Multi-period multi-weight power quality real-time optimization decision-making method, equipment and medium

Abstract

The invention discloses a multi-period multi-weight electric energy quality real-time optimization decision-making method, device and medium, through dynamically generating a multi-period weight matrix and carrying out optimization calculation, voltage deviation, harmonic pollution and three-phase imbalance of a transformer area can be effectively treated, and based on a double-label classification system of power utilization characteristics and topological positions, the real-time optimization decision-making method of the multi-period multi-weight electric energy quality can be realized. According to the method, the first classification label of the power utilization characteristics of the transformer area and the second classification label of the topological position are obtained, multi-period and multi-target refined modeling of the running state of the transformer area is achieved, the strategy of combining multi-period optimization and rolling optimization is adopted, and the bus voltage is used as a decision variable to conduct multi-period optimization. And the optimal bus voltage curve and the reactive plan in the future time period are output, so that the network loss is reduced, the voltage qualification rate is improved, a rolling optimization algorithm is adopted, the gear shifting action can be dynamically adjusted according to real-time data, the change of the running state of the power distribution network can be responded in time, and the timeliness and the effectiveness of a treatment strategy are ensured.

Description

Technical Field

[0001] This invention relates to the field of transformer area voltage data processing technology, specifically to a multi-time period, multi-weight power quality real-time optimization decision-making method, device, and medium. Background Technology

[0002] In the current power distribution network, the penetration rate of renewable energy is constantly increasing, resulting in large voltage fluctuations for users, significant differences between peak and valley voltages, and frequent daytime voltage fluctuations. Existing analytical methods cannot accurately analyze power quality operation and maintenance plans. Addressing only problematic users and transformer areas through proactive maintenance can easily impact the power quality of other users, leading to multiple adjustments or upgrades for the same issue. Power quality issues require a holistic consideration of the distribution network. Analyzing system voltage issues solely based on power supply voltage regulations ignores the electricity consumption characteristics and habits of some users, failing to accurately grasp their electricity needs. Therefore, real-time optimization of power quality is necessary. However, existing optimization processes still rely on single time periods or fixed weights. When working on a single time period, the weight is usually adjusted once based on the current cross-section. However, daily electricity consumption has peak and off-peak periods, and the weight of a single time period cannot represent the weight of other time periods. If the adjustment is based on a fixed weight, the fixed weight simply weights and sums voltage, harmonics, and imbalances. The weight coefficient usually relies on human experience. The manually set weight coefficient will not be able to be dynamically correlated with the electricity consumption characteristics of the transformer area, topological location, and sensitive users. This will lead to a mismatch in the allocation of governance resources, an inability to adapt to the drastic voltage fluctuations in multiple time periods during the day caused by the superposition of photovoltaic output and load peaks, and an inability to carry out differentiated governance for different peak periods. Summary of the Invention

[0003] The technical problem this invention aims to solve is that the optimization process remains stuck in a single time period or with fixed weights. The goal is to provide a multi-time period, multi-weight real-time power quality optimization decision-making method, device, and medium. By dynamically generating a multi-time period weight matrix and performing optimization calculations, it can effectively manage voltage deviation, harmonic pollution, and three-phase imbalance in distribution areas. Based on a dual-label classification system of power consumption characteristics and topological location (i.e., power consumption characteristics (first category label) and topological location (second category label), it achieves multi-time period, multi-objective refined modeling of the distribution area's operating status. Employing a strategy combining multi-time period optimization and rolling optimization, with bus voltage as the decision variable, it performs multi-time period optimization and outputs the optimal bus voltage curve and reactive power plan for future time periods. This helps reduce network losses and improve voltage qualification rate. The rolling optimization algorithm can dynamically adjust the shifting action based on real-time data, responding promptly to changes in the distribution network's operating status and ensuring the timeliness and effectiveness of the management strategy.

