Electric power data intelligent processing method and system

Abstract

The invention relates to the technical field of power load prediction, and particularly discloses a power data intelligent processing method and system, and the method comprises the steps: determining the time sequence analysis granularity of historical power grid load data based on a power grid load prediction target, and carrying out the local time sequence granularity feature extraction of the historical power grid load data based on the time sequence analysis granularity; the method comprises the following steps: acquiring a time sequence change rule and a characteristic mode of power grid load historical data in each local time domain, and then carrying out cross-modal joint coding on a power grid load prediction target and power grid load historical time sequence characteristics in each local time domain; therefore, under the constraint of a power grid load prediction target, power grid load time sequence characteristic modes highly related to the power grid load prediction target are mined and aggregated, and intelligent prediction of the power grid load is realized on the basis. In this way, the reliability and generalization ability of load prediction can be effectively improved, the method adapts to complex and changeable power grid load prediction scenes, and reliable support is provided for power grid dispatching and resource optimization.

Description

Technical Field

[0001] This application relates to the field of power load forecasting technology, and more specifically, to a method and system for intelligent processing of power data. Background Technology

[0002] Against the backdrop of rapid development in power grid intelligence and renewable energy integration, load forecasting has become a crucial link in power system planning, dispatching, operation, and electricity market transactions. The accuracy of load forecasting directly affects the economic efficiency and security of power grid operation, and is an important technical support for ensuring grid stability and adaptability. However, with the increasing complexity of factors such as renewable energy integration, electric vehicle charging, and demand-side response, traditional load forecasting methods have revealed significant limitations when dealing with the time-series nonlinear relationships and dynamic changes in power grid load, making it difficult to meet the increasingly complex power load forecasting needs.

[0003] Specifically, traditional load forecasting methods typically employ a fixed time resolution (e.g., hourly or daily) to globally model historical data, neglecting the compatibility between the forecasting objective and the temporal correlation structure of historical data. For example, short-term frequency regulation scenarios require capturing minute-level load mutation characteristics, while medium- to long-term capacity planning relies on weekly / monthly trend evolution patterns. Traditional rigid time series partitioning ignores the inherent differences in time resolution requirements for different forecasting tasks, easily leading to temporal pattern aliasing or dilution of key information during feature extraction, resulting in the submergence of local dynamic features or global trend deviations. Furthermore, while traditional power grid load time series feature extraction methods can characterize the periodicity and trend of load changes through techniques such as sliding window decomposition and wavelet transform, they lack in-depth exploration of the load forecasting objective in feature engineering. This may lead to cognitive discrepancies between the power grid load time series feature space and the forecast space, further limiting the reliability and generalization ability of power grid load forecasting.

[0004] Therefore, there is a need for an optimized intelligent processing method and system for power data. Summary of the Invention

[0005] To address the aforementioned technical problems, this application is proposed. Embodiments of this application provide an intelligent power data processing method and system. First, it determines the temporal analysis granularity of historical power grid load data based on the power grid load forecasting target. Then, based on this, it extracts local temporal granular features from the historical power grid load data to capture the temporal variation patterns and characteristic patterns of the historical power grid load data in various local time domains. Subsequently, by performing cross-modal joint encoding between the power grid load forecasting target and the historical temporal features of the power grid load in each local time domain, it mines and aggregates highly correlated temporal feature patterns of the power grid load under the constraints of the power grid load forecasting target, thereby achieving intelligent forecasting of the power grid load. This approach effectively improves the reliability and generalization ability of load forecasting, adapts to complex and ever-changing power grid load forecasting scenarios, and provides reliable support for power grid scheduling and resource optimization.

[0006] According to one aspect of this application, a method for intelligent processing of power data is provided, comprising:

[0007] Obtain the target for power grid load forecasting;

[0008] The power grid load prediction target is semantically embedded and encoded to obtain the power grid load prediction target semantic embedding and encoding vector;

[0009] Based on the semantic embedding encoding vector of the power grid load prediction target, the temporal segmentation granularity of the power grid load is determined;

[0010] Historical power grid load data is extracted from the power grid load storage database, and local time-series pattern feature extraction based on the time-series segmentation granularity of the power grid load is performed on the historical power grid load data to obtain a sequence of local time-series granularity pattern feature encoding vectors of the power grid load.

[0011] The sequence of the target semantic embedding encoding vector for power grid load prediction and the local temporal granularity pattern feature encoding vector for power grid load is subjected to cross-modal joint encoding based on load prediction target constraints to obtain the temporal pattern feature encoding vector of power grid load under prediction target constraints;

[0012] Feature decoding is performed on the feature encoding vector of the time-series pattern of power grid load under the constraint of the prediction target to obtain the power grid load prediction value corresponding to the power grid load prediction target.

[0013] According to another aspect of this application, a power data intelligent processing system is provided, comprising:

[0014] The power grid load forecasting target acquisition module is used to acquire the power grid load forecasting target;

[0015] A semantic embedding encoding module is used to perform semantic embedding encoding on the power grid load prediction target to obtain a semantic embedding encoding vector for the power grid load prediction target.

[0016] The time-series segmentation granularity determination module is used to determine the time-series segmentation granularity of the power grid load based on the target semantic embedding encoding vector of the power grid load prediction.

