model in the face of a changing environment are enhanced, the problem of multi-source data fusion and the superposition problem of prediction results in different provinces can be solved, and efficient and accurate main grid load prediction is achieved, providing strong support for the stable operation of the power grid.

[0137] Those skilled in the art can understand that in the above method of the specific embodiment, the writing order of each step does not mean a strict execution order and does not constitute any limitation on the implementation process. The specific execution order of each step should be determined by its function and possible internal logic.

[0138] It should be noted that in practical applications, all the above possible implementation manners can be combined in any combination to form possible embodiments of the present disclosure, which will not be elaborated herein one by one. The information (including but not limited to device information, user information, etc.) and data (including but not limited to data for analysis, storage, and display, etc.) involved in this application are all information and data authorized by users or fully authorized by all parties.

[0139] Based on the same concept, the embodiment of the present disclosure also provides a power grid load prediction device, which corresponds one-to-one to the power grid load prediction method in the above embodiment. Figure 4 The structural schematic diagram of the power grid load prediction device provided by the embodiment of the present disclosure is shown, see Figure 4 As shown, the power grid load prediction device 400 provided by the embodiment of the present disclosure includes:

[0140] The first acquisition module 401 is used to acquire the first single-region load prediction results reported by multiple sub-regions within the region to be predicted; the first single-region load prediction result is the output of the sub-region using its trained first load prediction model;

[0141] A second acquisition module 402, configured to acquire a set of index values corresponding to each of the sub-regions; the set of index values includes index values of at least one of the following indexes: a prediction result credibility index, a sub-region load influence degree index, and a sub-region load prediction difficulty index;

[0142] A calculation module 403, configured to calculate weight data of each sub-region according to each set of index values;

[0143] A prediction module 404, configured to calculate an overall load prediction result of the region to be predicted according to each of the first single-region load prediction results and each of the weight data.

[0144] In some embodiments, the device further includes a first calculation module, configured to: use a preset index weight determination algorithm to process at least one of the acquired first historical prediction accuracy, real-time prediction accuracy, and prediction accuracy under extreme weather, and calculate a historical prediction performance index value; use the index weight determination algorithm to process at least one of the acquired prediction model update frequency, prediction result continuity data, and abnormal prediction ratio, and calculate a prediction model stability index value; use the index weight determination algorithm to process at least one of the acquired data integrity rate, data timeliness data, and data consistency data, and calculate a data quality level index value of the data required for prediction; use the index weight determination algorithm to process at least one of the historical prediction performance index value, the prediction model stability index value, and the data quality level index value, and calculate an index value of the prediction result credibility index.

[0145] In some embodiments, the device further includes a second calculation module, configured to: use a preset index weight determination algorithm to process at least one of the acquired maximum load ratio, average load ratio, and peak load contribution rate, and calculate a load scale ratio index value; use the index weight determination algorithm to process at least one of the acquired peak shaving capacity ratio, new energy installed capacity ratio, and demand response ability data, and calculate an adjustment ability index value; use the index weight determination algorithm to process at least one of the load scale ratio index value and the adjustment ability index value, and calculate an index value of the sub-region load influence degree index.

[0146] In some embodiments, the device further includes a third calculation module, configured to: process at least one of the obtained daily load volatility, weekly load volatility, and seasonal fluctuation intensity by using a preset index weight determination algorithm to calculate a load fluctuation characteristic index value; process at least one of the obtained temperature sensitivity, humidity sensitivity, and holiday sensitivity by using the index weight determination algorithm to calculate an external factor sensitivity index value; process at least one of the obtained industrial electricity consumption ratio, number of key users, and number of user types by using the index weight determination algorithm to calculate an electricity consumption structure complexity index value; process at least one of the load fluctuation characteristic index value, the external factor sensitivity index value, and the electricity consumption structure complexity index value by using the index weight determination algorithm to obtain an index value of the sub-region load prediction difficulty index.

[0147] In some embodiments, the device further includes a load prediction module, configured to: obtain load prediction influencing factor data of each sub-region, and preprocess each load prediction influencing factor data to respectively obtain corresponding processed influencing factor data; for any target sub-region, use a corresponding second load prediction model to predict the corresponding target processed influencing factor data to obtain a corresponding second single-region load prediction result; the prediction module 404 is specifically configured to: for any sub-region, use corresponding weight data to perform weighted averaging on the corresponding first single-region load prediction result and the second single-region load prediction result to obtain a corresponding fusion load prediction result; add up each fusion load prediction result to obtain the overall load prediction result.

[0148] In some embodiments, the device further includes a coefficient calculation module, configured to: obtain the second historical prediction accuracy of each of the first load prediction models within a preset time period, multiply each second historical prediction accuracy by corresponding weight data to obtain a plurality of products; calculate a dynamic adjustment coefficient according to the plurality of products; the prediction module 404 is specifically configured to: for any sub-region, use corresponding weight data and the dynamic adjustment coefficient to perform weighted averaging on the corresponding first single-region load prediction result and the second single-region load prediction result to obtain the fusion load prediction result.