[0004] This invention is achieved through the following technical solution:

[0005] The first aspect of this invention provides a multi-time-period, multi-weighted real-time power quality optimization decision-making method, comprising the following specific steps:

[0006] Acquire transformer area data and preprocess the transformer area data;

[0007] Based on the preprocessed data, the first classification label characterizing the power consumption characteristics of the transformer area is obtained;

[0008] Based on the connection relationship between the main station transformer, the line, and the distribution area, the distribution network topology is reclassified to obtain a second classification label that represents the location of the line topology.

[0009] Based on the first and second category labels, a multi-time period weight matrix is ​​dynamically generated for each transformer area at different time periods, using the voltage-harmonic-imbalance weight coefficients.

[0010] Using the voltage-harmonic-unbalance weighting coefficient matrix of each transformer area in different time periods within a preset time period as input, and the bus voltage as the decision variable, multi-time period optimization calculation is performed to output the optimal bus voltage curve and reactive power plan for future time periods.

[0011] Using the voltage-harmonic-imbalance weighting coefficient matrix of each transformer area at different time periods and the optimal bus voltage curve as inputs, and the switching action of each transformer area as the decision variable, a rolling optimization algorithm is used to calculate and generate the real-time management strategy for each transformer area.

[0012] Furthermore, the acquisition of transformer area data and the preprocessing of the transformer area data specifically include:

[0013] Synchronously collect voltage and current time-series data from power users and substation bus voltage time-series data, and perform time alignment, outlier correction, missing value imputation, resampling and normalization preprocessing on the data to obtain a synchronous time-series dataset.

[0014] Furthermore, based on the preprocessed data, a first classification label characterizing the electricity consumption characteristics of the distribution area is obtained, specifically including:

[0015] Based on the preprocessed synchronous time series dataset, frequency deviation, voltage deviation, three-phase voltage imbalance, voltage fluctuation and flicker, voltage sag and total harmonic distortion rate are calculated to form an initial value sequence of indicators.

[0016] A sliding time window is used to extract statistical features from the initial value sequence of the index, and a multidimensional statistical feature vector is constructed.

[0017] Wavelet packet transform is used to decompose the bus voltage and user current into multiple scales, extract the energy proportion of each frequency band, and form a wavelet energy feature vector.

[0018] By concatenating the wavelet energy feature vector with the statistical feature vector, a comprehensive feature vector integrating time-domain statistical characteristics and frequency-domain energy distribution is obtained.

[0019] The comprehensive feature vector is mapped to a low-dimensional manifold space through nonlinear dimensionality reduction, and HDBSCAN density clustering is used to obtain the first classification label characterizing the power consumption characteristics of the transformer area. At the same time, a label dictionary for the first classification label is established.

[0020] Furthermore, after obtaining the first category label, it also includes:

[0021] The system updates the acquired transformer area data in real time and generates new multidimensional statistical feature vectors. It compares the distribution of the new multidimensional statistical feature vectors with the distribution of historical multidimensional statistical feature vectors to quantify whether the new clustering results have drifted with the old clustering results. If any indicator exceeds the set threshold, it is determined that the electricity consumption characteristic tag has drifted, triggering a hot update of the tag dictionary.

[0022] Furthermore, the reclassification of the distribution network topology based on the connection relationship between the main station transformer, the line, and the distribution area to obtain a second classification label representing the location of the line topology specifically includes:

[0023] A directed topology tree is constructed based on the transformer-line-transformer area connection relationship provided by the main station. The topology depth and power supply radius of each transformer area to the main transformer are calculated, and a two-dimensional topology vector is generated.

[0024] K-means clustering is performed on the two-dimensional topology vector to obtain a second classification label representing the topological location of the line.

[0025] Furthermore, based on the first and second classification labels, a multi-time period weight matrix is ​​dynamically generated for the voltage-harmonic-imbalance weight coefficients of each transformer area at different time periods.