[0017] The historical data feature extraction module is used to extract historical power grid load data from the power grid load storage database, and to perform local time series pattern feature extraction on the historical power grid load data based on the time series segmentation granularity of the power grid load to obtain a sequence of local time series granularity pattern feature encoding vectors of the power grid load.

[0018] A cross-modal joint coding module is used to perform cross-modal joint coding on the sequence of the target semantic embedding coding vector of the power grid load prediction and the local temporal granularity pattern feature coding vector of the power grid load based on the load prediction target constraint to obtain the temporal pattern feature coding vector of the power grid load under the prediction target constraint;

[0019] The feature decoding module is used to perform feature decoding on the feature encoding vector of the power grid load time series pattern under the prediction target constraint to obtain the power grid load prediction value corresponding to the power grid load prediction target.

[0020] Compared with existing technologies, the intelligent power data processing method and system provided in this application first determines the temporal analysis granularity of historical power grid load data based on the power grid load forecasting target. Then, based on this, it extracts local temporal granular features from the historical power grid load data to capture the temporal variation patterns and characteristic patterns of the historical power grid load data in various local time domains. Subsequently, by performing cross-modal joint encoding between the power grid load forecasting target and the historical temporal features of the power grid load in each local time domain, it mines and aggregates highly correlated temporal feature patterns of the power grid load under the constraints of the power grid load forecasting target, thereby achieving intelligent forecasting of the power grid load. This approach effectively improves the reliability and generalization ability of load forecasting, adapts to complex and ever-changing power grid load forecasting scenarios, and provides reliable support for power grid dispatching and resource optimization. Attached Figure Description

[0021] 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.

[0022] Figure 1This is a flowchart of an intelligent power data processing method according to an embodiment of this application.

[0023] Figure 2 This is a schematic diagram of the data flow in the intelligent power data processing method according to an embodiment of this application.

[0024] Figure 3 This is a flowchart of sub-step S4 of the intelligent power data processing method according to an embodiment of this application.

[0025] Figure 4 This is a flowchart of sub-step S5 of the intelligent power data processing method according to an embodiment of this application.

[0026] Figure 5 This is a flowchart of sub-step S51 of the intelligent power data processing method according to an embodiment of this application.

[0027] Figure 6 This is a flowchart of sub-step S52 of the intelligent power data processing method according to an embodiment of this application.

[0028] Figure 7 This is a block diagram of an intelligent power data processing system according to an embodiment of this application. Detailed Implementation

[0029] 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.

[0030] 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.

[0031] 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.

[0032] 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.

[0033] 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.

[0034] To address the technical problems described in the background, this application proposes an intelligent power data processing method. First, it determines the temporal analysis granularity of historical power grid load data based on the power grid load forecasting target. Then, based on this, it extracts local temporal granular features from the historical power grid load data to capture the temporal variation patterns and characteristic patterns of the historical power grid load data in various local time domains. Subsequently, by performing cross-modal joint encoding between the power grid load forecasting target and the historical temporal features of the power grid load in each local time domain, it mines and aggregates highly correlated temporal feature patterns of the power grid load under the constraint of the power grid load forecasting target, thereby achieving intelligent forecasting of the power grid load. This approach effectively improves the reliability and generalization ability of load forecasting, adapts to complex and ever-changing power grid load forecasting scenarios, and provides reliable support for power grid dispatching and resource optimization.

[0035] Figure 1 This is a flowchart of an intelligent power data processing method according to an embodiment of this application. Figure 2 This is a schematic diagram of the data flow in the intelligent power data processing method according to an embodiment of this application. Figure 1 and Figure 2 As shown, the intelligent power data processing method includes the following steps: S1, obtaining the power grid load prediction target; S2, performing semantic embedding encoding on the power grid load prediction target to obtain a power grid load prediction target semantic embedding encoding vector; S3, determining the power grid load time series segmentation granularity based on the power grid load prediction target semantic embedding encoding vector; S4, extracting historical power grid load data from the power grid load storage database, and performing local time series pattern feature extraction on the historical power grid load data based on the power grid load time series segmentation granularity to obtain a sequence of power grid load local time series granularity pattern feature encoding vectors; S5, performing cross-modal joint encoding based on load prediction target constraints on the sequence of the power grid load prediction target semantic embedding encoding vector and the power grid load local time series granularity pattern feature encoding vectors to obtain a power grid load time series pattern feature encoding vector under prediction target constraints; S6, performing feature decoding on the power grid load time series pattern feature encoding vector under prediction target constraints to obtain the power grid load prediction value corresponding to the power grid load prediction target.

[0036] In the aforementioned intelligent power data processing method, step S1 involves obtaining the power grid load forecast target. It should be understood that using a fixed time resolution for global modeling of historical data ignores the inherent differences in time resolution requirements for different forecasting tasks, leading to problems such as temporal pattern aliasing and dilution of key information in feature extraction, thus failing to meet the needs of complex load forecasting. Therefore, this application obtains the power grid load forecast target to provide directional guidance for subsequent analysis and processing based on the power grid load forecast target to obtain the power grid load forecast value, making the entire forecasting process more targeted and improving the reliability and generalization ability of load forecasting.

