Power resource allocation method and system based on big data

Abstract

The invention relates to the technical field of resource allocation, and particularly discloses an electric power resource allocation method and system based on big data, which retrieves historical power consumption data of a target power consumption unit similar to a target electric power scheduling date in weather based on weather similarity, and introduces a time sequence analysis technology based on deep learning. Multi-scale power consumption mode time sequence feature enhanced perception is performed on power consumption data of a target power consumption unit, and layered decoding and proofreading are performed on captured multi-scale power consumption mode time sequence features, so that a power consumption critical value of the target power consumption unit is determined; and power consumption unit power data with similar power consumption modes are screened based on the power consumption critical value so as to create a power consumption prediction model, and power consumption prediction and power regulation and control are performed on the power regulation and control area. According to the method, the real power consumption behavior mode of the target power consumption unit can be more effectively sensed, the reliability of the screened sample power consumption unit is improved, and thus the accuracy and adaptability of power consumption prediction are enhanced.

Description

Technical Field

[0001] This application relates to the field of resource allocation technology, and more specifically, to a power resource allocation method and system based on big data. Background Technology

[0002] With the continuous development of the global economy and the increasing dependence of various sectors of society on electricity, the efficient and rational allocation of power resources has become a key link in ensuring energy security, promoting stable economic growth, and improving the quality of life. In the traditional power resource allocation system, power load forecasting and resource allocation are mainly based on historical average electricity consumption data, seasonal patterns, and empirical judgments. However, this allocation method has gradually revealed many limitations when facing increasingly complex and volatile electricity demand scenarios.

[0003] In response, the invention patent with publication number CN118889414A proposes a power resource allocation method based on big data. This method selects multiple sample dates for analysis based on the weather similarity between historical dates and target power dispatch dates. It then determines the power consumption threshold based on the target power user's electricity consumption data within the sample dates, further selecting multiple sample power users with similar electricity consumption characteristics. Finally, it calculates the year-on-year and periodic electricity consumption data of each power user to create a power consumption prediction model. This model is used to predict and regulate electricity consumption in the power control area, thereby reducing the risk of power waste and supply shortages.

[0004] Specifically, in existing technologies, the mean and variance of daily electricity consumption of the target electricity user within a sample period are calculated, and a fixed threshold range for electricity consumption is generated based on a linear formula to screen sample electricity users with similar electricity consumption characteristics to the target electricity user. However, the mean and variance are statistical measures that are highly sensitive to outliers (extremely high or extremely low electricity consumption) in the data. They are easily affected by extreme electricity consumption data and are difficult to dynamically adapt to complex changes in electricity consumption behavior, such as seasonal fluctuations, sudden events, or gradual adjustments in users' electricity consumption habits. This may result in the set threshold values ​​failing to accurately reflect the actual electricity consumption behavior patterns, leading to the selected sample electricity users not truly possessing similar electricity consumption characteristics to the target electricity user, thus affecting the accuracy and reliability of subsequent electricity consumption prediction models.

[0005] Therefore, there is a need for an optimized method and system for power resource allocation based on big data. Summary of the Invention

[0006] To address the aforementioned technical problems, this application is proposed. Embodiments of this application provide a power resource allocation method and system based on big data. It retrieves historical electricity consumption data of target electricity users with weather similarities to the target power dispatch date based on weather similarity, and introduces deep learning-based time-series analysis technology. By enhancing the perception of multi-scale electricity consumption pattern time-series characteristics of the target electricity users' electricity consumption data, it captures the electricity consumption behavior patterns of the target electricity users at different time scales. Furthermore, by performing hierarchical decoding and verification of the captured multi-scale electricity consumption pattern time-series characteristics, it determines the electricity consumption threshold of the target electricity users. Based on the electricity consumption threshold, it filters electricity data of electricity users with similar electricity consumption patterns to create an electricity consumption prediction model, enabling electricity consumption prediction and power regulation in the power control area. This method can more effectively perceive the real electricity consumption behavior patterns of target electricity users, improve the reliability of the selected sample electricity users, and thus enhance the accuracy and adaptability of electricity consumption prediction.

[0007] According to one aspect of this application, a power resource allocation method based on big data is provided, comprising:

[0008] In the power dispatching area, sample dates are selected based on the weather similarity between historical dates and target power dispatching dates to obtain sample date data;

[0009] Based on the sample date data, a first electricity consumption threshold and a second electricity consumption threshold are obtained, and multiple sample electricity consumption units with similar electricity consumption characteristics to the target electricity consumption unit are selected based on the first electricity consumption threshold and the second electricity consumption threshold.

[0010] Calculate the year-on-year and periodic electricity consumption data of the target electricity user and multiple sample electricity users to create an electricity consumption prediction model to predict the electricity consumption of the power control area, obtain the predicted electricity consumption of the power control area, and carry out power control based on the predicted electricity consumption of the power control area.

[0011] The process of obtaining the first and second electricity consumption thresholds based on sample date data includes: performing time-series feature enhancement perception and hierarchical decoding correction based on multi-scale electricity consumption patterns on the electricity consumption data of the target electricity consumption unit in the sample date data to obtain the first and second electricity consumption thresholds.

[0012] According to another aspect of this application, a power resource allocation system based on big data is provided, comprising:

[0013] The sample date selection module is used to select sample dates in the power dispatching area based on the weather similarity between historical dates and target power dispatching dates to obtain sample date data;

[0014] The electricity consumption threshold screening module is used to obtain a first electricity consumption threshold and a second electricity consumption threshold based on the sample date data, and to screen out multiple sample electricity consumption units with similar electricity consumption characteristics to the target electricity consumption unit based on the first electricity consumption threshold and the second electricity consumption threshold.

[0015] The electricity consumption forecasting and control module is used to calculate the year-on-year and periodic electricity consumption data of the target electricity consumption unit and multiple sample electricity consumption units to create an electricity consumption forecasting model to predict the electricity consumption of the power control area, obtain the predicted electricity consumption of the power control area, and carry out power control based on the predicted electricity consumption of the power control area.

[0016] The electricity consumption threshold screening module is used to: perform time-series feature enhancement perception and hierarchical decoding correction based on multi-scale electricity consumption patterns on the electricity consumption data of the target electricity consumption unit in the sample date data to obtain the first electricity consumption threshold and the second electricity consumption threshold.