[0026] Based on the joint indexing of the first and second classification labels, the basic weight vectors of the power consumption characteristic cluster and the topological location cluster at a set threshold are read from the offline calibration library.

[0027] Obtain the user-preset time period template and correction coefficient, and perform time period correction on the basic weight vector;

[0028] The feedback gain factor is obtained by non-linearly mapping the sensitive user feedback score using the Sigmoid gain function.

[0029] The final weight vector for the current time period is obtained by multiplying the time-period-corrected weight vector by the feedback gain factor.

[0030] Iterate through all time periods within the cycle and stack them in chronological order to form a multi-time period weight coefficient matrix.

[0031] Furthermore, the output of the optimal bus voltage curve and reactive power plan for future time periods specifically includes:

[0032] The main station collects the voltage-harmonic-imbalance weight coefficient matrix uploaded by each transformer area and constructs a global weight dataset according to the transformer area number and time stamp.

[0033] Using the voltage amplitude of the low-voltage busbar on the main transformer and the output of the reactive power compensation equipment in each future time period as decision variables, a multi-time period optimization model is established. The objective function is to maximize the weighted voltage qualification rate and minimize the number of low-voltage users. The constraints include: upper and lower limits of busbar voltage safety, limit of the number of tap changers of the main transformer, capacity and ramping limit of reactive power equipment, and power flow equation of the distribution network.

[0034] The multi-period optimization model is solved to obtain the optimal bus voltage curve and reactive power plan for each future period.

[0035] Furthermore, the generation of real-time governance strategies for each distribution area specifically includes:

[0036] In each rolling time slot, the master station loads the weight coefficient matrix for the next M time periods and the corresponding optimal bus voltage curve into the rolling window;

[0037] Using the number of on-load tap changer taps in each transformer area within the window as discrete decision variables, a short-term rolling optimization model is established. The objective function is set as minimizing the weighted sum of the transformer area voltage over-limit risk and the tap change operation cost, where the voltage over-limit risk is dynamically weighted by the voltage weight coefficient of each transformer area. The constraints include: tap upper and lower limits, remaining amount of maximum daily adjustment times, tap change limit between adjacent time periods, and voltage reference trajectory deviation tolerance derived from the optimal bus voltage curve.

[0038] Solve the objective function and output the optimal dispatching command for each station area in the current time period.

[0039] A second aspect of the present invention provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement a multi-time-period, multi-weighted real-time power quality optimization decision-making method.

[0040] A third aspect of the present invention provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements a multi-time-period, multi-weighted real-time power quality optimization decision-making method.

[0041] Compared with the prior art, the present invention has the following advantages and beneficial effects:

[0042] By dynamically generating and optimizing multi-time period weight matrices, voltage deviation, harmonic pollution, and three-phase imbalance in distribution areas can be effectively addressed. Based on a dual-label classification system of power consumption characteristics and topological location (i.e., power consumption characteristics (first category label) and topological location (second category label), a refined multi-time period and multi-objective modeling of distribution area operation status is achieved. A strategy combining multi-time period optimization and rolling optimization is adopted, with bus voltage as the decision variable for multi-time period optimization, outputting the optimal bus voltage curve and reactive power plan for future periods. This helps reduce network losses and improve voltage qualification rate. The rolling optimization algorithm can dynamically adjust the shifting action based on real-time data, responding promptly to changes in the distribution network operation status and ensuring the timeliness and effectiveness of the governance strategy. Attached Figure Description

[0043] To more clearly illustrate the technical solutions of the exemplary embodiments of the present invention, the accompanying drawings used in the embodiments will be briefly described below. It should be understood that the following drawings only show some embodiments of the present invention and should not be considered as a limitation of the scope. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort. In the drawings:

[0044] Figure 1 This is a real-time power quality optimization decision-making method in an embodiment of the present invention. Detailed Implementation

[0045] To make the objectives, technical solutions, and advantages of the present invention clearer, the present invention will be further described in detail below with reference to the embodiments and accompanying drawings. The illustrative embodiments and descriptions of the present invention are only used to explain the present invention and are not intended to limit the present invention.