[0037] Specifically, obtaining grid load forecasting targets first relies on a deep understanding of power system operating scenarios. Load forecasting needs differ significantly across scenarios. For example, short-term frequency regulation scenarios focus on minute-level or hourly load fluctuation characteristics, while medium- and long-term capacity planning emphasizes weekly or monthly trend evolution patterns. Therefore, clarifying the forecast timescale is the first step in obtaining forecasting targets. Through communication with stakeholders such as grid dispatching departments and market trading institutions, the time frame of the forecasting task can be accurately defined, and the specific content of the forecasting targets can be further refined. For example, for short-term frequency regulation scenarios, forecasting targets might include the load peak value and its changing trend within the next 15 minutes; while for medium- and long-term planning, it might involve the seasonal fluctuation patterns and average annual growth rate of load over the next year.

[0038] In practical implementation, obtaining grid load forecast targets also requires full consideration of the target-oriented nature of the forecasting task. That is, the setting of forecast targets should be closely centered on specific business needs. For example, in the electricity market trading scenario, the forecast target may focus more on the peak-valley difference of the load and its distribution pattern to optimize power generation plans and resource allocation; while in the grid dispatching scenario, the forecast target may focus on the real-time fluctuation characteristics of the load and its impact on system stability. Furthermore, obtaining grid load forecast targets also requires attention to the introduction of dynamic adjustment and feedback mechanisms. Since grid load is affected by various uncertainties, the setting of forecast targets cannot remain static. Therefore, a rolling update mechanism can be established to periodically revise the forecast targets based on the latest operating data and changes in the external environment. For example, when a sudden weather event occurs, the forecast target can be quickly adjusted to reflect its potential impact on the load; when the proportion of renewable energy integration changes, the forecast target can also be updated accordingly to adapt to the new load characteristics. Simultaneously, by introducing a closed-loop feedback mechanism, the actual performance of the forecast results can be compared and analyzed with the expected target to identify problems and continuously optimize the definition of the forecast target. This dynamic adjustment capability is particularly important for coping with complex and ever-changing grid environments.

[0039] In the aforementioned intelligent power data processing method, step S2 involves semantic embedding encoding of the power grid load prediction target to obtain a semantic embedding encoding vector for the power grid load prediction target. That is, semantic embedding encoding establishes the connection between the power grid load prediction target and subsequent power grid load data analysis and processing, providing a crucial digital foundation for subsequent analysis and processing around the power grid load prediction target. This enables the mining of power grid load time-series feature patterns closely related to the power grid load prediction target based on the generated semantic embedding encoding vector, thereby effectively improving the reliability and generalization ability of load prediction. In a specific example of this application, the power grid load prediction target is input into a semantic embedding encoder based on the BERT model to obtain the semantic embedding encoding vector for the power grid load prediction target. It should be understood that the BERT model employs a bidirectional Transformer encoder structure, capable of simultaneously capturing the left and right context information of words. Utilizing its powerful language understanding and feature extraction capabilities, it performs deep semantic analysis on the power grid load prediction target. The resulting semantic embedding encoding vector for the power grid load prediction target accurately represents the semantic connotation of the power grid load prediction target, providing a reliable basis for subsequent analysis and processing based on prediction target constraints.

[0040] In the aforementioned intelligent power data processing method, step S3 determines the granularity of power grid load time series segmentation based on the semantic embedding encoding vector of the power grid load prediction target. Specifically, according to the semantic characteristics of the power grid load prediction target, a granularity of power grid load time series segmentation that meets its requirements is determined to avoid deviations in time series feature extraction. The resulting granularity of power grid load time series segmentation can specifically capture the time series variation patterns and characteristic patterns of historical power grid load data in an appropriate time domain, avoiding the dilution of key information or the aliasing of time series patterns due to the use of inappropriate time resolution. This lays the foundation for improving the reliability and generalization ability of power grid load prediction, and better adapts to complex and ever-changing power grid load prediction scenarios. In the embodiments of this application, a multilayer perceptron model is used to perform feature analysis on the semantic embedding encoding vector of the power grid load prediction target. Through multilayer nonlinear transformation, the potential semantic information of the power grid load prediction target is deeply mined, and the granularity parameters of power grid load time series segmentation that meet the requirements of the prediction target are generated accordingly, thereby improving the pertinence and adaptability of the subsequent time series analysis process of historical power grid load data.

[0041] In the aforementioned intelligent power data processing method, step S4 involves extracting historical power grid load data from the power grid load storage database and performing local time-series pattern feature extraction based on the granularity of the power grid load time-series segmentation on the historical power grid load data to obtain a sequence of local time-series granularity pattern feature encoding vectors for the power grid load. Specifically, since using a fixed time resolution for global modeling of historical data neglects the adaptability issue between the power grid load prediction target and the temporal correlation structure of the historical power grid load data, and lacks in-depth mining of the guidance of the power grid load prediction target in feature engineering, this application performs local time-series pattern feature extraction based on the granularity of the power grid load time-series segmentation on the historical power grid load data. This captures the temporal variation patterns and characteristic patterns of the historical power grid load data in various local time domains, thereby mining more effective features related to the power grid load prediction target and providing a foundation for achieving more reliable power grid load prediction. Figure 3 This is a flowchart of sub-step S4 of the intelligent power data processing method according to an embodiment of this application. Figure 3 As shown, step S4 includes the following steps: S41, performing data segmentation on the historical power grid load data based on the power grid load time-series segmentation granularity to obtain a sequence of local time-series granularity distributions of power grid load data; S42, extracting the power grid load time-series pattern features of each local time-series granularity distribution of power grid load data from the sequence of local time-series granularity distributions of power grid load data to obtain a sequence of local time-series granularity pattern feature encoding vectors of power grid load data.