[0017] Compared with existing technologies, the power resource allocation method and system based on big data provided in this application retrieves historical electricity consumption data of target power users with weather similar to the target power dispatch date based on weather similarity. It also introduces deep learning-based time series analysis technology to enhance the perception of multi-scale electricity consumption pattern time series characteristics of the target power users' electricity consumption data, thereby capturing the electricity consumption behavior patterns of the target power users at different time scales. Furthermore, by performing hierarchical decoding and verification of the captured multi-scale electricity consumption pattern time series characteristics, the electricity consumption threshold of the target power users is determined. Based on the electricity consumption threshold, power data of power users with similar electricity consumption patterns are selected to create an electricity consumption prediction model, enabling electricity consumption prediction and power regulation in the power control area. This method can more effectively perceive the real electricity consumption behavior patterns of target power users, improve the reliability of the selected sample power users, and thus enhance the accuracy and adaptability of electricity consumption prediction. Attached Figure Description

[0018] The above and other objects, features, and advantages of this application will become more apparent from the more detailed description of the embodiments of this application in conjunction with the accompanying drawings. The drawings are provided to further illustrate the embodiments of this application and form part of the specification. They are used together with the embodiments of this application to explain this application and do not constitute a limitation thereof. In the drawings, the same reference numerals generally represent the same components or steps.

[0019] Figure 1 This is a flowchart of a power resource allocation method based on big data according to an embodiment of this application.

[0020] Figure 2 This is a flowchart of sub-step S2 of the big data-based power resource allocation method according to an embodiment of this application.

[0021] Figure 3 This is a schematic diagram of the data flow in sub-step S2 of the big data-based power resource allocation method according to an embodiment of this application.

[0022] Figure 4 This is a flowchart of sub-step S22 of the power resource allocation method based on big data according to an embodiment of this application.

[0023] Figure 5 This is a flowchart of sub-step S222 of the big data-based power resource allocation method according to an embodiment of this application.

[0024] Figure 6 This is a flowchart of sub-step S23 of the big data-based power resource allocation method according to an embodiment of this application.

[0025] Figure 7 This is a block diagram of a big data-based power resource allocation system according to an embodiment of this application. Detailed Implementation

[0026] As indicated in this application and claims, unless the context clearly indicates otherwise, the words "a," "an," "an," and / or "the" are not specifically singular and may include plural forms. Generally speaking, the terms "comprising" and "including" only indicate the inclusion of explicitly identified steps and elements, which do not constitute an exclusive list, and the method or apparatus may also include other steps or elements.

[0027] While this application makes various references to certain modules of the systems according to embodiments of this application, any number of different modules can be used and run on user terminals and / or servers. The modules described are merely illustrative, and different aspects of the systems and methods may use different modules.

[0028] Flowcharts are used in this application to illustrate the operations performed by the system according to embodiments of this application. It should be understood that the preceding or following operations are not necessarily performed in exact order. Instead, various steps can be processed in reverse order or simultaneously as needed. Furthermore, other operations can be added to these processes, or one or more steps can be removed from them.

[0029] Hereinafter, exemplary embodiments according to this application will be described in detail with reference to the accompanying drawings. Obviously, the described embodiments are merely some embodiments of this application, and not all embodiments of this application. It should be understood that this application is not limited to the exemplary embodiments described herein.

[0030] It is worth noting that all data acquisition actions in this application were carried out in compliance with the relevant data protection laws and policies of the country where the application is located, and with the authorization granted by the owner of the relevant device.

[0031] As mentioned in the background section above, patent CN118889414A proposes a power resource allocation method based on big data. This method selects multiple sample dates for analysis based on the weather similarity between historical dates and target power dispatch dates. It then determines the power consumption threshold based on the target power user's electricity consumption data within the sample dates, further selecting multiple sample power users with similar electricity consumption characteristics to the target power user. Finally, it calculates the year-on-year and periodic electricity consumption data of each power user to create a power consumption prediction model. This model is used to predict and regulate electricity consumption in the power control area, thereby reducing the risk of power waste and supply shortages.

[0032] In existing technologies, the mean and variance of daily electricity consumption of a target electricity user within a sample period are calculated, and a fixed threshold range is determined using a linear formula to select sample electricity users similar to the target user. However, the mean and variance are sensitive to outliers (such as extremely high or low electricity consumption) and are easily affected by extreme data. Furthermore, this method struggles to adapt to complex changes in electricity consumption behavior, such as seasonal fluctuations, sudden events, or gradual adjustments in user electricity usage habits. This may result in the set threshold failing to accurately reflect actual electricity consumption patterns, leading to insufficient similarity between the selected sample electricity users and the target user, ultimately affecting the accuracy and reliability of subsequent electricity consumption prediction models. To address the aforementioned technical issues, this application proposes a big data-based power resource allocation method. This method retrieves historical electricity consumption data of target power-consuming units with weather similarities to the target power dispatch date based on weather similarity. It also introduces deep learning-based time-series analysis technology to enhance the perception of multi-scale electricity consumption pattern time-series characteristics of the target power-consuming units' electricity consumption data, thereby capturing their electricity consumption behavior patterns at different time scales. Furthermore, by performing hierarchical decoding and verification of the captured multi-scale electricity consumption pattern time-series characteristics, the method determines the electricity consumption threshold of the target power-consuming units. Based on this threshold, it filters electricity data from power-consuming units with similar consumption patterns to create an electricity consumption prediction model, enabling electricity consumption prediction and power regulation in the power control area. This method can more effectively perceive the actual electricity consumption behavior patterns of target power-consuming units, improve the reliability of the selected sample power-consuming units, and thus enhance the accuracy and adaptability of electricity consumption prediction.

[0033] Figure 1 This is a flowchart of a big data-based power resource allocation method according to an embodiment of this application. Figure 1As shown, the power resource allocation method based on big data includes the following steps: S1, in the power control area, sample dates are selected based on the weather similarity between historical dates and target power dispatch dates to obtain sample date data; S2, a first power consumption threshold and a second power consumption threshold are obtained based on the sample date data, and multiple sample power consumption units with similar power consumption characteristics to the target power consumption unit are selected based on the first power consumption threshold and the second power consumption threshold; S3, the year-on-year power consumption data and periodic power consumption data of the target power consumption unit and multiple sample power consumption units are calculated to create a power consumption prediction model to predict the power consumption in the power control area, obtain the predicted power consumption in the power control area, and perform power control based on the predicted power consumption in the power control area.

