Active transformer area voltage out-of-limit early warning and photovoltaic bearing capacity evaluation method

Abstract

The invention provides an active transformer area voltage out-of-limit early warning and photovoltaic bearing capacity evaluation method, which comprises the following steps: acquiring multi-source data of a power distribution transformer area, and generating a standardized multi-source time sequence data set through time alignment processing, data quality evaluation and feature fusion; extracting voltage out-of-limit features based on the time sequence similarity matrix and the node influence matrix; the periodicity, volatility and randomness characteristics of photovoltaic output are analyzed, a coupling relation matrix and a sensitivity vector are established, and an influence evaluation model is constructed; performing out-of-limit early warning based on an early warning index system and a deep learning model; and establishing a multi-dimensional evaluation index system for carrying out bearing capacity evaluation in combination with power flow analysis and the dynamic influence factors. Through multi-source data fusion and multi-dimensional feature analysis, the accuracy of voltage out-of-limit early warning is improved, accurate evaluation of the photovoltaic bearing capacity is realized, and effective technical support is provided for safe operation of the power distribution network.

Description

Technical Field

[0001] The present invention relates to the technology related to distribution networks, in particular to an active substation voltage over-limit warning and photovoltaic carrying capacity evaluation method. Background Art

[0002] With the large-scale access of photovoltaic power generation to the distribution network, active distribution substations are facing increasingly complex voltage management challenges. The intermittent and fluctuating characteristics of photovoltaic power generation make it difficult for traditional voltage regulation methods to meet the operation requirements under the new situation. Especially in distribution substations with a high photovoltaic penetration rate, the risk of voltage over-limit increases significantly, which not only affects the power quality and power consumption safety, but also may lead to the frequent disconnection of photovoltaic power generation units. Therefore, establishing an effective voltage over-limit warning mechanism and a scientific photovoltaic carrying capacity evaluation method has important practical significance for improving the operation reliability of the distribution network and promoting the consumption of photovoltaic energy.

[0003] Currently, domestic and foreign scholars have carried out a large number of studies on distribution network voltage over-limit warning and photovoltaic carrying capacity evaluation. In terms of voltage over-limit warning, mainly statistical analysis-based methods are used to establish warning indicators, or simple machine learning models are used to evaluate the over-limit risk. These methods often rely on a single data source, and the construction of warning indicators is relatively simple, making it difficult to fully reflect the dynamic characteristics of the system. In terms of photovoltaic carrying capacity evaluation, traditional methods mostly determine the theoretical access capacity based on static power flow calculation or make rough estimates through empirical formulas. These evaluation methods mainly consider steady-state constraint conditions and insufficiently consider the dynamic impact of photovoltaic output fluctuations, and the evaluation results are often too conservative or radical.

[0004] However, the existing technologies still have the following technical problems in practical applications: First, in terms of data processing, the time alignment and quality evaluation mechanisms for multi-source heterogeneous data are imperfect. Especially when processing data with different sampling periods such as photovoltaic power generation, meteorology, and load, it is difficult to ensure the temporal consistency and reliability of the data. Second, the voltage feature extraction method is relatively single and fails to effectively combine temporal features and topological features, resulting in insufficient accuracy in over-limit feature recognition. Third, the existing photovoltaic impact evaluation methods often ignore the fluctuation characteristics of different frequency scales and do not fully consider the regulatory effect of load changes, affecting the accuracy of the evaluation results. In addition, the self-adaptability of the warning model is insufficient, making it difficult to dynamically adjust model parameters according to data quality, and lacking a reliable model performance evaluation mechanism. Finally, in terms of carrying capacity evaluation, the existing methods lack a systematic consideration of dynamic constraint conditions and fail to establish an effective multi-dimensional evaluation index system, making it difficult to provide specific technical support for system optimization. The existence of these technical problems seriously restricts the safety of distribution network operation and the economy of photovoltaic access. Summary of the Invention

[0005] The object of the present invention is to propose a method for active substation voltage over - limit warning and photovoltaic carrying capacity assessment. Through multi - source data fusion and multi - dimensional feature analysis, the accuracy of voltage over - limit warning is improved, the accurate assessment of photovoltaic carrying capacity is realized, and effective technical support is provided for the safe operation of the distribution network.

