A power distribution system energy efficiency dynamic optimization and energy saving control method for mass time series data

Abstract

The application belongs to the technical field of data processing and management, and specifically discloses a power distribution system energy efficiency dynamic optimization and energy-saving control method for massive time-series data, which comprises the following steps: obtaining massive time-series data of a power distribution system and performing adaptive hierarchical compression to generate compressed data; dynamically extracting time, energy efficiency and device state multi-dimensional features of the compressed data to generate a dynamic feature vector; inputting the feature vector into an adaptive optimization model to predict an energy efficiency trend and generate a prediction result through a dynamic weight mechanism; generating a real-time control strategy based on the prediction result and executing device parameter adjustment; collecting feedback data, and when an energy efficiency deviation exceeds a threshold, using feedback to synchronously optimize compression operation and the prediction model. The application significantly improves the integrity and continuity of energy efficiency optimization, greatly improves the prediction accuracy through differentiated data compression and a dynamic weight mechanism, and realizes collaborative optimization of energy saving, stability and cost targets.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of data processing and management, and relates to a power distribution system energy efficiency dynamic optimization and energy saving control method for massive time series data. BACKGROUND

[0002] In modern commercial and public utility management, the power distribution system as a key link connecting the power grid and the end user, its energy efficiency management level is directly related to the energy utilization efficiency and operating cost. With the development of smart grid technology, the power distribution system has generated massive, multi-source and heterogeneous time series data, which contains great potential for improving system operation efficiency. How to use advanced data processing methods to analyze, predict and manage these data to realize fine energy efficiency optimization and energy saving control has become an important issue in the field of power operation and management.

[0003] At present, the technical scheme for power distribution system energy efficiency optimization usually adopts a relatively fixed data processing and analysis mode. For example, in the data processing link, a unified compression algorithm is often used to process all time series data without distinguishing the dynamic characteristics of the data. In terms of energy efficiency prediction, the existing models are mostly based on single or limited features for trend extrapolation, and the model weight or structure is usually kept unchanged after deployment. In terms of control strategy generation, most methods focus on a single energy saving goal, such as minimizing line loss, by adjusting device parameters.

[0004] However, the above existing technical scheme has obvious defects in actual application. The unified data compression method cannot take into account the characteristics of different data, which may lead to the loss of key mutation information or the coexistence of redundant slowly changing data, affecting the data quality. The static prediction model relying on limited features is difficult to capture the complex dynamic correlation of the power distribution system, and its prediction accuracy will decrease significantly when the operating conditions change. In addition, the control strategy focusing on a single energy saving goal often ignores the impact on system voltage stability or device operating cost, which may cause potential operation risks or increase long-term maintenance costs, lacking comprehensive operation and management considerations. SUMMARY

[0005] In order to overcome the above-mentioned defects of the prior art and achieve the above-mentioned purpose, the application provides the following technical scheme: a power distribution system energy efficiency dynamic optimization and energy saving control method for massive time series data, comprising: A1, acquiring massive time series data of the power distribution system, and performing adaptive hierarchical compression operation on the massive time series data to generate compressed data.

[0006] A2, performing multi-dimensional feature dynamic extraction on the compressed data to generate a dynamic feature vector.

[0007] A3, input the dynamic feature vector into the adaptive optimization model, and perform energy efficiency trend prediction through a dynamic weight mechanism to generate an energy efficiency prediction result.

[0008] A4, generate a real-time control strategy through a multi-objective collaborative optimization framework according to the energy efficiency prediction result, and execute the real-time control strategy to adjust the power distribution equipment operation parameters.

[0009] A5, collect feedback data after the adjustment of the equipment parameters, and when the energy efficiency deviation of the feedback data and the energy efficiency prediction result exceeds a threshold defined by a function, use the feedback data to simultaneously optimize the adaptive hierarchical compression operation and the adaptive optimization model.

[0010] Compared with the prior art, the present application has the following advantages: (1) The present application significantly improves the integrity and sustainability of power distribution system energy efficiency optimization by constructing a complete intelligent management process from data processing to control execution to closed-loop feedback. The method organically integrates adaptive data compression, multi-dimensional feature extraction, dynamic weight prediction, multi-objective control, and self-updating mechanism into one, forming a dynamic system that can continuously learn and adapt to environmental changes, ensuring that the energy efficiency optimization strategy remains efficient and robust in long-term operation, and achieving high resilience in system management.

[0011] (2) The present application enhances the accuracy and foresight of energy efficiency management decisions. By using differentiated adaptive compression strategies for different types of data and combining time, energy efficiency, equipment state and other multi-dimensional features for deep mining, high-quality and information-rich inputs are provided for the prediction model. The dynamic weight mechanism enables the model to intelligently focus on the most influential factors, greatly improving the accuracy of energy efficiency trend prediction and providing reliable decision-making basis for developing scientific and reasonable energy-saving control strategies.

[0012] (3) The present application realizes the collaborative optimization and balanced management of the power distribution system under multiple objectives of energy saving, stability and cost. The multi-objective collaborative control framework can dynamically balance between multiple mutually restrictive operation objectives based on accurate energy efficiency prediction results, generating control strategies that not only pursue energy efficiency, but also take into account the safety and stability of the system and the economy of operation, avoiding potential risks that may be caused by single objective optimization, making the overall operation and management of the power distribution system more scientific, comprehensive and efficient. BRIEF DESCRIPTION OF DRAWINGS

[0013] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the following will briefly introduce the drawings needed for the embodiment description. Obviously, the drawings in the following description are only some embodiments of the present application, and other drawings can be obtained by those skilled in the art without creative labor.

[0014] Figure 1 The flow chart of the method is shown in the figure. DETAILED DESCRIPTION

[0015] The technical solutions in the embodiments of the present application will be clearly and completely described with reference to the accompanying drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by a person of ordinary skill in the art without creative work fall within the protection scope of the present application.