[0046] As one possible implementation method, such as Figure 1 As shown, since the traditional zone-time-target governance strategy relies on fixed weights and offline plans, it is difficult to adapt to rapid fluctuations in source and load. Therefore, this embodiment provides a multi-time-period, multi-weight real-time power quality optimization decision-making method, which can accurately identify problems and dynamically allocate resources to achieve comprehensive management of distribution network voltage, harmonics, and three-phase imbalance, thereby improving the power quality, operating efficiency, and intelligence level of the distribution network. The specific steps include the following:

[0047] Data of distribution areas is acquired and preprocessed. Based on the preprocessed data, a first classification label representing the power consumption characteristics of the distribution area is obtained. Based on the connection relationship between the main station transformer, line and distribution area, the distribution network topology is reclassified to obtain a second classification label representing the location of the line topology. According to the first and second classification labels, a multi-time period weight matrix of voltage-harmonic-unbalance weight coefficients of each distribution area at different time periods is dynamically generated.

[0048] Using the voltage-harmonic-imbalance weighting coefficient matrix of each transformer area at different time periods within a preset time period as input, and the bus voltage as the decision variable, multi-time period optimization calculations are performed to output the optimal bus voltage curve and reactive power plan for future time periods. Using the voltage-harmonic-imbalance weighting coefficient matrix of each transformer area at different time periods and the optimal bus voltage curve as input, and the switching action of each transformer area as the decision variable, a rolling optimization algorithm is used to calculate and generate the real-time management strategy for each transformer area.

[0049] This embodiment proposes a dual-label system of electricity consumption characteristics and topological location to achieve differentiated modeling of distribution transformer areas; it constructs a dynamic weight matrix generation mechanism to quantify the priority of multiple objectives; and it designs a two-level rolling optimization model of main and distribution transformers, taking into account both global optimization and local executability. Dynamically generating and optimizing multi-time period weight matrices effectively addresses voltage deviation, harmonic pollution, and three-phase imbalance in distribution transformer areas. Based on the dual-label classification system of electricity consumption characteristics and topological location (i.e., electricity consumption characteristics (first category label) and topological location (second category label), it achieves refined multi-time period and multi-objective modeling of distribution transformer operating status. Employing a strategy combining multi-time period optimization and rolling optimization, with bus voltage as the decision variable, it performs multi-time period optimization, outputting the optimal bus voltage curve and reactive power plan for future periods. This helps reduce network losses and improve voltage qualification rate. The rolling optimization algorithm dynamically adjusts the switching action based on real-time data, responding promptly to changes in the operating status of the distribution network and ensuring the timeliness and effectiveness of the governance strategy.

[0050] In some possible implementations, voltage and current time-series data of the power user side and voltage time-series data of the substation bus are collected synchronously, and the data are preprocessed by time alignment, outlier correction, missing value imputation, resampling and normalization to obtain a synchronous time-series dataset. The transformer area data includes: real-time voltage, current, active / reactive power, power factor, voltage deviation, harmonic distortion rate (THD), three-phase voltage imbalance, etc.

[0051] In some possible implementations, based on the preprocessed synchronous time-series dataset, frequency deviation, voltage deviation, three-phase voltage imbalance, voltage fluctuation and flicker, voltage sag, and total harmonic distortion (THD) are calculated to form an initial value sequence of indicators. This results in frequency deviation curves, voltage deviation curves, imbalance curves, fluctuation / Pst curves, sag event tables, and THD curves, all with the time axes fully aligned, providing sequence unification for subsequent sliding window feature extraction.