[0149] In some embodiments, in the above device, the second load prediction model corresponding to the target sub-region is trained and generated based on the i-transformer model.

[0150] In some embodiments, the device further includes a training module, configured to: respectively obtain data sets under multiple power grid operation scenarios; the power grid operation scenarios are divided according to at least one dimension of whether it is a holiday or a season type; for any target power grid operation scenario, use the corresponding data set to train at least two models among ARIMA, Holt-Winters, LSTM, GRU, Transformer, and i-Transformer respectively to obtain multiple candidate load prediction models; determine the prediction model with the best performance among the multiple candidate load prediction models as the load prediction model under the target power grid operation scenario; match the power grid operation scenario according to the power grid operation scenario information of the target load prediction influencing factor data, and use the load prediction model corresponding to the matched power grid operation scenario as the second load prediction model corresponding to the target sub-region.

[0151] The present invention provides a power grid load prediction device. First, it obtains the first single-region load prediction results reported by multiple sub-regions within the region to be predicted, as well as at least one of the prediction result credibility index value, sub-region load influence degree index value, and sub-region load prediction difficulty index value corresponding to each sub-region. These index values can reflect various characteristics such as the load prediction quality, influence degree, and load prediction difficulty of each sub-region. Then, according to the first single-region load prediction results and the index value set, the weight data of each sub-region is calculated, so that the proportion of the first single-region load prediction results of each sub-region in the overall load prediction is reasonably allocated, reflecting the importance differences of different sub-regions. Finally, by combining the first single-region load prediction results and the weight data, a more accurate and reliable overall load prediction result can be calculated for the region to be predicted.

[0152] For the specific limitations of the power grid load prediction device, reference can be made to the limitations on the power grid load prediction method in the above text, which will not be elaborated here. Each module in the above power grid load prediction device can be implemented in whole or in part through software, hardware, and their combinations. The above modules can be embedded in the processor of the computer device in hardware form or be independent of it, or can be stored in the memory of the computer device in software form, so that the processor can call and execute the operations corresponding to the above modules.

[0153] Figure 5 FIG. shows a schematic structural diagram of an electronic device provided by an embodiment of the present disclosure. As Figure 5As shown, at the hardware level, the electronic device includes a processor, and optionally also includes an internal bus, a network interface, and a memory. Among them, the memory may include memory, such as high-speed random access memory (RAM), and may also include non-volatile memory, such as at least one disk memory, etc. Of course, the electronic device may also include other hardware required for other services.

[0154] The processor, network interface, and memory can be interconnected through the internal bus. The internal bus can be an ISA (Industry Standard Architecture) bus, a PCI (Peripheral Component Interconnect) bus, or an EISA (Extended Industry Standard Architecture) bus, etc. The bus can be divided into an address bus, a data bus, a control bus, etc. For ease of representation, Figure 5 only a two-way arrow is used in the figure, but it does not mean that there is only one bus or one type of bus.

[0155] The memory is used to store programs. Specifically, the program may include program code, and the program code includes computer operation instructions. The memory can include memory and non-volatile memory, and provide instructions and data to the processor.

[0156] The processor reads the corresponding computer program from the non-volatile memory into the memory and then runs it, forming a power grid load prediction device at the logical level. The processor executes the program stored in the memory and is specifically used to execute the foregoing method.

[0157] A processor may be an integrated circuit chip with the ability to process signals. In the implementation process, each step of the above method can be completed by the integrated logic circuit of the hardware in the processor or the instructions in the form of software. The above-mentioned processor may be a general-purpose processor, including a central processing unit (CPU), a network processor (NP), etc.; it may also be a digital signal processor (DSP), an application specific integrated circuit (ASIC), a field-programmable gate array (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components. It can implement or execute each method, step, and logic block diagram disclosed in the embodiments of the present disclosure. The general-purpose processor may be a microprocessor or the processor may also be any conventional processor, etc. The steps of the method disclosed in combination with the embodiments of the present disclosure can be directly embodied as being executed and completed by the hardware decoding processor, or executed and completed by the combination of the hardware and software modules in the decoding processor. The software module may be located in a mature storage medium in the art such as a random access memory, a flash memory, a read-only memory, a programmable read-only memory, or an electrically erasable programmable memory, a register, etc. The storage medium is located in the memory, and the processor reads the information in the memory and combines its hardware to complete the steps of the above method.

[0158] The electronic device can execute the power grid load forecasting method provided by multiple embodiments of the present disclosure and be implemented as a power grid load forecasting device in Figure 4 the functions of the illustrated embodiments, which will not be elaborated herein in the embodiments of the present disclosure.