[0016] Please refer to Figure 1 As shown in the figure, the power distribution system energy efficiency dynamic optimization and energy saving control method for massive time series data proposed by the present application comprises: A1, acquiring massive time series data of the power distribution system, and performing adaptive hierarchical compression operation on the massive time series data to generate compressed data.

[0017] In a preferred embodiment, the adaptive hierarchical compression operation on the massive time series data to generate compressed data comprises: identifying mutation type data, slow change type data and correlation type data in the massive time series data.

[0018] The lossless compression method is used to process the mutation type data to generate first compressed sub-data.

[0019] The lossy compression method is used to process the slow change type data to generate second compressed sub-data.

[0020] The master-slave elimination strategy is used to process the correlation type data to generate third compressed sub-data.

[0021] The first compressed sub-data, the second compressed sub-data and the third compressed sub-data are integrated to generate compressed data.

[0022] Specifically, the adaptive hierarchical compression operation is performed on the massive time series data. The massive time series data mainly refers to a large amount of monitoring data continuously changing with time generated in the operation process of the power distribution system, such as electrical parameter data of voltage, current, active power, etc., device operation state data of device temperature, operating hours, switch action times, load rate, etc., and energy efficiency related data of power factor, line loss rate, transformer load rate. The method for generating compressed data first performs real-time analysis and classification on the acquired massive time series data stream of the power distribution system. This step identifies the data type by calculating the change rate of the data points in the continuous time window.

[0023] The mass time-series data types include mutation type data, slow change type data or correlation type data. The mutation type data refers to data that changes significantly in a short time, the change rate of which is much higher than the normal level, and usually reflects a burst event or abnormal state in the power distribution system. The slow change type data refers to data that changes slowly in a long time, the change rate of which is low, and reflects the steady-state operation characteristics or gradual change process of the power distribution system. The correlation type data refers to data that has a dependent or correlation relationship with other data, and the change of the data is often affected by other data, or the data together with other data reflects a certain characteristic of the system.

[0024] For example, a data sequence V, the change rate of which at time point t can be represented as: wherein, is the data value at the current time, is the data value at the previous time, is the time interval. When exceeds the preset high change threshold, the data sequence is identified as mutation type data; when continues to be lower than the preset low change threshold, it is identified as slow change type data; for the identification of correlation type data, the correlation coefficient between different sensor data sequences is calculated, for example, the Pearson correlation coefficient, and when the coefficient is higher than the preset correlation threshold, the data sequences are identified as correlation type data.

[0025] After classification, different compression strategies are performed for different data types: for the identified mutation type data, in order to ensure that the key information contained therein is not distorted, a lossless compression method such as LZW algorithm is used to process it to generate a first compressed sub-data. The LZW algorithm is a lossless data compression algorithm based on dictionary encoding, which replaces repeated strings by constructing a string dictionary, thereby realizing data compression.

[0026] For slow change type data, lossy compression is implemented, and the compression ratio is dynamically adjusted according to the data change rate. The lower the change rate, the higher the compression ratio, and vice versa. The compression ratio is reduced to retain more details, and a second compressed sub-data is generated after processing.

[0027] For correlation type data, a master-slave elimination strategy is implemented. First, a sensor with the highest data quality or the most critical position is selected as the master data source, and the remaining highly correlated sensors are slave data sources. The system only saves the data of the master data source completely, and the slave data sources only record the difference with the master data source or the data points when the difference exceeds a certain range. In this way, redundant information is removed, and a third compressed sub-data is generated.

[0028] Finally, the generated first, second and third compressed sub-data, together with the respective metadata tags, are integrated into a structured data stream, i.e. the final compressed data, for use in subsequent steps.

[0029] This method greatly improves the data compression efficiency and processing speed by dynamically classifying data and applying targeted compression strategies while ensuring the lossless of key information. Compared with traditional methods that use a single compression algorithm, this hierarchical compression mechanism can intelligently distinguish the importance of data, avoid the precision damage of sudden key signals, and effectively eliminate a large amount of redundancy in slowly varying and related data, thereby providing high-quality and lightweight input for subsequent energy efficiency analysis and prediction models. This fine-grained preprocessing not only reduces the storage and computing burden of the system, but also directly improves the accuracy of subsequent feature extraction and the reliability of energy efficiency prediction by retaining effective information, filtering noise and redundancy, laying a solid data foundation for dynamic optimization and energy-saving control of the entire system, and achieving a synergistic improvement in processing efficiency and analysis accuracy.

[0030] In a further preferred embodiment, the slowly varying data is processed using a lossy compression method to generate second compressed sub-data, comprising: monitoring the data change rate of the slowly varying data to generate a change rate index.

[0031] The compression ratio is dynamically calculated according to the change rate index to generate a dynamic compression ratio.

[0032] The dynamic compression ratio is calculated according to the change rate index by mapping the change rate index to the dynamic compression ratio wherein the dynamic compression ratio is determined by the following formula: wherein, and are the maximum and minimum compression ratios allowed by the system, is a set sensitivity adjustment coefficient, is the change rate index, is an exponential function with the natural constant as the base.

[0033] The slowly varying data is lossily compressed using the dynamic compression ratio to generate the second compressed sub-data.

[0034] Specifically, in the embodiment, the slowly varying data is processed by a lossy compression method, in which the compression ratio is dynamically adjusted according to the data variation rate to generate the second compressed sub-data. The execution process is as follows. First, the system continuously monitors the identified slowly varying data stream to generate a variation rate index. This step is achieved by applying a fixed length sliding time window to the data stream. In each window, the system calculates the variance of the data as a quantitative data variation rate.