[0042] Specifically, step S41 involves segmenting the historical power grid load data based on the time-series segmentation granularity to obtain a sequence of local time-series granularity distributions of the power grid load data. More specifically, by segmenting the historical power grid load data based on the time-series segmentation granularity (e.g., minute-level, weekly / monthly-level) determined by the power grid load prediction target, the global data is divided into local time-domain data units that meet the time resolution requirements of different prediction tasks. This forms a sequence reflecting the distribution pattern of the power grid load data at a specific granularity, allowing subsequent analysis to focus on local time-domain features that match the prediction target, avoiding feature confusion or information loss caused by rigid time-series segmentation.

[0043] Specifically, step S42 involves extracting the power grid load time-series pattern features of each local time-series granularity distribution of the power grid load data from the sequence of local time-series granularity distributions to obtain a sequence of local time-series granularity pattern feature encoding vectors for the power grid load. Specifically, directly modeling historical data globally can lead to the loss of key dynamic features due to aliasing or information dilution of time-series patterns at different time granularities. Therefore, this application independently extracts features for each local time-series granularity distribution of the power grid load data to avoid the problem of local dynamic features being submerged under rigid global partitioning. This yields a sequence of local time-series granularity pattern feature encoding vectors for the power grid load, representing the regularity of load time-series changes within the segment. This explicitly separates time-series patterns at different time granularities and eliminates the interference of pattern confusion in global time-series modeling. Simultaneously, local feature encoding preserves the nonlinear dynamic details of the original power grid load data, providing target-oriented local time-series features for subsequent cross-modal joint encoding, thereby improving the load forecasting model's ability to capture complex dynamic scenarios. In a specific example of this application, the local temporal granular distribution of the power grid load data is subjected to one-dimensional convolutional coding-based extraction of power grid load temporal pattern features to obtain the local temporal granular pattern feature encoding vector of the power grid load. That is, by using the self-learning weights of the one-dimensional convolutional kernel to provide differentiated feature responses to load fluctuations at each local temporal granularity, it is possible to simultaneously capture high-frequency spikes and low-frequency slowly varying features in the load data, avoiding pattern aliasing caused by manually pre-setting the temporal segmentation granularity. Furthermore, the convolutional layer, through nonlinear activation and cross-layer feature transfer, can enhance the model's ability to learn complex temporal features, hierarchically decoupling noise disturbances and key patterns in the original load temporal series, forming a semantically robust local temporal granular pattern feature encoding vector of the power grid load.

[0044] In the aforementioned intelligent power data processing method, step S5 involves performing cross-modal joint encoding based on load prediction target constraints on the sequences of the target semantic embedding encoding vector and the local temporal granularity pattern feature encoding vector of the power grid load to obtain the power grid load temporal pattern feature encoding vector under the prediction target constraints. It should be understood that by establishing a correlation feature representation between the target semantic embedding information and the local temporal granularity pattern features through cross-modal joint encoding, the load temporal feature patterns highly correlated with the power grid load prediction target are mined and strengthened, while irrelevant or secondary features are eliminated. The resulting power grid load temporal pattern feature encoding vector under the prediction target constraints contains both historical load dynamics and feature representations closely related to the prediction target, providing a more accurate and reliable feature foundation for subsequent feature decoding and power grid load prediction, thereby improving the reliability and generalization ability of load prediction. Figure 4 This is a flowchart of sub-step S5 of the intelligent power data processing method according to an embodiment of this application. Figure 4As shown, step S5 includes the following steps: S51, performing cross-scale feature alignment and semantic interactive encoding on each local temporal granularity pattern feature encoding vector of the power grid load prediction target semantic embedding encoding vector and the local temporal granularity pattern feature encoding vector of the power grid load to obtain a set of power grid load semantic query score encoding vectors; S52, based on the feature set self-distribution characteristics of the set of the set of power grid load semantic query score encoding vectors, performing semantic association dynamic gating fusion on the set of the set of power grid load semantic query score encoding vectors to obtain the power grid load temporal pattern feature encoding vector under the prediction target constraint.

[0045] Figure 5 This is a flowchart of sub-step S51 of the intelligent power data processing method according to an embodiment of this application. Figure 5 As shown, step S51 includes the following steps: S511, performing feature enhancement based on deconvolution coding on the target semantic embedding encoding vector for power grid load prediction to obtain a target semantic feature enhancement encoding vector for power grid load prediction, wherein the target semantic feature enhancement encoding vector for power grid load prediction has the same feature scale as each local temporal granularity pattern feature encoding vector of power grid load in the sequence of local temporal granularity pattern feature encoding vectors for power grid load; S512, inputting each local temporal granularity pattern feature encoding vector of power grid load in the sequence of the target semantic feature enhancement encoding vector for power grid load prediction and the local temporal granularity pattern feature encoding vector of power grid load into a single semantic query unit to obtain a set of semantic query score encoding vectors for power grid load.