[0052] A 24-hour sliding time window was used to extract statistical features from each initial value sequence of indicators window by window, and a multidimensional statistical feature vector was constructed. The multidimensional statistical feature vector includes: mean, standard deviation, maximum value, minimum value, peak-to-peak value, 95th percentile, 5th percentile, skewness, and kurtosis. In other words, the original data was compressed, but the central tendency, dispersion, extreme values ​​and distribution pattern of the data were preserved.

[0053] Wavelet packet transform is used to perform multi-scale decomposition on the average bus voltage sequence and the average user current sequence, extract the energy proportion of each frequency band, and form a wavelet energy feature vector.

[0054] By concatenating wavelet energy feature vectors with statistical feature vectors, a comprehensive feature vector is obtained that integrates time-domain statistical characteristics and frequency-domain energy distribution. Since the comprehensive feature vector carries both time-domain statistical morphology and frequency-domain energy distribution, the complementary information makes subsequent clustering no longer solely dependent on amplitude magnitude. It can also identify complex electricity consumption characteristics such as large harmonics but small fluctuations or many transients but normal THD, thereby improving the inter-class distance.

[0055] The comprehensive feature vector is mapped to a 2-dimensional low-dimensional manifold space using nonlinear dimensionality reduction UMAP, and HDBSCAN density clustering is used to automatically identify clusters to obtain the first classification label characterizing the power consumption characteristics of the transformer area. Simultaneously, a label dictionary for the first classification label is established. In the label dictionary of the first classification label, the statistical-energy center of each cluster core sample needs to be looked up in reverse to give a semantic name: Class 0 = stable and high quality, Class 1 = light load harmonics, Class 2 = impulse fluctuations, Class 3 = frequent sags, Class 4 = three-phase imbalance, Class 5 = high harmonics and fluctuations, Class 6 = overvoltage tendency, −1 = abnormal and mixed. The triplet of number-semantic meaning-center vector is written into a YAML dictionary.

[0056] In some possible implementations, after obtaining the first classification label, the process further includes: updating the acquired transformer area data in real time, calling the same set of statistical scripts to form a new multidimensional statistical feature vector, comparing the difference between the distribution of the new multidimensional statistical feature vector and the distribution of the historical multidimensional statistical feature vector, quantifying whether the new clustering result has drifted compared with the old clustering result, and if any indicator exceeds a set threshold, determining that the electricity consumption characteristic label has drifted, triggering a hot update of the label dictionary, which can result in substantial changes in the electricity consumption characteristics of the transformer area and reduce the drift detection time.

[0057] In some possible implementations, based on the transformer-line-distribution area connection relationship provided by the main station, a directed topology tree is constructed with the main transformer as the root node. The nodes include: main transformer, line, branch switch, distribution area (distribution transformer), and the edges represent the power supply direction, which is from the main transformer to the distribution area. The tree structure is traversed using graph theory algorithms (such as DFS / BFS) to establish parent-child relationships and obtain the directed topology tree.

[0058] Construct a directed topology tree, calculate the topology depth and power supply radius from each transformer substation to the main transformer, and generate a two-dimensional topology vector:

[0059] Topology depth calculation: Starting from the root node (main transformer), use BFS / DFS to calculate the depth of each node; Power supply radius calculation: Obtain the number of levels traversed from the main transformer to the distribution area; Accumulate the lengths of each line segment on the path to obtain the electrical distance from the main transformer to the distribution area, a two-dimensional topology vector;

[0060] K-means clustering is performed on the two-dimensional topology vectors of all transformer areas to obtain a second classification label representing the topological location of the line. The second classification label includes three types of topological location labels: near end (close to the main transformer, small depth and short radius), middle section (middle position), and end (far from the main transformer, large depth and long radius). The complex network structure is simplified into three types of locations: "near-middle-end", which can better optimize the segmented power quality.

[0061] In some possible implementations, a two-dimensional key-value pair <Electrical characteristics, topological location> is constructed, with the first category label (electrical characteristics cluster ID) as the row number and the second category label (topological location cluster ID, obtained by clustering of transformer feeders and switch levels) as the column number.