[0035] The variance of the data, i.e. the variation rate index , the calculation formula is: , wherein is the number of data points in the window, is the i-th data point in the window, i is the number of data points in the window, is the average value of all data points in the window. The variation rate index objectively reflects the recent volatility of the data.

[0036] Subsequently, the system dynamically calculates the compression ratio according to the variation rate index. This calculation is completed by a pre-set mapping function, which maps the variation rate index to a specific compression ratio value to generate a dynamic compression ratio . The mapping function is designed to be inversely proportional, i.e. the smaller the variation rate index, the higher the dynamic compression ratio.

[0037] The mapping function is represented as: , wherein and are the maximum and minimum compression ratios allowed by the system, used to map the variation rate index into the adjustment range of the compression ratio; is an exponential function with base e, used to map the variation rate index into the adjustment range of the compression ratio C. The exponential function has a nonlinear characteristic, which can quickly increase or decay according to the size of the input value, making it very suitable for controlling the severity of the change in the compression ratio with the variation rate; is a sensitivity adjustment coefficient, used to control the severity of the change in the compression ratio with the variation rate. It is usually determined by the system designer through experiments or simulations according to the historical operation data of the power distribution system, equipment characteristics and energy efficiency optimization goals. For example, in a similar scale power distribution network, if the data fluctuation is sensitive to energy efficiency, the coefficient is set to a larger value of 0.7 to respond quickly to changes; if the system has a high tolerance to data fluctuations, the coefficient is set to a smaller value of 0.3.

[0038] Finally, the system applies this dynamic compression ratio to the current window of the slowly-varying data to generate the second compressed sub-data. The specific compression technique can be adaptive sampling, i.e., if the dynamic compression ratio is C, then only one representative data point is retained in every C data points; or piecewise aggregation, i.e., the data in the window is represented by one or a few parameters, such as the slope and intercept of a linear fit.

[0039] This method realizes the intelligentization and refinement of data compression. Compared with the traditional method of using a fixed compression ratio, this method can adaptively adjust the compression strength according to the real-time dynamic characteristics of the data. When the power distribution system runs smoothly and the data changes slightly, the system automatically uses a high compression ratio, greatly reducing the redundancy of data storage and transmission, and improving the processing efficiency. When the system load fluctuates or state transitions occur, the data change rate increases, and the system automatically reduces the compression ratio to retain sufficient data details, ensuring that these key change information will not be lost due to excessive compression. This dynamic balancing strategy maximizes the retention of information essential for subsequent energy efficiency analysis and prediction while ensuring data processing efficiency, providing a high-quality data source for the accuracy and reliability of the entire optimization control system, and achieving the best trade-off between data precision and storage efficiency.

[0040] A2, dynamically extracting multi-dimensional features from the compressed data to generate a dynamic feature vector.

[0041] In a preferred embodiment, the dynamically extracting multi-dimensional features from the compressed data to generate a dynamic feature vector includes extracting time series fluctuation features from the compressed data to generate a time feature sub-vector.

[0042] Extracting energy efficiency correlation features composed of power factor, load rate, and device input-output energy ratio from the compressed data to generate an energy efficiency feature sub-vector.

[0043] Extracting device state features composed of device temperature, operating hours, and quantitative indicators of switch action times from the compressed data to generate a state feature sub-vector.

[0044] By splicing and fusing the time feature sub-vector, the energy efficiency feature sub-vector, and the state feature sub-vector in the vector space, a dynamic feature vector is generated.

[0045] Specifically, the method of dynamically extracting multi-dimensional features from the compressed data to generate a dynamic feature vector in this embodiment starts with analyzing the compressed data generated in the previous step. The compressed data stream contains different types of time series information.

[0046] First, to extract the time series fluctuation features, the system uses sliding time window technology to segment the data sequence. Within each window, a series of statistical indicators such as mean, variance, kurtosis and skewness are calculated. These indicators collectively quantify the data's short-term concentration trend, dispersion degree and distribution form, thus capturing its dynamic volatility. Combining these statistical indicators forms a time feature sub-vector.

[0047] Next, the energy efficiency correlation features are extracted. This step focuses on mining the internal relationship between energy consumption data and other operating parameters. For example, by calculating the ratio of active power to total apparent power, the power factor, a key energy efficiency indicator, is obtained. At the same time, the system also analyzes parameters such as load rate and device input-output energy ratio, which reflect the efficiency of energy conversion and utilization. These calculated energy efficiency indicators are integrated to form an energy efficiency feature sub-vector.

[0048] Then, the device state features are extracted, which aims to quantify the health status and operating pressure of power distribution equipment. By analyzing the device temperature, operating hours, switch action times and other information recorded in the compressed data, and comparing them with the preset device normal operation baseline or threshold, deviation values or state scores are generated. For example, the normalized difference between the current temperature of a device and its historical average temperature can be used as a feature. These quantified state indicators collectively form a state feature sub-vector.

[0049] Finally, the generated time feature sub-vector, energy efficiency feature sub-vector and state feature sub-vector are spliced and fused in the vector space to form a unified high-dimensional vector, namely the dynamic feature vector, which comprehensively describes the multi-dimensional operating state of the power distribution system at a specific time point.

[0050] This method transforms the original, unstructured compressed data into a high-information-density, structured dynamic feature vector, providing a comprehensive and deep system state insight for subsequent energy efficiency prediction models. Compared to traditional methods that only use single time series data for analysis, this method constructs a more complete system portrait by fusing features from three dimensions: time fluctuation, energy efficiency correlation and device state. The synergistic effect of these multi-dimensional features enables the prediction model not only to perceive the surface energy consumption changes, but also to understand the physical reasons behind the changes and the device health status, thus accurately identifying the root causes of energy efficiency reduction and predicting future trends. This greatly improves the robustness and accuracy of the prediction model, avoiding misjudgments due to incomplete information, and lays a solid foundation for achieving precise energy-saving control strategies.