[0046] More specifically, step S511 is expressed by the formula:

[0047]

[0048] Where V1 represents the semantic embedding encoding vector of the power grid load forecast target, V 1′ W represents the enhanced encoding vector of semantic features of the power grid load forecast target. deconv Let f represent the deconvolution weight matrix. deconv (·) represents the feature enhancement function based on deconvolution coding, and ‖·‖ represents the computation norm.

[0049] In other words, since there may be differences in feature scale between the target semantic embedding encoding vector of power grid load prediction and the local temporal granularity pattern feature encoding vector of power grid load, this application introduces deconvolution encoding to enable the model to adaptively expand the feature dimension of the target semantic embedding encoding vector of power grid load prediction. While enhancing its semantic expressive power, it also enables the enhanced encoding vector of the target semantic features of power grid load prediction to form an interactive alignment structure with the local temporal granularity pattern feature encoding vector of power grid load in the spatial dimension.

[0050] More specifically, step S512 is expressed by the formula:

[0051] V2={V 21 V 22 ,...,V 2i ,...,V 2n}

[0052] R i =tanh{W Ri [V 1′ V 2i ]+b i}

[0053] Where V2 represents the set of local time-series granularity pattern feature encoding vectors of the power grid load, V 21 V 22 V 2i and V 2n Let represent the 1st, 2nd, 1st, and 1st local time-series granularity pattern feature encoding vectors of the power grid load, respectively, where n is the number of such vectors, tanh(·) denotes the tanh function, and b i W represents the bias term. Ri R represents the weight matrix of a neural network layer, [·; ·] denotes cascading operations. i V represents 1′ and V 2i The semantic query score encoding vector of the power grid load between them.

[0054] In other words, by using a single semantic query unit, the implicit semantic relationship between the semantic feature enhancement encoding vector of each power grid load prediction target and the local temporal granularity pattern feature encoding vector of the power grid load is modeled in multiple dimensions, rather than relying solely on a single similarity index. This comprehensively characterizes the degree of semantic relationship between the prediction target and each local temporal feature in different dimensions. The resulting set of power grid load semantic query score encoding vectors provides fine-grained semantic relationship basis for subsequent feature aggregation based on prediction target constraints, enabling the model to specifically focus on temporal patterns that are highly related to the prediction target.

[0055] Figure 6 This is a flowchart of sub-step S52 of the intelligent power data processing method according to an embodiment of this application. Figure 6As shown, step S52 includes the following steps: S521, performing prediction constraint optimization based on the prediction target-grid load interaction mechanism on each grid load semantic query score encoding vector in the set of grid load semantic query score encoding vectors to obtain a set of optimized grid load semantic query score encoding vectors; S522, performing gated aggregation on the set of optimized grid load semantic query score encoding vectors to obtain the grid load time series pattern feature encoding vector under the prediction target constraint.

[0056] Here, the single semantic query unit uses the enhanced encoding vector V of the target semantic features of the power grid load prediction to strengthen the encoding vector. 1′ and the corresponding local temporal granularity pattern feature encoding vector V of the power grid load 2i When encoding a single semantic query score by directly concatenating features, it is desirable to improve the representation accuracy of the single semantic query score through optimization of the dynamic characteristic representation from the feature space to the semantic query encoding space and enhancement of the inherent alignment characteristics between the feature space and the semantic query encoding space. Based on this, in a preferred example of this application, step S52 includes: S521, performing prediction constraint optimization based on the prediction target-grid load interaction mechanism on each power grid load semantic query score encoding vector in the set of power grid load semantic query score encoding vectors to obtain an optimized set of power grid load semantic query score encoding vectors.

[0057] Specifically, if the target semantic feature enhancement encoding vector V of the power grid load prediction is... 1′ and the corresponding local temporal granularity pattern feature encoding vector V of the power grid load 2i The cascade feature is denoted as V 3i V 3i =[V 1′ V 2i First, construct the interaction potential vector of the mapping space:

[0058] V 4i =W i V 3i -V 3i

[0059] Among them, V 3i Indicate V1' and V 2i The concatenated vector of grid load forecasting and local time-series features between W i V represents the interaction potential weight matrix. 4i Indicate V1' and V 2i The interaction potential vector between them.

[0060] That is, the cascaded feature ontology is used as a time-varying feature basis to obtain its feature state mapping representation in the spatial mapping process.

[0061] Then, an inherent alignment is performed between the feature space and the semantic query encoding space:

[0062] R′ i =R i +gS i V 4i

[0063] Where g represents the coupling constant, S i Let R' be a covariant matrix. i This represents the i-th optimized power grid load semantic query score encoding vector in the set of optimized power grid load semantic query score encoding vectors.

[0064] Finally, the coupling constant g is calculated, for example, in the same way as during deconvolution enhancement, to maintain symmetry, i.e.:

[0065]

[0066] And S i Let V be a covariant matrix. 4i =S i R i To obtain.

[0067] Therefore, in the interaction potential vector V 4i As an inherent alignment generator, constructing a covariant mapping trajectory under metric conservation constraints as a guarantee of projective metric constraints significantly improves the representation accuracy of the monolithic semantic query score.