[0034] In the aforementioned big data-based power resource allocation method, step S1 involves selecting sample dates within the power control area based on the weather similarity between historical dates and the target power dispatch date to obtain sample date data. It should be understood that there is a significant correlation between weather and user electricity demand; for example, factors such as temperature, humidity, wind speed, and sunshine duration can all influence user electricity consumption behavior. By selecting historical dates with weather similar to the target power dispatch date as sample dates, it helps ensure that the analyzed data is consistent with the weather conditions of the target scenario, reducing the interference of external environmental differences on electricity consumption pattern analysis. Specifically, measuring the weather similarity between historical characteristic dates and the target power dispatch date is an existing technology. It can be performed using the method disclosed in Chinese Patent CN118889414A, calculating the weather matching coefficient between the two based on the sum of the absolute differences in temperature and wind intensity data between the historical characteristic dates and the target power dispatch date, and selecting dates with a weather matching coefficient below a threshold as sample dates to ensure that the weather conditions of the sample dates are close to those of the target date. Of course, other weather similarity measurement methods can also be used, and this application is not limited to these methods.

[0035] Specifically, to accurately capture the weather characteristics of target power dispatch dates, a deep understanding of the impact of meteorological factors such as temperature, humidity, wind speed, and sunshine duration on power consumption behavior is essential. Each weather condition can trigger different changes in electricity demand. For example, high temperatures may lead to increased air conditioning usage, thereby increasing overall power consumption; while strong winds or low humidity environments may reduce the operating frequency of certain equipment, lowering the power load. Therefore, when determining sample dates, it is necessary to carefully consider the interactions between these variables and their specific impact on power demand.

[0036] In the implementation process, selecting historical dates that meet the criteria from a massive historical dataset is crucial. This requires establishing a detailed weather database that not only contains long-term accumulated daily, hourly, and even minute-level temperature records, but also covers other meteorological parameters such as wind speed and precipitation. By inputting the weather conditions of the target power dispatch date into a pre-set algorithm model, the system can automatically identify and calculate which historical dates experienced similar power consumption patterns under the same meteorological conditions. It is worth noting that the emphasis here is not only on the matching degree of individual meteorological factors, but more importantly, on whether the overall weather conditions under the combined effect of multiple meteorological factors are similar.

[0037] When constructing weather similarity measurement models, it is common practice to use the sum of absolute differences to measure the differences in weather conditions between two dates. For example, for key indicators such as temperature and wind speed, the absolute differences between the target power dispatch date and historical dates can be calculated separately, and these differences can be summed to obtain a total weather matching coefficient. If this coefficient is lower than a preset threshold, the two dates are considered to have sufficient weather similarity and can be included as potential sample dates. In addition, considering that the meteorological characteristics of different regions may vary significantly, the criteria for judging weather similarity need to be adjusted according to the specific geographical location to ensure that the selected sample dates can truly reflect the local climate characteristics.

[0038] Meanwhile, in practical applications, the impact of seasonal variations on weather similarity assessments must also be considered. Since meteorological conditions in the same region often exhibit significant periodic fluctuations, such as hot and rainy summers and cold and dry winters, this means that even seemingly similar dates in different seasons may reveal substantial differences in underlying electricity consumption patterns. Therefore, when selecting sample dates, in addition to focusing on the similarity of the weather itself, special attention must be paid to the role of seasonal factors to avoid errors caused by ignoring seasonal characteristics.

[0039] Furthermore, unforeseen events are also a significant factor influencing weather similarity measurements. Extreme weather phenomena such as sudden cold waves, heat waves, or torrential rains, while perhaps short-lived, can have a substantial impact on short-term electricity demand. Traditional methods based on averages or simple statistical indicators are insufficient for these special situations, necessitating the introduction of more flexible analytical approaches. For example, real-time monitoring systems can be used to obtain the latest meteorological information and quickly integrate it into existing weather similarity evaluation systems. This allows for rapid adjustments to sample date selection strategies, ensuring the timeliness and reliability of forecast results.

[0040] In the aforementioned big data-based power resource allocation method, step S2 involves obtaining a first and second electricity consumption threshold based on the sample date data, and then filtering out multiple sample electricity consumption units with similar electricity consumption characteristics to the target electricity consumption unit based on the first and second thresholds. In a specific example of this application, step S2 includes: performing time-series feature enhancement perception and hierarchical decoding correction based on multi-scale electricity consumption patterns on the electricity consumption data of the target electricity consumption unit in the sample date data to obtain the first and second thresholds. Specifically, this application considers that existing technologies mainly calculate electricity consumption data thresholds based on the statistical characteristics of electricity consumption data as a screening standard for the electricity consumption patterns of electricity consumption units. However, statistical characteristics are easily affected by extreme data and are difficult to accurately reflect the actual electricity consumption behavior patterns and changes of the target electricity consumption unit. Therefore, to improve the accuracy of the screening, this application introduces deep learning technology. By performing multi-timescale analysis on the electricity consumption data of the target electricity user, it identifies electricity consumption patterns under different scenarios (such as baseline load, periodic peaks, and sudden events), thereby generating dynamic and adaptive electricity consumption thresholds. This avoids threshold bias caused by outliers or single-scale analysis, overcoming the limitations of fixed statistics. Furthermore, based on the determined electricity consumption thresholds, multiple sample electricity users with similar electricity consumption characteristics to the target electricity user are selected as analysis objects to construct an electricity consumption prediction model, thus avoiding model construction bias due to excessive differences in electricity consumption scale. Figure 2 This is a flowchart of sub-step S2 of the big data-based power resource allocation method according to an embodiment of this application. Figure 3 This is a schematic diagram of the data flow in sub-step S2 of the big data-based power resource allocation method according to an embodiment of this application. For example... Figure 2 and Figure 3 As shown, step S2 includes the following steps: S21, performing multi-scale electricity consumption pattern time-series encoding on the electricity consumption data of the target electricity consumption unit in the sample date data to obtain a first-scale electricity consumption pattern time-series encoded feature vector and a second-scale electricity consumption pattern time-series encoded feature vector; S22, performing feature time-series expression enhancement on the first-scale electricity consumption pattern time-series encoded feature vector and the second-scale electricity consumption pattern time-series encoded feature vector respectively to obtain a first-scale electricity consumption pattern time-series enhanced encoded feature vector and a second-scale electricity consumption pattern time-series enhanced encoded feature vector; S23, performing feature decoding and double criticality verification on the first-scale electricity consumption pattern time-series enhanced encoded feature vector and the second-scale electricity consumption pattern enhanced encoded feature vector respectively to obtain a first electricity consumption critical value and a second electricity consumption critical value.