[0006] To achieve the above object, the technical solution of the present invention is: a method for active substation voltage over - limit warning and photovoltaic carrying capacity assessment, comprising the following steps:

[0007] Step S1: Obtain the original voltage monitoring data, photovoltaic power generation data, meteorological data, and load data of the distribution substation, generate a time - aligned data set through time alignment processing; perform data quality assessment and correction on the time - aligned data set to generate a cleaned data set; extract voltage and photovoltaic features from the cleaned data set and perform data fusion to form a fusion feature set; perform standardization processing on the fusion feature set and associate it with the distribution network topology information, and output a standardized multi - source time - series data set.

[0008] Step S2: Read the standardized multi - source time - series data set, obtain time - window sequences through time - window segmentation processing, calculate voltage similarity to obtain a time - series similarity matrix; construct a network connection matrix according to the distribution network topology information, analyze the node influence relationship to generate a node influence matrix; extract voltage over - limit features based on the time - series similarity matrix and the node influence matrix, and output a voltage over - limit feature set.

[0009] Step S3: Read the photovoltaic data in the standardized multi - source time - series data set, analyze its periodic, fluctuating, and random characteristics to generate a photovoltaic feature combination; establish the correlation between photovoltaic output and voltage change to form a coupling relationship matrix and a sensitivity vector; construct an impact assessment model based on the coupling relationship matrix and the sensitivity vector, and output a photovoltaic impact feature set.

[0010] Step S4: Read the voltage over - limit feature set, construct an early - warning index system to obtain an early - warning index set, design an early - warning threshold in combination with the photovoltaic impact feature set to obtain a threshold matrix; train an early - warning model based on the standardized multi - source time - series data set to generate a model parameter set; perform early - warning analysis on real - time monitoring data and output an over - limit early - warning signal.

[0011] Step S5: Read the distribution network parameters for power flow analysis to generate a theoretical capacity matrix; analyze the influence of photovoltaic output fluctuations to obtain a dynamic influence factor matrix; establish a multi - dimensional assessment index system, and perform carrying capacity assessment based on the theoretical capacity matrix and the dynamic influence factor matrix, and output a photovoltaic carrying capacity assessment report.

[0012] Preferably, step S1 is specifically:

[0013] Step S11: Receive the original voltage monitoring data of the distribution substation area, and extract the timestamp, measurement point identifier, and voltage value information therein; receive the photovoltaic power generation data, and extract the timestamp, installed capacity, and actual power generation information therein; receive the meteorological data, and extract the timestamp, irradiance, temperature, and cloud cover information therein; receive the load data, and extract the timestamp, active power, and reactive power information therein; align the above data according to a unified time scale, and output a time-aligned data set;

[0014] Step S12: Read the time-aligned data set, calculate the missing degree of various types of data to obtain a missing rate index; complete the data according to the missing rate index to generate a completed data set; perform outlier processing on the completed data set, and output a cleaned data set; quantitatively calculate the credibility of the cleaned data set to generate a data quality evaluation matrix;

[0015] Step S13: Read the voltage data in the cleaned data set, extract its statistical features to generate a voltage feature set; process the photovoltaic data in the cleaned data set, extract its fluctuation features to generate a photovoltaic feature set; calculate the expected value of photovoltaic power generation in combination with the meteorological data in the cleaned data set to generate a photovoltaic prediction set; fuse the voltage feature set, photovoltaic feature set, and photovoltaic prediction set to output a fused feature set;

[0016] Step S14: Read the fused feature set, perform numerical normalization processing to generate a standardized feature set; read the distribution network topology information, establish an association between the standardized feature set and the topology information to generate an associated feature set; integrate the standardized feature set, associated feature set, and data quality evaluation matrix, and output a standardized multi-source time series data set.

[0017] Preferably, step S2 is specifically as follows:

[0018] Step S21: Read the standardized multi-source time series data set, segment the data using a sliding window method to generate a time window sequence; calculate the similarity degree between the voltage data sequences of each monitoring point to form a time series similarity matrix; extract the time series features of voltage changes from the time series similarity matrix, and output a time series pattern set;

[0019] Step S22: Read the distribution network topology information, construct a network connection matrix representing the node connection relationship; analyze the voltage influence degree between each node to generate a node influence matrix; identify the nodes that have a significant impact on voltage over-limit from the node influence matrix, and output a key node set;

[0020] Step S23: Read the time series pattern set and the node influence matrix, extract the feature information of voltage over-limit to generate an over-limit feature vector; analyze the correlation features of voltage over-limit in the time and space dimensions to generate an over-limit correlation matrix; integrate the over-limit feature vector and the over-limit correlation matrix, and output a voltage over-limit feature set.