[0051] In a further preferred embodiment, the extracting energy efficiency related features from the compressed data to generate an energy efficiency feature sub-vector comprises: analyzing the correlation between energy consumption and device operating states in the compressed data to generate correlation parameters as regression model coefficients.

[0052] Based on the electrical parameters extracted from the compressed data, a set of energy efficiency indicators including power factor, transformer load rate, line loss rate are calculated, and the set of energy efficiency indicators are taken as a set of energy efficiency calculation values.

[0053] The correlation parameters and the set of energy efficiency calculation values are vectorized to generate an energy efficiency feature sub-vector.

[0054] Specifically, the embodiment aims to deeply mine and quantify the energy utilization efficiency of the power distribution system, and realizes the conversion from raw time series data to high information density energy efficiency features. This process first analyzes the input compressed data to identify and model the inherent correlation between energy consumption and device operating states. The system will select key energy consumption indicators such as instantaneous active power and related operating state variables such as voltage, current, load rate, and use statistical methods such as multiple regression analysis or correlation analysis to generate correlation parameters. These correlation parameters, such as the coefficients of the regression model, quantify the contribution of different operating state variables to energy consumption, thus revealing the key factors affecting energy efficiency.

[0055] Next, based on the understanding of these correlations, the system proceeds to calculate a series of specific energy efficiency indicators to generate energy efficiency calculation values. This is a process of converting raw electrical parameters into efficiency measures with clear physical meaning. For example, the system will calculate the power factor based on the active power P and the apparent power S in the compressed data, whose formula is: , where represents the actual power used to do work, represents the total apparent power, both of which can be directly calculated or extracted from the voltage and current values in the compressed data.

[0056] As a key energy efficiency calculation value, the power factor directly reflects the degree of effective utilization of electrical energy. In addition to the power factor, the system will also calculate other energy efficiency indicators such as transformer load rate, line loss rate, etc., to form a set of energy efficiency calculation values.

[0057] The last step, the system integrates and formats this set of scalar form energy efficiency calculation values, organizes them into an ordered numerical array, i.e. generates an energy efficiency feature sub-vector. In this process, each indicator may also be normalized to eliminate dimensional differences and ensure their weight fairness in subsequent models.

[0058] This method extracts and transforms the energy efficiency information hidden and dispersed in massive time series data into a set of structured and high information density energy efficiency related features. Compared with directly using raw electrical parameters as input, this method greatly enhances the interpretability and effectiveness of input data by generating energy efficiency feature sub-vectors with clear physical meaning. It enables the subsequent energy efficiency prediction model to learn and reason directly based on the "efficiency state" of the system, rather than starting from scratch to discover complex electrical principles. This not only significantly reduces the learning difficulty of the model and its dependence on training data, but more importantly, it makes the model's prediction results closely linked to the actual physical process, greatly improving the accuracy and robustness of the prediction, and providing direct and profound insights for formulating precise and effective energy-saving control strategies.

[0059] A3, input the dynamic feature vector into an adaptive optimization model to perform energy efficiency trend prediction through a dynamic weight mechanism, and generate an energy efficiency prediction result.

[0060] In a preferred embodiment, the step of inputting the dynamic feature vector into an adaptive optimization model to perform energy efficiency trend prediction through a dynamic weight mechanism, and generate an energy efficiency prediction result, comprises: analyzing the continuity of the feature sequences in the dynamic feature vector to generate a weight distribution parameter, wherein the weight distribution parameter is a set of weights obtained by normalizing the evaluation scores of the quantized features in the dynamic feature vector using a function.

[0061] According to the weight distribution parameter, the dynamic feature vector is weighted to generate a weighted feature vector.

[0062] The weighted feature vector is input into the adaptive optimization model for prediction to generate an energy efficiency prediction result.

[0063] Specifically, in this embodiment, the method of inputting the dynamic feature vector into an adaptive optimization model to perform energy efficiency trend prediction through a dynamic weight mechanism to generate an energy efficiency prediction result, the core lies in enabling the prediction model to dynamically identify and focus on the most important information. The process first analyzes the input dynamic feature vector in depth to evaluate the continuity and importance of each feature sequence in a particular time context. The continuity here refers to the stability and contribution of a feature in predicting future energy efficiency. The system uses an attention module to quantify this continuity, calculating a real-time relevance score for each feature in the dynamic feature vector. The relevance score is dynamically generated based on the value of the feature itself and its interaction with the internal state of the model.

[0064] Then, the system normalizes the relevance scores of all features using a Softmax function to generate the weight assignment parameters. The Softmax function is an activation function widely used in machine learning and deep learning, especially as the activation function of the output layer in multi-class classification problems. It can map a vector to a probability distribution, where each element's value is between 0 and 1, and the sum of all elements is 1.

[0065] The formula for calculating the weights is as follows: , where is the weight of the rth feature in the dynamic feature vector, and r is the number of each feature in the dynamic feature vector; is the relevance score of the rth feature in the dynamic feature vector; is the relevance score of the jth feature in the dynamic feature vector, and when using the Softmax function, j ranges over all dimensions of the dynamic feature vector, i.e., all features are traversed; is the sum of the relevance scores of all features, ensuring the normalization of the output weights. By dividing the relevance score of each feature by the sum of the relevance scores of all features, the Softmax function ensures that the output weights have a value range of 0 to 1, and the sum of all weights is 1. This set of constitutes the weight assignment parameters.

[0066] Next, the system applies these weight assignment parameters to the original dynamic feature vector. Specifically, each feature component in the dynamic feature vector is multiplied element-wise with its corresponding weight to generate a weighted feature vector. In this new vector, feature components with high relevance scores are amplified, while feature components with low relevance scores are suppressed.