[0062] Based on the first and second category labels, a joint index (electricity characteristics, topology location) is performed. The basic weight vectors of the electricity characteristics cluster and the topology location cluster at a set threshold are read from the offline calibration library. The basic weight vectors include six indicators: frequency, voltage, imbalance, fluctuation, sag, and THD.

[0063] Obtain the user-preset time period template and correction coefficient, and perform time period correction on the basic weight vector;

[0064] The feedback gain factor is obtained by non-linearly mapping the sensitive user feedback score using the Sigmoid gain function.

[0065] The final weight vector for the current time period is obtained by multiplying the time-period-corrected weight vector by the feedback gain factor.

[0066] Iterate through all time periods within the cycle and stack them in chronological order to form a multi-time period weight coefficient matrix.

[0067] In some possible implementations, the master station aggregates the voltage-harmonic-imbalance weighting coefficient matrices uploaded by each transformer substation, constructs a global weighted dataset according to the substation number and time stamp, and thus calculates the weighting coefficient matrix for each substation i in time period t:

[0068] ;

[0069] in: Indicates voltage deviation. Indicates the degree of harmonic exceedance. This represents the degree of imbalance, α+β+γ=1, which can be adjusted according to business priorities.

[0070] Output: Global weight dataset Indexed by station area number + time stamp.

[0071] A multi-period optimization model is established using the voltage amplitude of the low-voltage side bus of the main transformer and the output of the reactive power compensation equipment in the next T time periods as decision variables. , ;in, For the weighted voltage pass rate, For the number of low-voltage users, Indicates voltage. Indicates an indicator function, The objective function is to maximize the weighted voltage qualification rate and minimize the number of low-voltage users; the constraints include: upper and lower limits of bus voltage safety, limit of the number of main transformer tap changers, capacity and ramping limit of reactive power equipment, and power flow equation of distribution network.

[0072] Using the voltage amplitude of the low-voltage busbar on the main transformer and the output of reactive power compensation equipment in each future time period as decision variables, a multi-time period optimization model is established. The objective function is to maximize the weighted voltage qualification rate and minimize the number of low-voltage users. The constraints include: upper and lower limits of busbar voltage safety, limits on the number of tap changers of the main transformer, limits on the capacity and ramp-up limits of reactive power equipment, and power flow equations of the distribution network. The multi-time period optimization model is solved to obtain the optimal busbar voltage curve and reactive power plan for each future time period.

[0073] In some possible implementations, in each rolling time slot, the master station loads the weight coefficient matrix of the next M time periods and the corresponding optimal bus voltage curve into the rolling window. The master station is responsible for global voltage curve optimization (day-ahead / rolling), and the distribution area (including OLTC distribution transformers) makes distributed discrete adjustment decisions within the rolling window based on local weights and the master station's reference trajectory.

[0074] Using the number of on-load tap changer taps in each transformer area within the window as discrete decision variables, a short-term rolling optimization model is established. The objective function is set as minimizing the weighted sum of the transformer area voltage over-limit risk and the tap changer operation cost, where the voltage over-limit risk is dynamically weighted by the voltage weight coefficient of each transformer area. The objective function is: ,in, This indicates a risk of voltage exceeding limits. , This represents the cost of gear adjustment, where constraints include: upper and lower limits of the gear position, remaining amount of the maximum number of adjustments per day, limit of gear position change between adjacent time periods, and tolerance for voltage reference trajectory deviation derived from the optimal bus voltage curve; solve the objective function and output the optimal gear adjustment command for each transformer area in the current time period.