[0021] Preferably, step S3 is specifically as follows:

[0022] Step S31: Read the photovoltaic data in the standardized multi-source time series dataset, analyze its periodic change characteristics, and generate a periodic feature set; extract the fluctuation information of the photovoltaic output to generate a fluctuation feature set; analyze the random change characteristics of the photovoltaic output, and output a random feature set;

[0023] Step S32: Read the photovoltaic output data and voltage change data, establish the corresponding relationship between the two, and generate a coupling relationship matrix; calculate the influence degree of the photovoltaic output change on the voltage to form a sensitivity vector; analyze the influence differences in different time periods, and output a time period influence matrix;

[0024] Step S33: Read the coupling relationship matrix and the sensitivity vector, construct an influence evaluation model; quantitatively analyze the influence degree of photovoltaic power generation on the voltage to generate an influence evaluation matrix; integrate all evaluation results, and output a photovoltaic influence feature set.

[0025] Preferably, step S4 is specifically as follows:

[0026] Step S41: Read the voltage over-limit feature set, construct a multi-dimensional early warning index system, and generate an early warning index set; read the photovoltaic influence feature set, calculate the dynamic early warning threshold, and form a threshold matrix; determine the division standard of the early warning level, and output an early warning level set;

[0027] Step S42: Read the standardized multi-source time series dataset, the early warning index set and the threshold matrix, and construct a deep learning early warning model; optimize the model training process in combination with the data quality evaluation matrix to generate a model parameter set;

[0028] Step S43: Read the real-time monitoring data and the model evaluation index, and analyze them using the trained early warning model; correct the credibility of the early warning result based on the model evaluation index, calculate the voltage over-limit risk degree, and generate a risk evaluation vector; according to the risk evaluation result, output an over-limit early warning signal.

[0029] Preferably, step S5 is specifically as follows:

[0030] Step S51: Read the distribution network parameter information, establish a power flow analysis model, and generate a network model; calculate the theoretical maximum carrying capacity considering the voltage margin constraint to form a theoretical capacity matrix; analyze the network structure characteristics, and output a network feature set;

[0031] Step S52: Read the photovoltaic influence feature set, analyze the influence of the output fluctuation, and generate a dynamic influence factor matrix; read the over-limit early warning signal, analyze the historical over-limit situation, and form a historical over-limit feature set; analyze the influence of the load characteristics, and output a load influence matrix;

[0032] Step S53: Read the theoretical capacity matrix, dynamic impact factor matrix, historical overlimit feature set, network feature set, and data quality assessment matrix, construct a multi-dimensional evaluation index system; evaluate the rationality of the existing photovoltaic access capacity, generate a capacity evaluation result; analyze the system optimization space based on the capacity evaluation result, generate an optimization recommendation set; integrate the capacity evaluation result and the optimization recommendation set, and output a photovoltaic bearing capacity evaluation report.

[0033] Preferably, step S11 is specifically as follows:

[0034] Step S111: Read the original voltage monitoring data, extract the timestamp information to generate a time index set; detect the continuity of the timestamps, mark the time interval abnormal points, generate a time anomaly marking set; perform time dimension preprocessing on the original voltage monitoring data based on the time anomaly marking set to generate a preprocessed voltage data set;

[0035] Step S112: Read the photovoltaic power generation data, match its timestamp with the time index set to generate a photovoltaic time mapping table; extract the installed capacity and actual power generation information, and reconstruct the data sequence in combination with the photovoltaic time mapping table to generate a reconstructed photovoltaic data set;

[0036] Step S113: Read the meteorological data, establish a time correspondence relationship using the time index set to generate a meteorological time mapping table; extract the irradiance, temperature, and cloud cover information, and reconstruct the data sequence in combination with the meteorological time mapping table to generate a reconstructed meteorological data set;

[0037] Step S114: Read the load data, establish a time correspondence relationship using the time index set to generate a load time mapping table; extract the active power and reactive power information, and reconstruct the data sequence in combination with the load time mapping table to generate a reconstructed load data set;

[0038] Step S115: Read the preprocessed voltage data set, reconstructed photovoltaic data set, reconstructed meteorological data set, and reconstructed load data set, construct a data alignment matrix with a unified time scale to generate an initial alignment matrix; detect the data alignment quality in the initial alignment matrix to generate an alignment quality index set; optimize the initial alignment matrix based on the alignment quality index set, and output a time-aligned data set.