[0067] Finally, this weighted feature vector is input into an adaptive optimization model. This adaptive optimization model is a machine learning model capable of time series prediction, such as a long short-term memory network, whose internal parameters can be continuously adjusted based on new feedback data. After receiving the weighted feature vector, the model performs forward propagation calculations and finally outputs the prediction of the energy efficiency value at one or more future time steps. This output is the energy efficiency prediction result. For example, the active power and reactive power prediction values at key nodes such as feeder head, main branch point, and large user in the power distribution network from 15 minutes to 24 hours in the future, and these predictions may also include voltage prediction values at key nodes.

[0068] This method introduces a dynamic weight mechanism, giving the energy efficiency prediction model "attention", enabling it to intelligently judge and amplify the most influential features according to the context of the real-time operation of the power distribution system, while ignoring or weakening the interference of noise and irrelevant features. Compared with static prediction models, this method can more accurately capture key changes in system status, such as automatically increasing attention to current and voltage fluctuations during load surges, and focusing more on temperature, loss, and other status features during equipment aging. This adaptive focusing ability greatly improves the accuracy and robustness of energy efficiency prediction results, making them more accurately reflect future energy efficiency trends. This not only provides high-quality decision-making basis for subsequent control decisions, but also enhances the intelligence level and adaptability of the entire optimization system to complex working conditions.

[0069] In a further preferred embodiment, the analysis of the continuity of the feature sequence in the dynamic feature vector generates a weight distribution parameter, including: retrieving a series of historical energy efficiency target values as historical operation results from the database to generate a historical energy efficiency target value sequence.

[0070] The continuity of the time series fluctuation feature in the dynamic feature vector is evaluated by calculating the moving average of the mutual information or Pearson correlation coefficient between it and the historical energy efficiency target value sequence to generate a time weight factor.

[0071] The continuity of the energy efficiency correlation feature in the dynamic feature vector is evaluated by calculating the moving average of the mutual information or Pearson correlation coefficient between it and the historical energy efficiency target value sequence to generate an energy efficiency weight factor.

[0072] The continuity of the device status feature in the dynamic feature vector is evaluated by calculating the moving average of the mutual information or Pearson correlation coefficient between it and the historical energy efficiency target value sequence to generate a status weight factor.

[0073] The time weight factor, energy efficiency weight factor, and status weight factor are combined to guide the calculation of the function to generate the weight distribution parameter. The calculation of the function generates the weight distribution parameter.

[0074] Specifically, the method of analyzing the continuity of the feature sequence in the dynamic feature vector in this embodiment to generate the weight distribution parameter is based on evaluating the contribution of different feature categories to future energy efficiency prediction. This process first decomposes the input dynamic feature vector into its three basic components: time series fluctuation features, energy efficiency correlation features, and device status features.

[0075] For each feature category, the system performs an independent coherence evaluation. Taking the coherence evaluation of time series fluctuation features as an example, the system examines the correlation strength and stability between the feature sub-vector and the actual energy efficiency change in a past time window. To achieve this goal, the system obtains a quantitative evaluation score by calculating the moving average of the mutual information or Pearson correlation coefficient between the feature sub-vector and the historical energy efficiency target value sequence. The higher the score, the stronger the coherence of the feature category, and the higher the value of the feature category as a predictor. This evaluation score is the generated time weight factor.

[0076] Similarly, the system independently evaluates the coherence of energy efficiency related features and device status features with historical energy efficiency changes using the same method, respectively generating energy efficiency weight factors and status weight factors. The time weight factor, energy efficiency weight factor, and status weight factor represent the relative importance of the three feature categories in the current system operating context.

[0077] Finally, the system combines the three generated weight factors into a vector, which is the weight allocation parameter. This weight allocation parameter will be passed to the next step as the macro guidance basis for fine-grained weight allocation by the dynamic weight mechanism.

[0078] This method establishes a structured and physically meaningful prior basis for subsequent dynamic weight allocation. Compared with the method of directly learning weights on all features without discrimination, this method achieves hierarchical attention allocation from macro to micro by pre-evaluating the overall importance of different feature categories. This method enables the energy efficiency prediction model to no longer blindly explore each independent feature when adjusting weights, but can first identify whether time fluctuations, energy efficiency indicators, or device health status are dominating energy efficiency changes in the current system state. For example, when there are signs of device failure, the coherence of device status features will significantly increase, and the system will accordingly increase its weight factor, guiding the model to pay more attention to device status information. This grouping evaluation and combination based on feature category coherence greatly improves the intelligence level and convergence speed of the dynamic weight mechanism, enabling it to focus more quickly and accurately on the core of the problem, thereby improving the adaptability and interpretability of the entire energy efficiency prediction model.

[0079] A4、According to the energy efficiency prediction result, a real-time control strategy is generated through a multi-objective collaborative optimization framework, and the real-time control strategy is executed to adjust the operation parameters of the power distribution equipment.

[0080] In a preferred embodiment, the generation of a real-time control strategy through a multi-objective collaborative optimization framework based on the energy efficiency prediction result includes determining energy saving target constraints based on the energy efficiency prediction result, and generating energy saving optimization parameters including the expected adjustment of reactive power compensation or transformer load rate.

[0081] Based on the energy efficiency prediction results, determine the stability target constraints, generate the stability optimization parameters including the upper and lower limits of voltage and the thermal limits of equipment.

[0082] Obtain the operation cost model, and determine the cost target constraints in combination with the energy efficiency prediction results, generate the cost optimization parameters including the limit of the number of switchgear operations.

[0083] Fuse the energy saving optimization parameters, stability optimization parameters and cost optimization parameters, and generate the real-time control strategy through a multi-objective optimization algorithm.

[0084] Specifically, in the embodiment, the method of generating a real-time control strategy according to energy efficiency prediction results through a multi-objective collaborative optimization framework is a process of converting abstract prediction data into specific executable equipment operation instructions. This process starts with taking the energy efficiency prediction results generated in the previous step as input, and quantizing them into different dimension optimization target functions or constraint conditions.