[0159] The embodiments of the present disclosure also propose a computer-readable storage medium that stores one or more programs. The one or more programs include instructions that, when executed by an electronic device including multiple application programs, can enable the electronic device to execute the power grid load forecasting method provided by multiple embodiments of the present disclosure.

[0160] Those skilled in the art should understand that the embodiments of the present disclosure can be provided as a method, a system, or a computer program product. Therefore, the present disclosure can adopt the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware aspects. Moreover, the present disclosure can adopt the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk memories, CD-ROMs, optical memories, etc.) containing computer-usable program codes.

[0161] This disclosure is described with reference to the flowcharts and / or block diagrams of methods, apparatuses (systems), and computer program products according to embodiments of the disclosure. It should be understood that each flow and / or block in the flowchart and / or block diagram, and combinations of flows and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to the processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing device to produce a machine, such that the instructions executed by the processor of the computer or other programmable data processing device generate means for implementing the functions specified in one or more of the flows Figure 1 one or more of the flows and / or blocks Figure 1 or means for implementing the functions specified in one or more of the blocks.

[0162] These computer program instructions can also be stored in a computer-readable memory that can direct a computer or other programmable data processing device to operate in a particular manner, such that the instructions stored in the computer-readable memory produce a manufacture including instruction means that implement the functions specified in one or more of the flows Figure 1 one or more of the flows and / or blocks Figure 1 or means for implementing the functions specified in one or more of the blocks.

[0163] These computer program instructions can also be loaded onto a computer or other programmable data processing device, such that a series of operational steps are performed on the computer or other programmable device to produce a computer-implemented process, and thus the instructions executed on the computer or other programmable device provide steps for implementing the functions specified in one or more of the flows Figure 1 one or more of the flows and / or blocks Figure 1 or means for implementing the functions specified in one or more of the blocks.

[0164] In a typical configuration, a computing device includes one or more processors (CPUs), an input / output interface, a network interface, and memory.

[0165] The memory may include non-permanent memory in the form of computer-readable media, random access memory (RAM), and / or non-volatile memory, such as read-only memory (ROM) or flash memory (flash RAM). The memory is an example of computer-readable media.

[0166] A computer-readable medium includes permanent and non-permanent, removable and non-removable media that can implement information storage by any method or technology. The information can be computer-readable instructions, data structures, program modules, or other data. Examples of computer storage media include, but are not limited to, phase change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technologies, compact disc read-only memory (CD-ROM), digital versatile disc (DVD) or other optical storage, magnetic cassette tapes, tapes, disk storage or other magnetic storage devices, or any other non-transitory medium that can be used to store information accessible by a computing device. As defined herein, a computer-readable medium does not include transitory computer-readable media, such as modulated data signals and carrier waves.

[0167] It should also be noted that the term "comprising", "including" or any other variant thereof is intended to cover non-exclusive inclusion, such that a process, method, article or device comprising a series of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article or device. Without further limitation, an element defined by the statement "comprising an..." does not exclude the presence of additional identical elements in the process, method, article or device comprising the element.

[0168] Those skilled in the art should understand that the embodiments of the present disclosure may be provided as a method, system, or computer program product. Therefore, the present disclosure may take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware aspects. Moreover, the present disclosure may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk memory, CD-ROM, optical memory, etc.) containing computer-usable program code.

[0169] The above are only embodiments of the present disclosure and are not used to limit the present disclosure. For those skilled in the art, various changes and modifications can be made to the present disclosure. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principles of the present disclosure shall be included within the scope of the claims of the present disclosure.

Claims

1. A power grid load forecasting method, characterized in that, The method includes: Obtaining first single - area load prediction results reported by multiple sub - areas within the area to be predicted; the first single - area load prediction results are output by the sub - areas using their trained first load prediction models; Obtaining the set of index values corresponding to each of the sub - areas; the set of index values includes the index values of at least one of the following indexes: prediction result credibility index, sub - area load influence degree index, and sub - area load prediction difficulty index; Calculating the weight data of each sub - area according to each set of index values; Calculating the overall load prediction result of the area to be predicted according to each of the first single - area load prediction results and each of the weight data.

2. The method according to claim 1, wherein The index value of the prediction result credibility index is calculated according to the following method: Using a preset index weight determination algorithm to process at least one of the obtained first historical prediction accuracy, real - time prediction accuracy, and prediction accuracy under extreme weather conditions, and calculating the historical prediction performance index value; Using the index weight determination algorithm to process at least one of the obtained prediction model update frequency, prediction result continuity data, and abnormal prediction ratio, and calculating the prediction model stability index value; Using the index weight determination algorithm to process at least one of the obtained data integrity rate, data timeliness data, and data consistency data, and calculating the data quality level index value of the data required for prediction; Using the index weight determination algorithm to process at least one of the historical prediction performance index value, the prediction model stability index value, and the data quality level index value, and calculating the index value of the prediction result credibility index.