[0039] Step S13 is specifically as follows:

[0040] Step S131: Read the voltage data in the cleaned dataset, calculate the maximum value, minimum value, and mean value to generate a basic statistic set; use the basic statistic set to calculate the standard deviation and variance to generate a fluctuation statistic set; read the voltage data in the cleaned dataset, calculate the difference sequence and change rate of the data to generate a change feature set; read the voltage data in the cleaned dataset, count the distribution parameters and quantiles of the data to generate a distribution feature set; combine the basic statistic set, fluctuation statistic set, change feature set, and distribution feature set, and output the voltage feature set;

[0041] Step S132: Read the photovoltaic data in the cleaned dataset, calculate the power difference sequence at adjacent moments to generate a power change sequence; perform statistical analysis on the power change sequence to generate a change statistic set; read the photovoltaic data in the cleaned dataset, calculate the fluctuation components at different time scales to generate a multi-scale fluctuation set; read the multi-scale fluctuation set, calculate the fluctuation amplitude and frequency characteristics to generate a fluctuation feature set; combine the change statistic set, multi-scale fluctuation set, and fluctuation feature set, and output the photovoltaic feature set;

[0042] Step S133: Read the meteorological data in the cleaned dataset, extract irradiance, temperature, and cloud cover data to generate a meteorological element set; use the temperature data in the meteorological element set to calculate the temperature coefficient of the photovoltaic panel to generate a temperature coefficient set; read the cloud cover data in the meteorological element set to calculate the irradiance attenuation coefficient to generate an attenuation coefficient set; read the meteorological element set, temperature coefficient set, and attenuation coefficient set, and combine with the standard parameters of the photovoltaic panel to calculate the predicted power generation value, and output the photovoltaic prediction set;

[0043] Step S134: Read the voltage feature set, photovoltaic feature set, and photovoltaic prediction set, calculate the correlation coefficients between the features to generate a correlation coefficient matrix; use the correlation coefficient matrix to screen the features to generate a feature screening set; read the feature screening set, construct feature combination items to generate a combined feature set; integrate the feature screening set and the combined feature set, and output the fusion feature set.

[0044] Preferably, step S21 is specifically as follows:

[0045] Step S211: Read the standardized multi-source time series dataset, calculate the time interval features of the data to generate a time feature sequence; use the time feature sequence to determine the optimal window length to generate a window parameter set; segment the data according to the window parameter set to generate an initial segmented sequence; optimize the boundary of the initial segmented sequence, and output the time window sequence;

[0046] Step S212: Read the voltage data in the time window sequence, calculate the numerical distribution characteristics of each monitoring point, and generate a distribution characteristic matrix; calculate the Euclidean distance between monitoring points using the distribution characteristic matrix to generate a distance matrix; read the time window sequence, calculate the dynamic time warping distance between monitoring points, and generate a dynamic distance matrix; perform weighted fusion on the distance matrix and the dynamic distance matrix, and output a time series similarity matrix.

[0047] Step S213: Read the time window sequence and the time series similarity matrix, identify typical voltage change segments, and generate a set of typical sequences; perform clustering analysis on the set of typical sequences to generate a set of clustering centers; read the set of typical sequences and the set of clustering centers, extract pattern feature parameters, and generate a set of pattern parameters; perform pattern matching on all sequences according to the set of pattern parameters, and output a time series pattern set.

[0048] Step S23 specifically includes:

[0049] Step S231: Read the time series pattern set, count the over-limit amplitude information of each monitoring point, and generate an over-limit amplitude sequence; read the node influence matrix, extract the influence weights between nodes, and generate a node weight vector; perform feature calculation by combining the over-limit amplitude sequence and the node weight vector to generate a basic feature matrix; perform dimensionality reduction processing on the basic feature matrix, and output an over-limit feature vector.