[0085] Firstly, based on the energy efficiency prediction results, the system evaluates the expected energy loss level and compares it with the ideal energy efficiency benchmark, thereby determining the energy saving target constraints, which aim to minimize the predicted energy loss, thereby generating a set of energy saving optimization parameters, such as the expected amount of reactive power compensation or transformer load rate.

[0086] The core of this step is to evaluate the expected energy loss level. The evaluation process starts with receiving the energy efficiency prediction results from the previous step, which can specifically include the predicted sequences of active power and reactive power of each key load node in the power distribution network for one or more future time steps. To convert the predicted load data into specific energy loss, the system first calls the pre-stored digital twin model or topology parameter database of the power distribution network, which accurately describes the network topology and contains the resistance and reactance parameters of each line segment, as well as the rated loss and impedance parameters of each transformer. Subsequently, for each time step in the prediction time domain, the system performs a power flow calculation once, which is a very mature and well-known technology in the field of power system analysis. By taking the predicted load value of this time step as input, the power flow calculation solves the current of each line in the network and the load current of each transformer. Based on the results of the power flow calculation, the system calculates the expected energy loss of this time step according to the following physical formula: 1) line active loss calculation: for all line segments in the network, their losses are accumulated as , 1) Resistance of the line segment; 2) Transformer loss calculation: including constant iron loss and copper loss proportional to the square of the load. Iron loss, also known as no-load loss, is a constant loss that a transformer will generate as long as it is connected to the power grid, and is independent of the load size. This value is stored in the database as an inherent parameter of the transformer; Copper loss, also known as load loss, is the load current. The losses generated when current flows through the transformer windings are proportional to the square of the load current. The total loss is the sum of all line losses and all transformer losses, yielding the instantaneous total power loss at that time step. Finally, the system integrates the instantaneous total power loss over all time steps in the predicted time domain to obtain the total expected energy loss level. , It is the energy lost during the t-th time step. The number representing the time step. This represents the duration of each time step. This quantification provides a direct and accurate numerical basis for subsequently determining energy-saving target constraints. The system compares this assessed energy loss level with an ideal energy efficiency benchmark to determine energy-saving target constraints designed to minimize predicted energy losses, thereby generating a set of energy-saving optimization parameters, such as the desired adjustment of reactive power compensation or transformer load rate.

[0087] Furthermore, the system analyzes the potential impact of energy efficiency forecasts on grid stability, such as whether predicted load changes might lead to voltage overshoots or line overloads. Based on this, the system sets stability target constraints, such as the requirement to control voltage fluctuations within standard ranges, and generates corresponding stability optimization parameters, including upper and lower voltage limits and equipment thermal limits.

[0088] The specific process of this step is as follows: For each time step in the prediction time domain, the system extracts the predicted voltage amplitude of all key nodes from the power flow calculation results. And the predicted current amplitude of all lines / transformers The system compares these predicted values ​​with preset stability optimization parameters in real time. These stability optimization parameters are a set of hard boundary conditions stored in the system configuration library, including: 1) the maximum allowable voltage limit for each node. and minimum voltage limit This limit follows national or industry standards, such as ±5% of the nominal voltage; 2) The long-term allowable current carrying capacity, i.e., the thermal stability limit, calculated for each line and transformer based on its model, material, and ambient temperature. Furthermore, the system transforms these alignment results into a quantifiable stability penalty function. The function is designed to: when all nodes... All in Within the interval, and all devices are all less than their , , once any of the predicted values exceeds its corresponding safety boundary, , the value of

[0089] Further, the system combines the energy efficiency prediction results with an operation cost model, which contains information such as electricity price periods, equipment maintenance costs, etc., to evaluate the economy of different control schemes, thereby determining the cost target constraint and generating cost optimization parameters, such as limiting the number of operations of switching devices to extend their service life.

[0090] The operation cost model is a quantitative evaluation system composed of multiple sub-models. Specifically, the calculation of the operation cost function includes the following steps: 1) calculation of electricity cost: the system obtains the total active power consumption of the entire distribution network including load consumption and network loss at each future time step from the power flow calculation. At the same time, the system calls the built-in time-of-use electricity price database, which stores the electricity prices of peak, flat, and valley periods . By multiplying and accumulating the two by time step, the total electricity cost is calculated. 2) calculation of device operation and maintenance cost: the system analyzes each control scheme to be optimized, such as a set of device operation instructions, and counts the number of operations of various types of devices involved, such as the number of transformer tap adjustments and the number of capacitor switching . The cost optimization parameter is the equivalent depreciation or maintenance cost of each type of operation preset by the system . The system calculates the total operation and maintenance cost by multiplying the number of operations by the unit cost. Finally, the total operation cost function under the control scheme is obtained by adding the two parts of the cost . This function serves as a target to be minimized in the multi-objective optimization framework, guiding the system to find the most economically beneficial control strategy under the premise of meeting energy saving and stability.

[0091] After generating the energy saving optimization parameters, stability optimization parameters, and cost optimization parameters, the system integrates these three into a unified multi-objective collaborative optimization framework.

[0092] The multi-objective collaborative optimization framework balances through a comprehensive objective function, which can be represented as: , where represents the comprehensive objective function; This represents the energy loss function defined by energy-saving optimization parameters; This represents the stability penalty function defined by the stability optimization parameters; its value increases sharply when the system state deviates from the stability region. This represents the operating cost function defined by the cost optimization parameters. These are preset weighting coefficients, representing the importance of the three objectives of energy saving, stability, and cost. These coefficients can be adjusted according to the overall operating strategy of the system.

[0093] Finally, the system uses an optimization algorithm to find the minimum value of the objective function and finds a set of power distribution equipment operating parameters that minimize the comprehensive objective function. This set of parameters constitutes the final real-time control strategy and is sent to the relevant equipment for execution.