3. The method according to claim 1, wherein The index value of the sub - area load influence degree index is calculated according to the following method: Using a preset index weight determination algorithm to process at least one of the obtained maximum load ratio, average load ratio, and peak load contribution rate, and calculating the load scale ratio index value; Using the index weight determination algorithm to process at least one of the obtained peak - shaving capacity ratio, new energy installation ratio, and demand response ability data, and calculating the regulation ability index value; Using the index weight determination algorithm to process at least one of the load scale ratio index value and the regulation ability index value, and calculating the index value of the sub - area load influence degree index.

4. The method according to claim 1, wherein The index value of the sub - area load prediction difficulty index is calculated according to the following method: Using a preset index weight determination algorithm to process at least one of the obtained daily load volatility, weekly load volatility, and seasonal fluctuation intensity, and calculating the load fluctuation characteristic index value; Using the index weight determination algorithm to process at least one of the obtained temperature sensitivity, humidity sensitivity, and holiday sensitivity, and calculating the external factor sensitivity index value; Using the index weight determination algorithm to process at least one of the obtained industrial electricity consumption ratio, number of key users, and number of user types, and calculating the electricity consumption structure complexity index value; Using the above-mentioned index weight determination algorithm, at least one of the load fluctuation characteristic index value, the external factor sensitivity index value, and the electricity consumption structure complexity index value is processed to obtain the index value of the load prediction difficulty index for the sub-region.

5. The method according to any one of claims 1-4, characterized in that The method further includes: Obtaining the load prediction influencing factor data for each sub-region, and preprocessing each piece of load prediction influencing factor data to obtain the corresponding processed influencing factor data; For any target sub-region, using the corresponding second load prediction model to predict the corresponding target processed influencing factor data to obtain the corresponding second single-region load prediction result; The calculating the overall load prediction result of the region to be predicted according to each of the first single-region load prediction results and each of the weight data includes: For any sub-region, using the corresponding weight data to perform weighted averaging on the corresponding first single-region load prediction result and the second single-region load prediction result to obtain the corresponding fused load prediction result; Adding up each of the fused load prediction results to obtain the overall load prediction result.

6. The method according to claim 5, characterized in that, The method further includes: Obtaining the second historical prediction accuracy of each of the first load prediction models within a preset time period, and multiplying each of the second historical prediction accuracies by the corresponding weight data to obtain a plurality of products; Calculating a dynamic adjustment coefficient according to the plurality of products; The performing weighted averaging on the corresponding first single-region load prediction result and the second single-region load prediction result for any sub-region using the corresponding weight data to obtain the corresponding fused load prediction result includes: For any sub-region, using the corresponding weight data and the dynamic adjustment coefficient to perform weighted averaging on the corresponding first single-region load prediction result and the second single-region load prediction result to obtain the fused load prediction result.

7. The method according to claim 5, characterized in that, The second load prediction model corresponding to the target sub-region is generated based on the i-transformer model.

8. The method according to claim 5, characterized in that, The second load prediction model corresponding to the target sub-region is generated according to the following method: Respectively obtaining data sets under various power grid operation scenarios; the power grid operation scenarios are divided according to at least one dimension of whether it is a holiday or the season type; For any target power grid operation scenario, using the corresponding data set to train at least two of the models ARIMA, Holt-Winters, LSTM, GRU, Transformer, and i-Transformer respectively to obtain a plurality of candidate load prediction models; Determining the prediction model with the best performance among the plurality of candidate load prediction models as the load prediction model under the target power grid operation scenario; According to the power grid operation scenario information of the target load prediction influencing factor data, matching the power grid operation scenario, and using the load prediction model corresponding to the matched power grid operation scenario as the second load prediction model corresponding to the target sub-region.

9. A power grid load forecasting device, characterized in that, The device includes: A first acquisition module, configured to acquire first single-region load prediction results reported by multiple sub-regions within a region to be predicted; the first single-region load prediction results are output by the sub-regions using their trained first load prediction models; A second acquisition module, configured to acquire a set of index values respectively corresponding to each of the sub-regions; the set of index values includes index values of at least one of the following indexes: prediction result credibility index, sub-region load influence degree index, and sub-region load prediction difficulty index; A calculation module, configured to calculate weight data for each sub-region according to each set of index values; A prediction module, configured to calculate an overall load prediction result of the region to be predicted according to each of the first single-region load prediction results and each of the weight data.

10. An electronic device, comprising: A processor; And A memory arranged to store computer-executable instructions, wherein the executable instructions, when executed, cause the processor to execute the steps of the method according to any one of claims 1-8.

Images