[0041] Specifically, in a specific example of this application, step S21 includes: using a multi-scale electricity consumption pattern sequence encoder based on an LSTM-1DCNN hybrid model to perform time-series encoding on the electricity consumption data of the target electricity user in the sample date data to obtain the first-scale electricity consumption pattern time-series encoded feature vector and the second-scale electricity consumption pattern time-series encoded feature vector. It should be understood that this application considers the multi-scale temporal characteristics of electricity data, meaning that electricity data exhibits different patterns and characteristics at different time scales. For example, at shorter time scales, electricity data may be affected by instantaneous factors such as users' daily activities and equipment start-up and shutdown, exhibiting high-frequency fluctuations; while at longer time scales (such as weekly or monthly), electricity data may be affected by long-term factors such as seasonal changes and holiday effects, exhibiting a more stable trend. Therefore, in order to comprehensively capture the temporal characteristics of electricity data, this application further employs an LSTM-1DCNN hybrid model to perform multi-scale analysis on the electricity consumption data of the target electricity user in the sample date data. Specifically, LSTM (Long Short-Term Memory) networks, as a special type of recurrent neural network, can effectively handle long-term dependencies in time-series data, and have significant advantages in capturing long-term trends and periodic changes in electricity data. 1DCNN (One-Dimensional Convolutional Neural Network), on the other hand, excels at extracting local features and patterns from time-series data. By using one-dimensional convolutional kernels to perform sliding convolution operations along the time dimension of electricity consumption data, it can effectively capture local change patterns in electricity consumption data and identify high-frequency fluctuations and instantaneous changes. Therefore, by combining LSTM and 1DCNN models, the advantages of both can be fully utilized to achieve comprehensive capture of multi-scale time-series features of electricity data, obtaining time-series encoded feature vectors of first-scale and second-scale electricity consumption patterns, thus providing a more comprehensive understanding of electricity consumption behavior patterns. The first-scale electricity consumption pattern time-series encoded feature vector is extracted by an LSTM model, representing the electricity consumption behavior pattern of the target electricity user over a longer time scale, such as monthly electricity consumption trends. The second-scale electricity consumption pattern time-series encoded feature vector is extracted by a 1DCNN model, primarily reflecting the electricity consumption fluctuations of the target electricity user in the short term, such as changes in electricity consumption over several consecutive days. Comprehensive analysis of the time-series features of the target electricity user's electricity consumption patterns at different time scales helps to more accurately determine the electricity consumption threshold of the target electricity user, providing strong support for subsequent screening of sample electricity users with similar electricity consumption characteristics.

[0042] Specifically, step S22 involves performing temporal representation enhancement on the first-scale power consumption pattern temporal coding feature vector and the second-scale power consumption pattern temporal coding feature vector to obtain the first-scale power consumption pattern temporal enhancement coding feature vector and the second-scale power consumption pattern temporal enhancement coding feature vector. Specifically, considering that the initially encoded first-scale power consumption pattern temporal coding feature vector and second-scale power consumption pattern temporal coding feature vector may contain redundant or noisy information (such as abnormal fluctuations caused by sensor acquisition errors), direct feature decoding can easily lead to biased critical value estimation. Therefore, this application further introduces an attention mechanism, which performs statistical analysis of the effective components of each local dimension feature within the first-scale power consumption pattern temporal coding feature vector and the second-scale power consumption pattern temporal coding feature vector to identify and enhance key power consumption pattern features (such as periodic peaks), suppress irrelevant noise, thereby achieving temporal representation enhancement processing of the first-scale power consumption pattern temporal coding feature vector and the second-scale power consumption pattern temporal coding feature vector, and improving feature discriminability. Figure 4 This is a flowchart of sub-step S22 of the big data-based power resource allocation method according to an embodiment of this application. Figure 4 As shown, step S22 includes the following steps: S221, recombining the feature components of the first-scale power consumption mode temporal coding feature vector to obtain a set of first-scale power consumption mode temporal feature local phase coding vectors; S222, based on the statistical characteristics of the effective components of each first-scale power consumption mode temporal feature local phase coding vector in the set of first-scale power consumption mode temporal feature local phase coding vectors, adjusting the feature component saliency of the set of first-scale power consumption mode temporal feature local phase coding vectors to obtain the first-scale power consumption mode temporal enhancement coding feature vector.

[0043] More specifically, in a specific example of this application, step S221 includes: recombining the feature components of the first-scale electricity consumption mode temporal coding feature vector based on one-dimensional convolutional coding to obtain a set of local phase coding vectors of the first-scale electricity consumption mode temporal features, expressed by the formula:

[0044] Conv l×1 (X) = {x1, x2, ..., x} i ,...,x n}

[0045] Where X represents the time-series encoded feature vector of the first-scale electricity consumption pattern, and Conv l×1 (·) denotes a one-dimensional convolutional encoding operation, where l is the scale of the one-dimensional convolutional kernel, and x1, x2, x... i and x nThese represent the 1st, 2nd, 1st, and 2nd local phase encoding vectors of the first-scale electricity consumption mode in the set of local phase encoding vectors of the first-scale electricity consumption mode, respectively, where n is the number of vectors in the set of local phase encoding vectors of the first-scale electricity consumption mode.

[0046] In other words, by using one-dimensional convolution operations, the local power consumption patterns at different positions in the first-scale power consumption pattern temporal coding feature vector are encoded to form a set of first-scale power consumption pattern temporal feature local phase coding vectors containing multi-view and multi-detail local structural features, which can provide a refined temporal feature representation for subsequent feature enhancement and critical value determination.

[0047] Figure 5 This is a flowchart of sub-step S222 of the big data-based power resource allocation method according to an embodiment of this application. Figure 5 As shown, step S222 includes the following steps: S2221, refining the information of each local phase encoding vector of the first-scale power consumption mode in the set of local phase encoding vectors of the first-scale power consumption mode to obtain a set of local phase encoding vectors extracted from the first-scale power consumption mode; S2222, calculating the feature saliency metric of each local phase encoding vector extracted from the first-scale power consumption mode in the set of local phase encoding vectors extracted from the first-scale power consumption mode; S2223, calculating the feature component recalibration gain factor of each local phase encoding vector extracted from the first-scale power consumption mode based on the feature saliency metric of each local phase encoding vector extracted from the first-scale power consumption mode; S2224, adjusting the feature component saliency of the set of local phase encoding vectors of the first-scale power consumption mode based on the feature component recalibration gain factor of each local phase encoding vector extracted from the first-scale power consumption mode to obtain the first-scale power consumption mode time-enhanced encoding feature vector.