[0050] Step S232: Read the time series pattern set, extract the timestamp information of over-limit events, and generate a time mark sequence; read the node influence matrix, construct the node connection relationship, and generate a connection relationship graph; calculate the propagation time series based on the time mark sequence and the connection relationship graph to generate a set of propagation sequences; perform statistical analysis on the set of propagation sequences, and output an over-limit correlation matrix.

[0051] Step S233: Read the over-limit feature vector, calculate the combination relationship between features, and generate a set of feature combinations; read the over-limit correlation matrix, extract key correlation patterns, and generate a set of correlation patterns; perform feature fusion on the set of feature combinations and the set of correlation patterns to generate a set of fusion features; perform standardization processing on the set of fusion features, and output a voltage over-limit feature set.

[0052] Preferably, Step S32 specifically includes:

[0053] Step S321: Read the photovoltaic output data, calculate the output change rate, and generate an output change sequence; read the voltage change data, calculate the voltage deviation value, and generate a voltage deviation sequence; align the output change sequence and the voltage deviation sequence in time to generate a time-aligned matrix; perform data matching on the time-aligned matrix, and output a coupling relationship matrix.

[0054] Step S322: Read the coupling relationship matrix, extract key change points, and generate a key point sequence; read the coupling relationship matrix, calculate the change trend characteristics, and generate a trend characteristic set; combine the key point sequence and the trend characteristic set to calculate the response characteristics, and generate a response characteristic matrix; perform statistical analysis on the response characteristic matrix and output a sensitivity vector.

[0055] Step S323: Read the coupling relationship matrix, divide the data by time period to generate a time-sharing data set; extract features for each time period of the time-sharing data set to generate a time period characteristic set; read the sensitivity vector, calculate the sensitivity distribution of each time period, and generate a time period distribution matrix; fuse the time period characteristic set and the time period distribution matrix and output a time period impact matrix.

[0056] Preferably, step S42 is specifically as follows:

[0057] Step S421: Read the standardized multi-source time series data set, divide it into a training set and a validation set in chronological order to generate a data set division matrix; read the early warning index set, construct sample labels, and generate a label sequence; read the data quality evaluation matrix, calculate sample weights, and generate a sample weight vector; combine the data set division matrix, the label sequence, and the sample weight vector and output a training data set.

[0058] Step S422: Read the training data set and the threshold matrix, construct the input layer structure, and generate an input feature matrix; design the network hierarchy according to the dimension of the input feature matrix to generate network structure parameters; read the sample weight vector, configure the loss function parameters, and generate an optimization parameter set; integrate the network structure parameters and the optimization parameter set and output initial model parameters.

[0059] Step S423: Read the training data set and the initial model parameters, perform model training, and generate a training process matrix; read the training process matrix, adjust the learning parameters, and generate a parameter adjustment sequence; update the model parameters according to the parameter adjustment sequence to generate an updated parameter set; perform optimal selection on the updated parameter set and output a model parameter set.

[0060] Step S424: Read the training data set and the model parameter set, calculate the prediction results, and generate a prediction result sequence; compare the prediction result sequence with the label sequence to generate an error statistics matrix; read the error statistics matrix, calculate the evaluation indicators, and generate an indicator calculation result; perform summary analysis on the indicator calculation result and output a model evaluation indicator.

[0061] Step S52 is specifically as follows:

[0062] Step S521: Read the photovoltaic impact feature set, extract the fluctuation feature sequence, and generate a fluctuation sequence matrix; perform frequency decomposition on the fluctuation sequence matrix to generate a frequency component set; read the frequency component set, calculate the impact weights of each frequency component, and generate a frequency weight vector; perform weighted combination on the frequency component set and the frequency weight vector, and output the dynamic impact factor matrix;

[0063] Step S522: Read the over-limit warning signal, count the time distribution of over-limit events, and generate a time distribution sequence; perform pattern recognition on the time distribution sequence to generate a pattern feature set; read the pattern feature set, extract key feature parameters, and generate a feature parameter matrix; perform feature integration on the feature parameter matrix, and output the historical over-limit feature set;