[0094] This method constructs a multi-objective collaborative optimization framework that integrates energy saving, stability, and cost. The real-time control strategy generated by this method is no longer a short-sighted decision based on a single objective, but rather a comprehensive optimal solution obtained from a global perspective. Compared to traditional control methods that only focus on energy saving, this scheme effectively avoids the potential risks of sacrificing system stability or increasing long-term operation and maintenance costs due to excessive pursuit of energy saving. For example, it makes a wise trade-off between an operation with significant energy saving effect but which may cause voltage fluctuations and an operation with slightly less energy saving effect but stable system operation. This collaborative optimization under multi-dimensional constraints ensures that the power distribution system maintains high operational reliability and economy while achieving high energy efficiency, significantly improving the overall resilience and intelligence level of the system, making the control strategy more comprehensive, safe, and forward-looking.

[0095] A5. Collect feedback data after the equipment parameters are adjusted. When the energy efficiency deviation between the feedback data and the energy efficiency prediction results exceeds the threshold defined by the function, use the feedback data to synchronously optimize the adaptive hierarchical compression operation and the adaptive optimization model.

[0096] In a preferred embodiment, the step of synchronously optimizing the adaptive hierarchical compression operation and the adaptive optimization model using feedback data includes: comparing the feedback data with the energy efficiency prediction results to generate an energy efficiency deviation index.

[0097] When the energy efficiency deviation index exceeds the threshold defined by the function, an update cycle is triggered.

[0098] The compression parameters of the adaptive stratified compression operation are adjusted based on the energy efficiency deviation index to generate an optimized compression operation.

[0099] The prediction parameters of the adaptive optimization model are adjusted based on the energy efficiency deviation index to generate an optimized prediction model.

[0100] Specifically, the method of this embodiment utilizes feedback data to synchronize the adaptive hierarchical compression operation and the adaptive optimization model, thereby constructing a closed-loop adaptive learning system. This process is initiated after the real-time control strategy is executed, and the system continuously collects the actual operating data after adjusting the device parameters as feedback data.

[0101] First, the system compares the actual energy efficiency values extracted from the feedback data with the energy efficiency prediction results generated before the control is executed point by point or time period by time period, in order to quantify the prediction accuracy of the model. Through this comparison, the system calculates the energy efficiency deviation index, which can be a normalized error value such as the root mean square error or the mean absolute percentage error. A simplified energy efficiency deviation index may be represented as: wherein, is the actual energy efficiency value obtained from the feedback data, is the corresponding energy efficiency prediction result.

[0102] When the calculated energy efficiency deviation index continuously or single-time significantly exceeds the threshold value preset by the system, it indicates that the performance of the current model has failed to meet the accuracy requirements, and the system automatically triggers the update cycle. In this cycle, the system first adjusts the compression parameters in the adaptive hierarchical compression operation according to the size and direction of the energy efficiency deviation index. For example, if the deviation is too large, it may mean that too much key information is lost in the lossy compression link, and the system will accordingly reduce the compression ratio of the slowly varying data or increase the sensitivity of the associated data elimination strategy to generate an optimized compression operation that retains more details.

[0103] At the same time, the system uses this feedback data and the corresponding input feature vector to form a new training sample pair. This sample pair is used to update the adaptive optimization model.

[0104] Specifically, the energy efficiency deviation index is used as part of the loss function, and the prediction parameters inside the model, such as the weights and biases of the neural network, are adjusted through the backpropagation algorithm to minimize the deviation of future predictions, thereby generating an optimized prediction model with better performance. These two optimization processes are carried out simultaneously to ensure the consistency and synergy of the data processing front end and the model prediction back end.

[0105] This method changes the entire energy efficiency optimization system from a static, one-time deployment process into a dynamic intelligent system that can continuously learn and evolve. By monitoring the deviation between prediction and reality in real time, this mechanism can timely discover and correct the model performance degradation problem caused by device aging, load mode changes or environmental factor changes, i.e. model drift. This closed-loop self-correction capability not only guarantees the long-term accuracy and reliability of energy efficiency prediction, but also ensures the effectiveness of the entire data link through reverse optimization of data compression strategies. Ultimately, it gives the system excellent environmental adaptability and robustness, enabling it to maintain efficient energy-saving control effects in a complex and variable power distribution system environment, achieving sustainable and resilient energy efficiency management.

[0106] It should be noted that the above formulas can convert physical quantities of different properties into unitless standard values or same-dimension superimposable parameters through dimensional consistency principles and mathematical standardization methods (such as normalization processing, dimensionless parameter conversion, or unit system unification), thereby eliminating the interference of different dimensions on the operation logic, making the formulas have mathematical operation rationality and objective law adaptability while preserving the original data distribution characteristics. The above is only an exemplary embodiment of the present application, and cannot limit the scope of the present application.