[0048] In a specific example of this application, S2221 is expressed by the formula:

[0049]

[0050] Where, x i This represents the i-th local phase encoding vector of the first-scale electricity consumption mode time series feature in the set of first-scale electricity consumption mode time series feature local phase encoding vectors, ‖·‖ 2 v represents the square of the vector norm. i The first-scale electricity consumption mode time series feature extraction local phase encoding vector in the set of first-scale electricity consumption mode time series feature extraction local phase encoding vectors, i.e., x iThe corresponding first-scale electricity consumption mode time-series features are extracted to extract the local phase encoding vector.

[0051] In other words, through feature distillation, key information closely related to electricity consumption pattern analysis is selectively retained, while redundancy and noise are suppressed or eliminated. This improves the efficiency and effectiveness of feature representation, making the features more concise and possessing stronger generalization ability, thus laying the foundation for accurately determining electricity consumption thresholds. Specifically, by selectively processing the local phase encoding vectors of the time-series features of each first-scale electricity consumption pattern, such as extracting core effective components based on vector norms and discarding invalid or repetitive information, a set of local phase encoding vectors extracted from the first-scale electricity consumption pattern time-series features containing high-value information is formed. This removes redundant dimensions and mixed noise from the features, highlighting key features that reflect real electricity consumption behavior patterns, making subsequent feature processing more accurate and efficient.

[0052] In a specific example of this application, S2222 is expressed by the formula:

[0053] en i =count i (v ij ),|v ij -v ij-1 |+|v ij -v ij+1 |>ε

[0054] Among them, v ij This represents the feature value at the j-th position in the local phase encoding vector extracted from the time-series features of the first-scale electricity consumption pattern, count. i (·) represents the counting function, ε represents the threshold parameter, |·| represents taking the absolute value, and v ij-1 and v ij+1 En represents the feature values ​​at positions j-1 and j+1 in the local phase encoding vector extracted from the time-series features of the first-scale electricity consumption pattern. i Indicates v i The corresponding feature significance measure.

[0055] In other words, through statistical analysis, the number of feature dimensions or combinations with high discriminative power and high relevance to identifying the electricity consumption patterns of target electricity users is quantified in the local phase encoding vectors extracted from the time-series features of each first-scale electricity consumption pattern. This forms a statistical indicator reflecting the value of feature information, providing data support for calculating the recalibration gain factor of feature components, thereby achieving targeted enhancement of key features. The feature significance measure obtained in this way, which characterizes the effective information content of the local phase encoding vectors extracted from the time-series features of each first-scale electricity consumption pattern, can accurately locate feature components that are important for electricity consumption pattern analysis, making the enhanced features more focused on the core patterns that reflect the actual electricity consumption behavior of target electricity users and reducing interference from invalid information.

[0056] In a specific example of this application, S2223 is expressed by the formula:

[0057]

[0058] Where, λ i x represents i The corresponding eigenphase reshaping intermediate parameters, arctan(·) represents the arctangent function, e i x represents i The corresponding characteristic components are recalibrated with gain factors.

[0059] In other words, by comprehensively considering the characteristics of the feature significance metric, the power consumption pattern analysis target, and the desired enhancement effect, a nonlinear mapping rule is designed to transform the quantification results of the effective components into differentiated gain weights. This enables fine-grained control of feature phase information and supports diverse feature enhancement strategies in complex scenarios. Through this calculation method, vectors with high statistical values ​​are recalibrated with higher feature components to highlight their significance in the feature space, thereby providing precise weight guidance for subsequent feature component significance adjustment.

[0060] In a specific example of this application, S2224 includes: First, optimizing the gain factor of the feature component recalibration of the local phase encoding vector extracted from the time-series features of each first-scale electricity consumption mode based on the effective component translation invariance to obtain a set of optimized first-scale electricity consumption mode time-series feature component recalibration gain factors, expressed by the formula:

[0061]

[0062] Among them, e (·) K represents an exponential function with the natural constant as its base. i T represents the complex manifold flatness factor. i Denotes the gauge field strength field, e ′i e i The corresponding optimized first-scale power consumption mode timing characteristic component recalibrates the gain factor.

[0063] In other words, by introducing metric decomposition and canonical connection construction of complex structure manifolds, the recalibration gain factor of feature components is combined with the spatial invariance of effective feature components. This ensures that the effective components in the set space remain invariant under translation transformation, thereby maintaining the consistency and stability of local phase encoding of first-scale electricity consumption mode time series features during feature enhancement. This ensures that the generated set of optimized first-scale electricity consumption mode time series feature component recalibration gain factors can avoid the loss or distortion of feature phase information due to spatial position changes during enhancement, and ensures that key electricity consumption mode features maintain a stable and symmetrical expression in the feature space.

[0064] Then, the set of recalibration gain factors for the time-series characteristic components of the optimized first-scale power consumption mode is normalized to obtain the set of phase reshaping gain weights for the time-series characteristic components of the first-scale power consumption mode, which is expressed by the formula:

[0065]

[0066] Where exp(·) represents the exponential function operation with base e, a i e ′i The corresponding first-scale power consumption mode timing characteristics phase reshaping gain weight.

[0067] In other words, a normalization algorithm is used to process the recalibration gain factor of the time-series feature components of the first-scale electricity consumption mode, forming a set of phase reshaping gain weights of the first-scale electricity consumption mode time-series features that can accurately characterize the importance of each feature vector, providing a scientific and reasonable weight allocation basis for subsequent feature weight fusion.

[0068] Finally, based on the set of phase reshaping gain weights of the first-scale electricity consumption pattern timing features, the set of local phase encoding vectors of the first-scale electricity consumption pattern timing features is weighted and fused to obtain the first-scale electricity consumption pattern timing enhancement encoding feature vector, which is expressed by the formula:

[0069]

[0070] Among them, v enhanced This represents the time-enhanced coding feature vector of the first-scale electricity consumption pattern.

[0071] In other words, by using weighted fusion, the effective information of local phase encoding vectors of multiple first-scale electricity consumption patterns is fused to form a more representative and discriminative feature representation. Specifically, the local phase encoding vectors of first-scale electricity consumption patterns are weighted and summed according to the phase reshaping gain weight of the first-scale electricity consumption patterns. This helps to highlight the contribution of important features and suppress the influence of secondary or noisy features. As a result, the aggregated first-scale electricity consumption pattern time-series enhanced encoding feature vector can more accurately reflect the electricity consumption pattern characteristics of the target electricity consumption unit at different time scales, providing high-quality feature input for the subsequent determination of electricity consumption thresholds and the construction of electricity consumption prediction models.