[0064] Step S523: Read the historical over-limit feature set, extract the load change information, and generate a load change sequence; perform correlation analysis on the load change sequence to generate a correlation coefficient matrix; read the correlation coefficient matrix, calculate the impact degree index, and generate an impact index set; convert the impact index set into an impact feature matrix, and output the load impact matrix;

[0065] Step S53 specifically includes:

[0066] Step S531: Read the theoretical capacity matrix and the network feature set, extract the capacity constraint conditions, and generate a constraint condition set; read the dynamic impact factor matrix, calculate the dynamic adjustment coefficient, and generate an adjustment coefficient matrix; read the historical over-limit feature set, extract the over-limit impact parameters, and generate an impact parameter set; perform index conversion on the constraint condition set, the adjustment coefficient matrix, and the impact parameter set, and output the evaluation index system;

[0067] Step S532: Read the evaluation index system, calculate the weight coefficients of each index, and generate a weight coefficient vector; read the theoretical capacity matrix, perform capacity analysis and calculation, and generate a capacity analysis matrix; perform weighted calculation on the capacity analysis matrix according to the weight coefficient vector, and generate a weighted result set; perform threshold judgment on the weighted result set, and output the capacity evaluation result;

[0068] Step S533: Read the capacity evaluation result and the evaluation index system, analyze the index differences, and generate a difference feature set; perform clustering analysis on the difference feature set, and generate a clustering result matrix; read the clustering result matrix, identify the optimization direction, and generate an optimization direction set; convert the optimization direction set into specific suggestions, and output the optimization suggestion set;

[0069] Step S534: Read the capacity evaluation result, extract the key evaluation information, and generate an evaluation summary set; read the optimization suggestion set, sort out the optimization measures, and generate a measure list set; perform formatting processing on the evaluation summary set and the measure list set, and generate a report structure set; fill in the content of the report structure set, and output the photovoltaic carrying capacity evaluation report.

[0070] Compared with the prior art, the present invention has the following beneficial effects:

[0071] Through multi-source data fusion and multi-dimensional feature analysis, the accuracy of voltage over-limit warning is improved, the accurate assessment of photovoltaic carrying capacity is realized, and effective technical support is provided for the safe operation of the distribution network. BRIEF DESCRIPTION OF THE DRAWINGS

[0072] Figure 1 It is a flowchart of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0073] The technical solution of the present invention will be specifically described below with reference to the drawings.

[0074] As Figure 1 shown, according to one aspect of the present application, an active substation voltage over-limit warning and photovoltaic carrying capacity assessment method includes the following steps:

[0075] Step S1: Obtain the original voltage monitoring data, photovoltaic power generation data, meteorological data and load data of the distribution substation, generate a time-aligned data set through time alignment processing; perform data quality assessment and correction on the time-aligned data set to generate a cleaned data set; extract voltage and photovoltaic features from the cleaned data set and perform data fusion to form a fusion feature set; perform standardization processing on the fusion feature set and associate it with the distribution network topology information, and output a standardized multi-source time series data set.

[0076] Step S2: Read the standardized multi-source time series data set, obtain time window sequences through time window segmentation processing, calculate voltage similarity to obtain a time series similarity matrix; construct a network connection matrix according to the distribution network topology information, analyze node influence relationships to generate a node influence matrix; extract voltage over-limit features based on the time series similarity matrix and the node influence matrix, and output a voltage over-limit feature set.

[0077] Step S3: Read the photovoltaic data in the standardized multi-source time series data set, analyze its periodic, fluctuating and random characteristics to generate a photovoltaic feature combination; establish an association relationship between photovoltaic output and voltage change to form a coupling relationship matrix and a sensitivity vector; construct an impact assessment model based on the coupling relationship matrix and the sensitivity vector, and output a photovoltaic impact feature set.

[0078] Step S4: Read the voltage over-limit feature set, construct a warning index system to obtain a warning index set, design a warning threshold in combination with the photovoltaic impact feature set to obtain a threshold matrix; train a warning model based on the standardized multi-source time series data set to generate a model parameter set; perform warning analysis on real-time monitoring data, and output an over-limit warning signal.

[0079] Step S5: Read the distribution network parameters for power flow analysis to generate a theoretical capacity matrix; analyze the impact of photovoltaic output fluctuations to obtain a dynamic impact factor matrix; establish a multi-dimensional evaluation index system, and perform bearing capacity evaluation based on the theoretical capacity matrix and the dynamic impact factor matrix, and output a photovoltaic bearing capacity evaluation report.