[0068] More specifically, in a specific example of this application, step S522 includes: first, calculating the power grid load semantic matching degree of each optimized power grid load semantic query score encoding vector in the set of optimized power grid load semantic query score encoding vectors to obtain a set of power grid load semantic matching degrees, expressed by the formula:

[0069]

[0070] Among them, R k Let k represent the k-th optimized power grid load semantic query score encoding vector in the set of optimized power grid load semantic query score encoding vectors, (·) T The vector transpose is represented by exp(·), which represents the exponential function operation with base e, and softmax(·) represents the normalized exponential function. i R′ i The corresponding semantic matching degree of the power grid load.

[0071] In other words, by optimizing the overall distribution characteristics of the semantic query score encoding vector for power grid load, the model adaptively adjusts each semantic matching degree in a context-aware manner. This allows the model to assign corresponding weights based on the prominence of features within the overall context, increasing the weight of features that are significantly above average and have strong semantic relevance, while reducing the impact of weakly correlated or noisy features. This constructs a feature importance ranking that meets the needs of the prediction target. This calculation method avoids the one-sidedness of evaluating features in isolation, enabling the generated power grid load semantic matching degree to accurately reflect the true correlation strength between the prediction target and each local time-series feature. This helps the model focus on high-value features and suppress interference information, thereby more efficiently selecting time-series patterns highly relevant to the prediction target during cross-modal feature aggregation, and improving the adaptability of the load prediction model to complex dynamic scenarios and the efficiency of feature utilization.

[0072] Then, the set of semantic matching degrees of the power grid load is input into the relation gating agent module to obtain the set of semantic self-attention weights of the power grid load query, which is expressed by the formula:

[0073]

[0074] Where τ represents the gating threshold, mask(·) represents the masking operation, and w i Indicates a i The corresponding self-attention weights for the power grid load query semantics.

[0075] In other words, by using a relation-gated proxy module as an intermediate layer, the semantic matching degree of the power grid load is selectively retained or filtered, enabling fine-grained control over the modeling of complex semantic relationships. This makes the set of generated power grid load query semantic self-attention weights more closely match the precise requirements of the self-attention mechanism for feature weight allocation. Specifically, through the dynamic adjustment of the gating mechanism, the model can more flexibly select core semantic matching information that is highly relevant to the prediction target and filter out irrelevant or weakly correlated interference signals. This allows for precise capture of the differences in the contribution of different local temporal features to the prediction target in the self-attention weight allocation, providing more targeted weight guidance for subsequent cross-modal feature aggregation. This enhances the model's ability to model multi-dimensional semantic relationships and feature interaction efficiency in complex power grid load scenarios, thereby improving the accuracy and robustness of load prediction.

[0076] Finally, based on the set of self-attention weights for the power grid load query semantics, the set of optimized power grid load semantic query score encoding vectors is aggregated to obtain the power grid load time-series pattern feature encoding vector under the prediction target constraint, expressed by the formula:

[0077] v p =∑ i w i R′ i

[0078] Among them, v p This represents the feature encoding vector of the time-series pattern of power grid load under the constraint of the prediction target.

[0079] In other words, by applying attention weights to the differentiated weighted aggregation of the semantic query score encoding vectors for each optimized power grid load, the model can proactively focus on and integrate local temporal pattern features highly relevant to the actual needs of the prediction target. This transforms scattered, multi-dimensional semantic matching information into a unified and strongly target-oriented comprehensive feature representation. The resulting power grid load temporal pattern feature encoding vector under the constraint of the prediction target retains key dynamic features and trend information at different time granularities while suppressing irrelevant or weakly correlated interference components. It can accurately map the semantic space of the load prediction target, providing a more targeted and discriminative feature foundation for subsequent feature decoding and generation. This effectively improves the model's ability to capture load change patterns in complex scenarios and enhances the reliability and generalization performance of the prediction results.

[0080] In the aforementioned intelligent power data processing method, step S6 involves feature decoding of the power grid load time-series pattern feature encoding vector under the prediction target constraint to obtain the power grid load prediction value corresponding to the power grid load prediction target. In a specific example of this application, step S6 includes: inputting the power grid load time-series pattern feature encoding vector under the prediction target constraint into a feature decoder based on an LSTM model to obtain the power grid load prediction value. It should be understood that the LSTM (Long Short-Term Memory) model is a special type of recurrent neural network specifically designed to process long-term dependencies in sequential data. In this application, the LSTM model, as the core component of the feature decoder, utilizes its unique gating mechanism (input gate, forget gate, output gate) to capture the long-distance temporal dependencies and nonlinear dynamic changes in the power grid load time-series pattern feature encoding vector under the prediction target constraint, transforming the abstract encoding vector, which integrates the semantic information of the prediction target and local temporal features, into a specific prediction value that conforms to the temporal evolution logic of the power grid load. By selectively retaining and forgetting historical states through memory units, LSTM models can effectively handle long-term dependent features such as periodicity and seasonality in power grid load data. At the same time, they can filter noise and enhance key time-series patterns, achieving accurate mapping from the feature space to the prediction space. This provides quantitative output for power grid load forecasting that is both time-series continuous and goal-oriented, improving the ability of the prediction results to fit actual load fluctuations and adapt to complex scenarios.