[0072] Figure 6 This is a flowchart of sub-step S23 of the big data-based power resource allocation method according to an embodiment of this application. Figure 6 As shown, step S23 includes the following steps: S231, performing feature decoding on the first-scale electricity consumption pattern temporal enhancement coding feature vector and the second-scale electricity consumption pattern temporal enhancement coding feature vector respectively to obtain a first electricity consumption critical upper limit estimate and a first electricity consumption critical lower limit estimate, as well as a second electricity consumption critical upper limit estimate and a second electricity consumption critical lower limit estimate; S232, calculating the average between the first electricity consumption critical upper limit estimate and the second electricity consumption critical upper limit estimate as the first electricity consumption critical value; S233, calculating the average between the first electricity consumption critical lower limit estimate and the second electricity consumption critical lower limit estimate as the second electricity consumption critical value. In a specific example of this application, step S231 includes: inputting the first-scale electricity consumption pattern temporal enhancement coding feature vector into a multilayer perceptron-based electricity consumption critical value decoder to obtain the first electricity consumption critical upper limit estimate and the first electricity consumption critical lower limit estimate.

[0073] Here, feature decoding is performed on the temporal enhanced coding feature vectors of electricity consumption patterns at the first and second scales respectively to obtain the electricity consumption threshold estimates of the target electricity consumption unit from different time perspectives. This helps to improve the accuracy and robustness of the electricity consumption threshold estimation through multi-perspective comprehensive analysis and verification. In the specific decoding process, a multilayer perceptron (MLP) is used as the decoder model. By performing multilayer nonlinear transformations on the input temporal enhanced coding feature vectors of electricity consumption patterns, its powerful nonlinear fitting ability is used to extract deep-level electricity consumption change feature representations and map them to the electricity consumption threshold numerical space. This outputs upper and lower limit estimates of the electricity consumption threshold that are adapted to the electricity consumption change trend of the target electricity consumption unit, revealing the highest and lowest reasonable electricity consumption under the normal electricity consumption mode in the current electricity consumption scenario.

[0074] It should be understood that the first upper limit estimate of electricity consumption and the first lower limit estimate of electricity consumption are the highest and lowest reasonable electricity consumption obtained by decoding the time-series enhanced coding feature vector of the first-scale electricity consumption pattern (i.e., the long-term electricity consumption behavior pattern); while the second upper limit estimate of electricity consumption and the second lower limit estimate of electricity consumption are the highest and lowest reasonable electricity consumption obtained by decoding the time-series enhanced coding feature vector of the second-scale electricity consumption pattern (i.e., the short-term electricity consumption fluctuation). Because electricity consumption behavior patterns at different scales have certain differences and complementarities, this application further conducts a comprehensive analysis and verification of the upper and lower limit estimation results of electricity consumption thresholds at long-term and short-term scales. The first electricity consumption threshold is obtained by calculating the average between the first and second estimated upper limit values ​​of the electricity consumption threshold, and the second electricity consumption threshold is obtained by calculating the average between the first and second estimated lower limit values ​​of the electricity consumption threshold. This balances the impact of long-term trends and short-term fluctuations on the estimation of electricity consumption thresholds, improves the rationality and adaptability of electricity consumption threshold setting, avoids the one-sidedness caused by relying solely on single-scale analysis, and provides a more reliable basis for subsequent selection of electricity users.

[0075] In the aforementioned big data-based power resource allocation method, step S3 involves calculating the year-on-year and periodic power consumption data of the target power-consuming unit and multiple sample power-consuming units to create a power consumption prediction model for power regulation areas, obtaining the predicted power consumption of the power regulation areas, and then performing power regulation based on the predicted power consumption of the power regulation areas. Specifically, the complexity of power resource allocation stems from the spatiotemporal heterogeneity of power consumption behavior. Power-consuming units are affected by multiple factors such as weather, season, and production cycle, and their power consumption exhibits both long-term trends (such as year-on-year growth or decline) and short-term fluctuations (such as weekly / monthly cycle changes). Traditional methods rely on static thresholds or historical averages of a single dimension for prediction, which makes it difficult to capture the nonlinear characteristics under dynamic changes, easily leading to insufficient power supply or resource waste. To address this, by calculating the year-on-year power consumption data of each power-consuming unit to characterize the long-term trend, calculating its periodic power consumption data to reflect short-term fluctuations, and combining multi-unit data to construct a prediction model, it helps to solve the problem of insufficient representation of complex power consumption patterns by traditional methods, providing a more granular decision-making basis for power dispatch.

[0076] In practical implementation, the calculation of year-on-year electricity consumption data firstly involves calculating the average annual electricity consumption growth rate of electricity-consuming units by grouping them by year, reflecting their long-term electricity consumption trend. For the calculation of periodic electricity consumption data, the intensity of short-term electricity consumption fluctuations is quantified by monitoring the continuous volatility of electricity consumption in the recent period (e.g., 30 days). The construction of the electricity consumption forecasting model relies on the fusion of these two types of data. Specifically, firstly, the year-on-year and periodic electricity consumption data of the target electricity-consuming unit and each sample electricity-consuming unit are input into a polynomial fitting model. Combined with the predicted baseline electricity consumption (the average periodic electricity consumption of the sample units), a nonlinear mapping relationship is established between the electricity consumption forecasting coefficients and the verification electricity consumption (actual value). For example, the polynomial coefficients are solved using the least squares method to make the model output value approximate the actual electricity consumption on the verification date. Finally, the electricity consumption forecasting coefficients of the target unit are substituted into the model to predict its electricity consumption on the target scheduling date, and the total electricity demand is obtained by summarizing the forecasts of all units in the region, guiding the power generation end to dynamically adjust the power supply. In this way, by conducting collaborative analysis of electricity consumption data of electricity-consuming units both longitudinally (year-on-year) and laterally (periodic fluctuations), we can effectively uncover the potential patterns in electricity consumption behavior, significantly improve the spatiotemporal resolution of electricity consumption forecasting, and ultimately achieve dynamic supply and demand balance scheduling of power resources.