Claims

1. A power distribution system energy efficiency dynamic optimization and energy saving control method for mass time series data, characterized in that, The method comprises the following steps: A1, acquiring mass time series data of a power distribution system, and performing adaptive hierarchical compression operation on the mass time series data to generate compressed data; The adaptive hierarchical compression operation on the mass time series data to generate compressed data comprises: identifying mutation type data, slow change type data and correlation type data in the mass time series data; processing the mutation type data by using a lossless compression method to generate first compressed sub-data; processing the slow change type data by using a lossy compression method to generate second compressed sub-data; processing the correlation type data by using a master-slave elimination strategy to generate third compressed sub-data; integrating the first compressed sub-data, the second compressed sub-data and the third compressed sub-data to generate the compressed data; The processing of the slow change type data by using a lossy compression method to generate second compressed sub-data comprises: monitoring the data change rate of the slow change type data to generate a change rate index; dynamically calculating a compression ratio according to the change rate index to generate a dynamic compression ratio; The dynamic calculation of the compression ratio according to the change rate index is mapping the change rate index to a dynamic compression ratio C, wherein the dynamic compression ratio C is determined by the following formula: wherein, and Cmax and Cmin are the maximum and minimum compression ratios allowed by the system, is a set sensitivity adjustment coefficient, is a change rate index, is an exponential function with the natural constant e as the base; applying the dynamic compression ratio to lossy compression of the slow change type data to generate the second compressed sub-data; A2, performing multi-dimensional feature dynamic extraction on the compressed data to generate a dynamic feature vector; A3, inputting the dynamic feature vector into an adaptive optimization model to perform energy efficiency trend prediction through a dynamic weight mechanism to generate an energy efficiency prediction result; The inputting of the dynamic feature vector into the adaptive optimization model to perform energy efficiency trend prediction through a dynamic weight mechanism to generate an energy efficiency prediction result comprises: analyzing the continuity of feature sequences in the dynamic feature vector to generate a weight distribution parameter, wherein the weight distribution parameter is a set of weights obtained by applying a Softmax function to normalized processing of evaluation scores of each feature in the dynamic feature vector; performing weighted processing on the dynamic feature vector according to the weight distribution parameter to generate a weighted feature vector; inputting the weighted feature vector into the adaptive optimization model for prediction to generate the energy efficiency prediction result; The analysis of the continuity of feature sequences in the dynamic feature vector to generate a weight distribution parameter comprises: retrieving a series of historical energy efficiency target values as historical running results from a database to form a historical energy efficiency target value sequence; evaluating the continuity of time sequence fluctuation features in the dynamic feature vector to generate a time weight factor by calculating the moving average of mutual information or Pearson correlation coefficient between the time sequence fluctuation features and the historical energy efficiency target value sequence; evaluating the continuity of energy efficiency correlation features in the dynamic feature vector to generate an energy efficiency weight factor by calculating the moving average of mutual information or Pearson correlation coefficient between the energy efficiency correlation features and the historical energy efficiency target value sequence; evaluating the continuity of device state features in the dynamic feature vector to generate a state weight factor by calculating the moving average of mutual information or Pearson correlation coefficient between the device state features and the historical energy efficiency target value sequence; combining the time weight factor, the energy efficiency weight factor and the state weight factor to guide the calculation of the Softmax function to generate the weight distribution parameter; A4, generating a real-time control strategy through a multi-objective collaborative optimization framework according to the energy efficiency prediction result, and executing the real-time control strategy to adjust the running parameters of the power distribution equipment. A5, the feedback data after the adjustment of the device parameter is collected, when the energy efficiency deviation between the feedback data and the energy efficiency prediction result exceeds the threshold defined by the function, the feedback data is used to synchronously optimize the adaptive hierarchical compression operation and the adaptive optimization model.

2. The power system energy efficiency dynamic optimization and energy saving control method for mass time series data according to claim 1, characterized in that, The multi-dimensional feature dynamic extraction on the compressed data generates a dynamic feature vector, which includes: The time sequence fluctuation feature is extracted from the compressed data to generate a time feature sub-vector; The energy efficiency correlation feature composed of the power factor, the load rate and the device input-output energy ratio is extracted from the compressed data to generate an energy efficiency feature sub-vector; The device state feature composed of the quantitative indexes of the device temperature, the operating hours and the switch action times is extracted from the compressed data to generate a state feature sub-vector; The dynamic feature vector is generated by splicing and fusing the time feature sub-vector, the energy efficiency feature sub-vector and the state feature sub-vector in the vector space.

3. The power distribution system energy efficiency dynamic optimization and energy saving control method for mass time series data according to claim 2, characterized in that, The energy efficiency correlation feature is extracted from the compressed data to generate the energy efficiency feature sub-vector, which includes: The correlation parameters serving as the regression model coefficients are generated by analyzing the correlation between the energy consumption and the device operating state in the compressed data; A group of energy efficiency indexes including the power factor, the transformer load rate and the line loss rate are calculated based on the electrical parameters extracted from the compressed data, and the group of energy efficiency indexes are taken as a group of energy efficiency calculation values; The correlation parameters and the group of energy efficiency calculation values are vectorized to generate the energy efficiency feature sub-vector.

4. The power system energy efficiency dynamic optimization and energy saving control method for mass time series data according to claim 1, characterized in that, The real-time control strategy is generated by the multi-objective collaborative optimization framework according to the energy efficiency prediction result, which includes: The energy-saving target constraint is determined based on the energy efficiency prediction result to generate the energy-saving optimization parameter including the expected adjustment of the reactive power compensation amount or the transformer load rate; The stability target constraint is determined based on the energy efficiency prediction result to generate the stability optimization parameter including the upper and lower limits of the voltage and the device thermal limit; The operation cost model is obtained, and the cost target constraint is determined based on the energy efficiency prediction result to generate the cost optimization parameter including the limitation on the switch device operation times; The real-time control strategy is generated by the multi-objective optimization algorithm by fusing the energy-saving optimization parameter, the stability optimization parameter and the cost optimization parameter.

5. The power system energy efficiency dynamic optimization and energy saving control method for mass time series data according to claim 1, characterized in that, The feedback data is used to synchronously optimize the adaptive hierarchical compression operation and the adaptive optimization model, which includes: The energy efficiency deviation index is generated by comparing the feedback data with the energy efficiency prediction result; When the energy efficiency deviation index exceeds the threshold defined by the function, the update cycle is triggered; The compression parameter of the adaptive hierarchical compression operation is adjusted according to the energy efficiency deviation index to generate the optimized compression operation; The prediction parameter of the adaptive optimization model is adjusted according to the energy efficiency deviation index to generate the optimized prediction model.

Images