[0081] In summary, the intelligent power data processing method based on the embodiments of this application is explained. First, it determines the temporal analysis granularity of historical power grid load data based on the power grid load forecasting target. Then, based on this, it extracts local temporal granular features from the historical power grid load data to capture the temporal variation patterns and characteristic patterns of the historical power grid load data in various local time domains. Subsequently, by performing cross-modal joint encoding between the power grid load forecasting target and the historical temporal features of the power grid load in each local time domain, it mines and aggregates highly correlated temporal feature patterns of the power grid load under the constraints of the power grid load forecasting target, thereby achieving intelligent forecasting of the power grid load. This approach effectively improves the reliability and generalization ability of load forecasting, adapts to complex and ever-changing power grid load forecasting scenarios, and provides reliable support for power grid scheduling and resource optimization.

[0082] Furthermore, an intelligent power data processing system is also provided.

[0083] Figure 7 This is a block diagram of an intelligent power data processing system according to an embodiment of this application. Figure 7 As shown, the intelligent power data processing system 100 according to an embodiment of this application includes: a power grid load prediction target acquisition module 110, used to acquire a power grid load prediction target; a semantic embedding encoding module 120, used to perform semantic embedding encoding on the power grid load prediction target to obtain a power grid load prediction target semantic embedding encoding vector; a time-series segmentation granularity determination module 130, used to determine the power grid load time-series segmentation granularity based on the power grid load prediction target semantic embedding encoding vector; and a historical data feature extraction module 140, used to extract historical power grid load data from a power grid load storage database and perform feature extraction on the historical power grid load data based on the target semantic embedding encoding vector. The system extracts local time-series pattern features at the time-series segmentation granularity of the power grid load to obtain a sequence of local time-series granularity pattern feature encoding vectors for the power grid load; a cross-modal joint encoding module 150 is used to perform cross-modal joint encoding based on load prediction target constraints on the sequence of the semantic embedding encoding vector of the power grid load prediction target and the local time-series granularity pattern feature encoding vector of the power grid load to obtain a power grid load time-series pattern feature encoding vector under the prediction target constraints; a feature decoding module 160 is used to perform feature decoding on the power grid load time-series pattern feature encoding vector under the prediction target constraints to obtain a power grid load prediction value corresponding to the power grid load prediction target.

[0084] Here, those skilled in the art will understand that the specific operations of each module in the aforementioned intelligent power data processing system have been referenced above. Figures 1 to 6 The method for intelligent processing of power data has been described in detail, and therefore, its repeated description will be omitted.

[0085] 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.

[0086] In the above embodiments, the descriptions of each embodiment have their own emphasis. Parts not detailed or described in a particular embodiment can be referred to in the relevant descriptions of other embodiments. In the several embodiments provided by this invention, it should be understood that the disclosed systems and methods can be implemented in other ways. For example, the system embodiments described above are merely illustrative; for example, the unit division is only a logical functional division, and other division methods may exist in actual implementation. The units described as separate components may or may not be physically separated. The components shown as units may or may not be physical units, that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0087] 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.

[0088] Furthermore, it is clear that the word "comprising" does not exclude other units or steps, and the singular does not exclude the plural. Multiple units stated in a system claim may also be implemented by a single unit through software or hardware.

[0089] 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 method for intelligent processing of power data, characterized in that, include: Obtain the target for power grid load forecasting; The power grid load prediction target is semantically embedded and encoded to obtain the power grid load prediction target semantic embedding and encoding vector; Based on the semantic embedding encoding vector of the power grid load prediction target, the temporal segmentation granularity of the power grid load is determined; Historical power grid load data is extracted from the power grid load storage database, and local time-series pattern feature extraction based on the time-series segmentation granularity of the power grid load is performed on the historical power grid load data to obtain a sequence of local time-series granularity pattern feature encoding vectors of the power grid load. The sequence of the target semantic embedding encoding vector for power grid load prediction and the local temporal granularity pattern feature encoding vector for power grid load is subjected to cross-modal joint encoding based on load prediction target constraints to obtain the temporal pattern feature encoding vector of power grid load under prediction target constraints; Feature decoding is performed on the feature encoding vector of the time-series pattern of power grid load under the constraint of the prediction target to obtain the power grid load prediction value corresponding to the power grid load prediction target.

2. The intelligent power data processing method according to claim 1, characterized in that, The semantic embedding encoding of the power grid load forecasting target is performed to obtain the power grid load forecasting target semantic embedding encoding vector, including: The target load prediction of the power grid is input into a semantic embedding encoder based on the Bert model to obtain the semantic embedding encoding vector of the target load prediction of the power grid.

3. The intelligent power data processing method according to claim 2, characterized in that, Historical power grid load data is extracted from the power grid load storage database, and local time-series pattern feature extraction based on the time-series segmentation granularity of the power grid load is performed on the historical power grid load data to obtain a sequence of local time-series granularity pattern feature encoding vectors of the power grid load, including: The historical data of the power grid load is segmented based on the time-series segmentation granularity of the power grid load to obtain a sequence of local time-series granularity distribution of the power grid load data; Extract the power grid load time-series pattern features of each power grid load data local time-series granularity distribution from the sequence of the power grid load data local time-series granularity distribution to obtain the sequence of the power grid load local time-series granularity pattern feature encoding vector.