[0077] In summary, the big data-based power resource allocation method based on the embodiments of this application is clarified. It retrieves historical electricity consumption data of target power-consuming units with weather similarity to the target power dispatch date based on weather similarity, and introduces deep learning-based time-series analysis technology. By enhancing the perception of multi-scale electricity consumption pattern time-series characteristics of the target power-consuming units' electricity consumption data, it captures the electricity consumption behavior patterns of the target power-consuming units at different time scales. Furthermore, by performing hierarchical decoding and verification of the captured multi-scale electricity consumption pattern time-series characteristics, it determines the electricity consumption threshold of the target power-consuming units. Based on the electricity consumption threshold, it filters electricity data of power-consuming units with similar electricity consumption patterns to create an electricity consumption prediction model, enabling electricity consumption prediction and power regulation in the power control area. This method can more effectively perceive the real electricity consumption behavior patterns of target power-consuming units, improve the reliability of the selected sample power-consuming units, and thus enhance the accuracy and adaptability of electricity consumption prediction.

[0078] Furthermore, a power resource allocation system based on big data is also provided.

[0079] Figure 7 This is a block diagram of a big data-based power resource allocation system according to an embodiment of this application. Figure 7As shown, the big data-based power resource allocation system 100 according to an embodiment of this application includes: a sample date selection module 110, used to select sample dates in a power control area based on the weather similarity between historical dates and target power dispatch dates to obtain sample date data; an electricity consumption threshold screening module 120, used to obtain a first electricity consumption threshold and a second electricity consumption threshold based on the sample date data, and to screen out multiple sample electricity consumption units with similar electricity consumption characteristics to the target electricity consumption unit based on the first electricity consumption threshold and the second electricity consumption threshold; and an electricity consumption prediction and control module 130, used to calculate the year-on-year electricity consumption data and periodic electricity consumption data of the target electricity consumption unit and multiple sample electricity consumption units to create an electricity consumption prediction model to predict the electricity consumption in the power control area, obtain the predicted electricity consumption in the power control area, and perform power control based on the predicted electricity consumption in the power control area. The electricity consumption threshold screening module 120 is used to: perform time-series feature enhancement perception and hierarchical decoding correction based on multi-scale electricity consumption patterns on the electricity consumption data of the target electricity consumption unit in the sample date data to obtain the first electricity consumption threshold and the second electricity consumption threshold.

[0080] Those skilled in the art will understand that the specific operations of each module in the aforementioned big data-based power resource allocation system have been referenced above. Figures 1 to 6 The description of the big data-based power resource allocation method is detailed here, and therefore, its repeated description will be omitted.

[0081] The basic principles of the present invention have been described above with reference to specific embodiments. However, it should be noted that the advantages, benefits, and effects mentioned in the present invention are merely examples and not limitations, and should not be considered as essential features of each embodiment of the present invention. Furthermore, the specific details of the above embodiments are for illustrative and facilitative purposes only, and are not limitations. These details do not limit the present invention to the necessity of employing the specific details described above.

[0082] In the above embodiments, the descriptions of each embodiment have their own emphasis. Parts not described in detail or in a particular embodiment can be referred to in the relevant descriptions of other embodiments. It should be understood that the disclosed systems and methods can be implemented in other ways within the several embodiments provided by this invention. For example, the system embodiments described above are merely illustrative; for instance, the unit division is only a logical functional division, and other division methods may be used in actual implementation.

[0083] It will be apparent to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above, and that the invention can be implemented in other specific forms without departing from its spirit or essential characteristics. Therefore, the embodiments should be considered in all respects as exemplary and non-limiting, and the scope of the invention is defined by the appended claims rather than the foregoing description. Thus, all variations falling within the meaning and scope of equivalents of the claims are intended to be embraced within the present invention. No appended diagram markings in the claims should be construed as limiting the scope of the claims.

[0084] Finally, it should be noted that the above description has been given for illustrative and descriptive purposes. Furthermore, the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the present invention.

Claims

1. A big data-based power resource allocation method, characterized in that, The method comprises the following steps: In the power regulation area, sample dates are selected based on the weather similarity between historical dates and the target power dispatch date to obtain sample date data; According to the sample date data, a first power consumption threshold and a second power consumption threshold are obtained, and a plurality of sample power consumption units with similar power consumption characteristics to the target power consumption unit are screened out according to the first power consumption threshold and the second power consumption threshold; The same period power consumption data and the periodic power consumption data of the target power consumption unit and the plurality of sample power consumption units are calculated to create a power consumption prediction model for power prediction of the power regulation area, and the predicted power consumption of the power regulation area is obtained, and power regulation is carried out based on the predicted power consumption of the power regulation area; Wherein, according to the sample date data, the first power consumption threshold and the second power consumption threshold are obtained, which comprises: the power consumption data of the target power consumption unit in the sample date data is subjected to multi-scale power consumption mode time sequence feature enhancement perception and hierarchical decoding correction based on multi-scale power consumption mode time sequence feature enhancement perception and hierarchical decoding correction to obtain the first power consumption threshold and the second power consumption threshold.

2. The big data based power resource dispatching method according to claim 1, characterized in that, The power consumption data of the target power consumption unit in the sample date data is subjected to multi-scale power consumption mode time sequence feature enhancement perception and hierarchical decoding correction based on multi-scale power consumption mode time sequence feature enhancement perception and hierarchical decoding correction to obtain the first power consumption threshold and the second power consumption threshold, which comprises: The power consumption data of the target power consumption unit in the sample date data is subjected to multi-scale power consumption mode time sequence coding to obtain first-scale power consumption mode time sequence coding feature vector and second-scale power consumption mode time sequence coding feature vector; The first-scale power consumption mode time sequence coding feature vector and the second-scale power consumption mode time sequence coding feature vector are subjected to feature time sequence expression enhancement respectively to obtain first-scale power consumption mode time sequence enhancement coding feature vector and second-scale power consumption mode time sequence enhancement coding feature vector; The first-scale power consumption mode time sequence enhancement coding feature vector and the second-scale power consumption mode time sequence enhancement coding feature vector are subjected to feature decoding and double threshold correction respectively to obtain the first power consumption threshold and the second power consumption threshold. 3.The big data based power resource dispatching method according to claim 2, characterized in that, The power consumption data of the target power consumption unit in the sample date data is subjected to multi-scale power consumption mode time sequence coding to obtain first-scale power consumption mode time sequence coding feature vector and second-scale power consumption mode time sequence coding feature vector, which comprises: The power consumption data of the target power consumption unit in the sample date data is subjected to time sequence coding using a multi-scale power consumption mode sequence encoder based on an LSTM-1DCNN hybrid model to obtain the first-scale power consumption mode time sequence coding feature vector and the second-scale power consumption mode time sequence coding feature vector.