[0075] Using the voltage-harmonic-imbalance weighting coefficient matrix of the transformer substation and the optimal bus voltage curve of the main station as dual inputs, and the number of tap positions of the on-load tap changer as discrete decision variables, the rolling optimization algorithm initiates short-time domain model predictive control at every interval t on the transformer substation side. Its technical effect goes beyond single-point voltage regulation; it simultaneously generates quantifiable, closed-loop, and perceptible systemic improvements along the three main lines of power quality: voltage, harmonics, and imbalance. The rolling optimization strategy translates the globally optimal bus voltage curve into the minimum number of local actions, simultaneously achieving early prevention, real-time correction, and minimal equipment movement along the three national standard red lines for voltage, harmonics, and imbalance. This elevates power quality from post-event remediation to pre-event self-governance, providing a replicable voltage base solution for the integration of high-proportion renewable energy and high-proportion power electronic loads in new power systems.

[0076] As one possible implementation, this embodiment provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the program, it implements a multi-time-period, multi-weighted real-time power quality optimization decision-making method.

[0077] As one possible implementation, this embodiment provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements a multi-time-period, multi-weighted real-time power quality optimization decision-making method.

[0078] The specific embodiments described above further illustrate the purpose, technical solution, and beneficial effects of the present invention. It should be understood that the above description is only a specific embodiment of the present invention and is not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A multi-time-period, multi-weighted real-time power quality optimization decision-making method, characterized in that, The specific steps include the following: Acquire transformer area data and preprocess the transformer area data; Based on the preprocessed data, the first classification label characterizing the power consumption characteristics of the transformer area is obtained; Based on the connection relationship between the main station transformer, the line, and the distribution area, the distribution network topology is reclassified to obtain a second classification label that represents the location of the line topology. Based on the first and second category labels, a multi-time period weight matrix is ​​dynamically generated for each transformer area at different time periods, using the voltage-harmonic-imbalance weight coefficients. Using the voltage-harmonic-unbalance weighting coefficient matrix of each transformer area in different time periods within a preset time period as input, and the bus voltage as the decision variable, multi-time period optimization calculation is performed to output the optimal bus voltage curve and reactive power plan for future time periods. Using the voltage-harmonic-imbalance weighting coefficient matrix of each transformer area at different time periods and the optimal bus voltage curve as inputs, and the switching action of each transformer area as the decision variable, a rolling optimization algorithm is used to calculate and generate the real-time management strategy for each transformer area.

2. The multi-time period, multi-weight power quality real-time optimization decision-making method according to claim 1, characterized in that, The acquisition of transformer area data and the preprocessing of the transformer area data specifically include: Synchronously collect voltage and current time-series data from power users and substation bus voltage time-series data, and perform time alignment, outlier correction, missing value imputation, resampling and normalization preprocessing on the data to obtain a synchronous time-series dataset.

3. The multi-time period, multi-weight power quality real-time optimization decision-making method according to claim 1, characterized in that, Based on the preprocessed data, a first classification label characterizing the electricity consumption characteristics of the transformer area is obtained, specifically including: Based on the preprocessed synchronous time series dataset, frequency deviation, voltage deviation, three-phase voltage imbalance, voltage fluctuation and flicker, voltage sag and total harmonic distortion rate are calculated to form an initial value sequence of indicators. A sliding time window is used to extract statistical features from the initial value sequence of the index, and a multidimensional statistical feature vector is constructed. Wavelet packet transform is used to decompose the bus voltage and user current into multiple scales, extract the energy proportion of each frequency band, and form a wavelet energy feature vector. By concatenating the wavelet energy feature vector with the statistical feature vector, a comprehensive feature vector integrating time-domain statistical characteristics and frequency-domain energy distribution is obtained. The comprehensive feature vector is mapped to a low-dimensional manifold space through nonlinear dimensionality reduction, and HDBSCAN density clustering is used to obtain the first classification label characterizing the power consumption characteristics of the transformer area. At the same time, a label dictionary for the first classification label is established.