4. The intelligent power data processing method according to claim 3, characterized in that, Extracting the power grid load time-series pattern features of each local time-series granularity distribution of the power grid load data from the sequence of local time-series granularity distributions of the power grid load data to obtain a sequence of local time-series granularity pattern feature encoding vectors of the power grid load, including: The local temporal granularity distribution of the power grid load data is subjected to one-dimensional convolutional coding to extract the power grid load temporal pattern feature to obtain the local temporal granularity pattern feature encoding vector of the power grid load.

5. The intelligent power data processing method according to claim 4, characterized in that, The sequence of the target semantic embedding encoding vector for power grid load prediction and the local temporal granularity pattern feature encoding vector for power grid load is subjected to cross-modal joint encoding based on load prediction target constraints to obtain the temporal pattern feature encoding vector of power grid load under prediction target constraints, including: Cross-scale feature alignment and semantic interactive encoding are performed on each local temporal granularity pattern feature encoding vector of the power grid load prediction target semantic embedding encoding vector and the local temporal granularity pattern feature encoding vector of the power grid load to obtain a set of power grid load semantic query score encoding vectors; Based on the self-distribution characteristics of the feature set of the set of semantic query score encoding vectors of the power grid load, the set of semantic association dynamic gating fusion is performed on the set of semantic query score encoding vectors of the power grid load to obtain the feature encoding vector of the time series pattern of the power grid load under the constraint of the prediction target.

6. The intelligent power data processing method according to claim 5, characterized in that, A set of power grid load semantic query score encoding vectors is obtained by performing cross-scale feature alignment and semantic cross-coding on each of the power grid load local time-series granularity pattern feature encoding vectors in the sequence of the power grid load prediction target semantic embedding encoding vector and the power grid load local time-series granularity pattern feature encoding vector, including: The target semantic embedding encoding vector for power grid load prediction is enhanced by deconvolutional coding to obtain a target semantic feature enhancement encoding vector for power grid load prediction. The target semantic feature enhancement encoding vector for power grid load prediction has the same feature scale as each local temporal granularity pattern feature encoding vector of power grid load in the sequence of local temporal granularity pattern feature encoding vectors of power grid load. Each local temporal granularity pattern feature encoding vector of the power grid load prediction target semantic feature enhancement encoding vector and the local temporal granularity pattern feature encoding vector of the power grid load are respectively input into the single semantic query unit to obtain the set of the power grid load semantic query score encoding vector.

7. The intelligent power data processing method according to claim 6, characterized in that, Based on the self-distribution characteristics of the feature set of the set of semantic query score encoding vectors of the power grid load, semantic association dynamic gating fusion is performed on the set of semantic query score encoding vectors of the power grid load to obtain the feature encoding vector of the time series pattern of the power grid load under the constraint of the prediction target, including: The set of power grid load semantic query score encoding vectors is optimized by prediction constraint based on the prediction target-power grid load interaction mechanism to obtain the set of optimized power grid load semantic query score encoding vectors. The set of optimized power grid load semantic query score encoding vectors is gated and aggregated to obtain the power grid load time-series pattern feature encoding vector under the prediction target constraint.

8. The intelligent power data processing method according to claim 7, characterized in that, Gated aggregation is performed on the set of optimized power grid load semantic query score encoding vectors to obtain the power grid load time-series pattern feature encoding vector under the prediction target constraint, including: Calculate the power grid load semantic matching degree of each optimized power grid load semantic query score encoding vector in the set of optimized power grid load semantic query score encoding vectors to obtain the set of power grid load semantic matching degrees; The set of semantic matching degrees of the power grid load is input into the relation gating agent module to obtain the set of semantic self-attention weights of the power grid load query. The set of optimized power grid load semantic query score encoding vectors is aggregated based on the set of self-attention weights of the power grid load query semantics to obtain the power grid load time-series pattern feature encoding vector under the prediction target constraint.

9. The intelligent power data processing method according to claim 8, characterized in that, Decoding the feature encoding vector of the time-series pattern of the power grid load under the constraint of the prediction target to obtain the predicted power grid load value corresponding to the prediction target includes: The feature encoding vector of the time-series pattern of power grid load under the predicted target constraint is input into the feature decoder based on the LSTM model to obtain the predicted value of power grid load.

10. An intelligent power data processing system, characterized in that, include: The power grid load forecasting target acquisition module is used to acquire the power grid load forecasting target; A semantic embedding encoding module is used to perform semantic embedding encoding on the power grid load prediction target to obtain a semantic embedding encoding vector for the power grid load prediction target. The time-series segmentation granularity determination module is used to determine the time-series segmentation granularity of the power grid load based on the target semantic embedding encoding vector of the power grid load prediction. The historical data feature extraction module is used to extract historical power grid load data from the power grid load storage database, and to perform local time series pattern feature extraction on the historical power grid load data based on the time series segmentation granularity of the power grid load to obtain a sequence of local time series granularity pattern feature encoding vectors of the power grid load. A cross-modal joint coding module is used to perform cross-modal joint coding on the sequence of the target semantic embedding coding vector of the power grid load prediction and the local temporal granularity pattern feature coding vector of the power grid load based on the load prediction target constraint to obtain the temporal pattern feature coding vector of the power grid load under the prediction target constraint; The feature decoding module is used to perform feature decoding on the feature encoding vector of the power grid load time series pattern under the prediction target constraint to obtain the power grid load prediction value corresponding to the power grid load prediction target.

Images