4. The big data based power resource dispatching method according to claim 3, characterized in that, The first-scale power consumption mode time sequence coding feature vector and the second-scale power consumption mode time sequence coding feature vector are subjected to feature time sequence expression enhancement respectively to obtain first-scale power consumption mode time sequence enhancement coding feature vector and second-scale power consumption mode time sequence enhancement coding feature vector, which comprises: The first-scale power consumption mode time sequence coding feature vector is subjected to feature component reorganization to obtain a set of first-scale power consumption mode time sequence feature local phase coding vectors; performing feature component saliency adjustment on the set of first scale power consumption pattern time series feature local phase coding vectors based on the feature effective component statistical characteristics of each first scale power consumption pattern time series feature local phase coding vector in the set of first scale power consumption pattern time series feature local phase coding vectors to obtain the first scale power consumption pattern time series enhanced coding feature vector.

5. The big data based power resource dispatching method according to claim 4, characterized in that, performing feature component saliency adjustment on the set of first scale power consumption pattern time series feature local phase coding vectors based on the feature effective component statistical characteristics of each first scale power consumption pattern time series feature local phase coding vector in the set of first scale power consumption pattern time series feature local phase coding vectors to obtain the first scale power consumption pattern time series enhanced coding feature vector. performing feature component saliency adjustment on the set of first scale power consumption pattern time series feature local phase coding vectors based on the feature effective component statistical characteristics of each first scale power consumption pattern time series feature local phase coding vector in the set of first scale power consumption pattern time series feature local phase coding vectors to obtain the first scale power consumption pattern time series enhanced coding feature vector. 6.The big data-based power resource dispatching method according to claim 5, wherein, performing feature component saliency adjustment on the set of first scale power consumption pattern time series feature local phase coding vectors based on the feature effective component statistical characteristics of each first scale power consumption pattern time series feature local phase coding vector in the set of first scale power consumption pattern time series feature local phase coding vectors to obtain the first scale power consumption pattern time series enhanced coding feature vector. performing feature component saliency adjustment on the set of first scale power consumption pattern time series feature local phase coding vectors based on the feature effective component statistical characteristics of each first scale power consumption pattern time series feature local phase coding vector in the set of first scale power consumption pattern time series feature local phase coding vectors to obtain the first scale power consumption pattern time series enhanced coding feature vector. calculating a feature saliency measure value of each first scale power consumption pattern time series feature extraction local phase coding vector in the set of first scale power consumption pattern time series feature extraction local phase coding vectors; calculating a feature component re-scaling gain factor of each first scale power consumption pattern time series feature extraction local phase coding vector based on the feature saliency measure value of the first scale power consumption pattern time series feature extraction local phase coding vector; performing feature component saliency adjustment on the set of first scale power consumption pattern time series feature local phase coding vectors based on the feature component re-scaling gain factor of each first scale power consumption pattern time series feature extraction local phase coding vector to obtain the first scale power consumption pattern time series enhanced coding feature vector. 7.The big data-based power resource dispatching method according to claim 6, wherein, performing feature component saliency adjustment on the set of first scale power consumption pattern time series feature local phase coding vectors based on the feature component re-scaling gain factor of each first scale power consumption pattern time series feature extraction local phase coding vector to obtain the first scale power consumption pattern time series enhanced coding feature vector, comprising: performing gain factor optimization based on effective component shift invariance on the feature component re-scaling gain factor of each first scale power consumption pattern time series feature extraction local phase coding vector to obtain a set of optimized first scale power consumption pattern time series feature component re-scaling gain factors; performing normalization processing on the set of optimized first scale power consumption pattern time series feature component re-scaling gain factors to obtain a set of first scale power consumption pattern time series feature phase reshaping gain weights; weight fusion is performed on the set of the first scale power consumption pattern time sequence local phase coding vectors to obtain a first scale power consumption pattern time sequence enhanced coding feature vector based on the set of the first scale power consumption pattern time sequence feature phase remodeling gain weights. 8.The big data based power resource dispatching method according to claim 7, wherein, The first scale power consumption pattern time sequence enhanced coding feature vector and the second scale power consumption pattern time sequence enhanced coding feature vector are respectively subjected to feature decoding and double criticality calibration to obtain the first power consumption criticality and the second power consumption criticality, including: The first scale power consumption pattern time sequence enhanced coding feature vector and the second scale power consumption pattern time sequence enhanced coding feature vector are respectively subjected to feature decoding to obtain a first power consumption criticality upper limit estimate and a first power consumption criticality lower limit estimate, and a second power consumption criticality upper limit estimate and a second power consumption criticality lower limit estimate. The mean value between the first power consumption criticality upper limit estimate and the second power consumption criticality upper limit estimate is calculated as the first power consumption criticality. The mean value between the first power consumption criticality lower limit estimate and the second power consumption criticality lower limit estimate is calculated as the second power consumption criticality. 9.The big data-based power resource dispatching method according to claim 8, wherein, The first scale power consumption pattern time sequence enhanced coding feature vector and the second scale power consumption pattern time sequence enhanced coding feature vector are respectively subjected to feature decoding to obtain a first power consumption criticality upper limit estimate and a first power consumption criticality lower limit estimate, and a second power consumption criticality upper limit estimate and a second power consumption criticality lower limit estimate, including: The first scale power consumption pattern time sequence enhanced coding feature vector is input into a power consumption criticality decoder based on a multi-layer perception machine to obtain the first power consumption criticality upper limit estimate and the first power consumption criticality lower limit estimate.

10. A big data based power resource allocation system, characterized in that, including: A sample date selection module is configured to select sample dates based on weather similarity between historical dates and a target power dispatch date to obtain sample date data in a power regulation area. A power consumption criticality screening module is configured to obtain a first power consumption criticality and a second power consumption criticality according to the sample date data, and screen a plurality of sample power consumption units having similar power consumption characteristics with a target power consumption unit according to the first power consumption criticality and the second power consumption criticality. A power consumption prediction and regulation module is configured to calculate same-period power consumption data and periodic power consumption data of the target power consumption unit and the plurality of sample power consumption units to create a power consumption prediction model for power prediction of the power regulation area, obtain a predicted power consumption of the power regulation area, and perform power regulation based on the predicted power consumption of the power regulation area. The power consumption criticality screening module is configured to perform multi-scale power consumption pattern time sequence feature enhancement perception and hierarchical decoding calibration on power consumption data of the target power consumption unit in the sample date data to obtain the first power consumption criticality and the second power consumption criticality.

Images