4. The multi-time period, multi-weight power quality real-time optimization decision-making method according to claim 3, characterized in that, After obtaining the first category label, it also includes: The system updates the acquired transformer area data in real time and generates new multidimensional statistical feature vectors. It compares the distribution of the new multidimensional statistical feature vectors with the distribution of historical multidimensional statistical feature vectors to quantify whether the new clustering results have drifted with the old clustering results. If any indicator exceeds the set threshold, it is determined that the electricity consumption characteristic tag has drifted, triggering a hot update of the tag dictionary.

5. The multi-time period, multi-weight power quality real-time optimization decision-making method according to claim 1, characterized in that, The distribution network topology is reclassified based on the connection relationship between the main station transformer, the line, and the transformer substation to obtain a second classification label representing the location of the line topology, specifically including: A directed topology tree is constructed based on the transformer-line-transformer area connection relationship provided by the main station. The topology depth and power supply radius of each transformer area to the main transformer are calculated, and a two-dimensional topology vector is generated. K-means clustering is performed on the two-dimensional topology vector to obtain a second classification label representing the topological location of the line.

6. The multi-time period, multi-weight power quality real-time optimization decision-making method according to claim 1, characterized in that, Based on the first and second category labels, a multi-time period weight matrix is ​​dynamically generated for each transformer area at different time periods, using the voltage-harmonic-imbalance weight coefficients. Based on the joint indexing of the first and second classification labels, the basic weight vectors of the power consumption characteristic cluster and the topological location cluster at a set threshold are read from the offline calibration library. Obtain the user-preset time period template and correction coefficient, and perform time period correction on the basic weight vector; The feedback gain factor is obtained by non-linearly mapping the sensitive user feedback score using the Sigmoid gain function. The final weight vector for the current time period is obtained by multiplying the time-period-corrected weight vector by the feedback gain factor. Iterate through all time periods within the cycle and stack them in chronological order to form a multi-time period weight coefficient matrix.

7. The multi-time period, multi-weight real-time power quality optimization decision-making method according to claim 1, characterized in that, The output of the optimal bus voltage curve and reactive power plan for future time periods specifically includes: The main station collects the voltage-harmonic-imbalance weight coefficient matrix uploaded by each transformer area and constructs a global weight dataset according to the transformer area number and time stamp. Using the voltage amplitude of the low-voltage busbar on the main transformer and the output of the reactive power compensation equipment in each future time period as decision variables, a multi-time period optimization model is established. The objective function is to maximize the weighted voltage qualification rate and minimize the number of low-voltage users. The constraints include: upper and lower limits of busbar voltage safety, limit of the number of tap changers of the main transformer, capacity and ramping limit of reactive power equipment, and power flow equation of the distribution network. The multi-period optimization model is solved to obtain the optimal bus voltage curve and reactive power plan for each future period.

8. The multi-time-period, multi-weighted real-time power quality optimization decision-making method according to claim 1, characterized in that, The generation of real-time governance strategies for each transformer area specifically includes: In each rolling time slot, the master station loads the weight coefficient matrix for the next M time periods and the corresponding optimal bus voltage curve into the rolling window; Using the number of on-load tap changer taps in each transformer area within the window as discrete decision variables, a short-term rolling optimization model is established. The objective function is set as minimizing the weighted sum of the transformer area voltage over-limit risk and the tap change operation cost, where the voltage over-limit risk is dynamically weighted by the voltage weight coefficient of each transformer area. The constraints include: tap upper and lower limits, remaining amount of maximum daily adjustment times, tap change limit between adjacent time periods, and voltage reference trajectory deviation tolerance derived from the optimal bus voltage curve. Solve the objective function and output the optimal dispatching command for each station area in the current time period.

9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the program, it implements the multi-time period multi-weight power quality real-time optimization decision-making method as described in any one of claims 1 to 8.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When executed by the processor, the program implements the multi-time period, multi-weight power quality real-time optimization decision-making method as described in any one of claims 1 to 8.

Images