[0080] In summary, in this embodiment, through the intelligent fusion and quality evaluation mechanism of multi-source data, the reliability of the analysis basic data is ensured; through the voltage over-limit analysis method based on time series characteristics and topological characteristics, the accuracy of over-limit feature recognition is improved; through the multi-dimensional quantitative evaluation method of photovoltaic impact, a scientific basis is provided for system optimization; through the early warning model and dynamic threshold mechanism, the accuracy and adaptability of early warning are improved; through the bearing capacity evaluation method considering static capacity and dynamic characteristics, reliable support is provided for photovoltaic access decision-making. The overall solution forms a closed-loop intelligent analysis system, significantly improving the safety and reliability of distribution network operation.

[0081] According to one aspect of the present application, step S1 is specifically as follows:

[0082] Step S11: Receive the original voltage monitoring data of the distribution substation area, and extract the timestamp, measurement point identifier, and voltage value information therein; receive the photovoltaic power generation data, and extract the timestamp, installed capacity, and actual power generation information therein; receive the meteorological data, and extract the timestamp, irradiance, temperature, and cloud cover information therein; receive the load data, and extract the timestamp, active power, and reactive power information therein; align the above data according to a unified time scale, and output a time-aligned data set.

[0083] Step S12: Read the time-aligned data set, calculate the missing degree of various data to obtain a missing rate index; perform data completion processing based on the missing rate index to generate a completed data set; perform outlier processing on the completed data set, and output a cleaned data set; perform quantitative calculation on the credibility of the cleaned data set to generate a data quality evaluation matrix.

[0084] Step S13: Read the voltage data in the cleaned data set, extract its statistical characteristics, and generate a voltage feature set; process the photovoltaic data in the cleaned data set, extract its fluctuation characteristics, and generate a photovoltaic feature set; calculate the expected value of photovoltaic power generation in combination with the meteorological data in the cleaned data set to generate a photovoltaic prediction set; perform feature fusion on the voltage feature set, the photovoltaic feature set, and the photovoltaic prediction set, and output a fused feature set.

[0085] Step S14: Read the fused feature set, perform numerical normalization processing to generate a standardized feature set; read the distribution network topology information, establish an association between the standardized feature set and the topology information to generate an associated feature set; integrate the standardized feature set, the associated feature set, and the data quality evaluation matrix, and output a standardized multi-source time series data set.

[0086] Through the time alignment and data quality assessment mechanism of multi-source data, high-quality data fusion is achieved. Specifically, alignment processing with a unified time scale is adopted to ensure the consistency of data from different sources (voltage monitoring, photovoltaic power generation, meteorological, and load data) in the time dimension, avoiding analysis biases caused by data time mismatches. By introducing a data quality assessment matrix, the credibility of the data is quantitatively evaluated, and the data quality is improved through completion processing and outlier processing, making the input data for subsequent analysis more reliable. During the feature fusion process, by extracting voltage statistical features and photovoltaic fluctuation features, and calculating the expected value of photovoltaic power generation in combination with meteorological data, a multi-dimensional feature expression is formed. This feature fusion method can more comprehensively describe the system operation state. Finally, through numerical normalization and topological correlation processing, features with different dimensions can be analyzed uniformly, while the physical connection relationship of the distribution network is retained, providing a standardized high-quality data basis for subsequent analysis. This multi-level data processing scheme significantly improves the usability and analysis accuracy of the data.

[0087] According to one aspect of the present application, step S2 is specifically as follows:

[0088] Step S21: Read the standardized multi-source time series data set, segment the data using a sliding window method to generate a time window sequence; calculate the similarity degree between the voltage data sequences of each monitoring point to form a time series similarity matrix; extract the time series features of voltage changes from the time series similarity matrix, and output a time series pattern set.

[0089] Step S22: Read the distribution network topology information, construct a network connection matrix representing the node connection relationship; analyze the voltage influence degree between each node to generate a node influence matrix; identify the nodes that have a significant impact on voltage over-limit from the node influence matrix, and output a key node set.

[0090] Step S23: Read the time series pattern set and the node influence matrix, extract the feature information of voltage over-limit to generate an over-limit feature vector; analyze the correlation features of voltage over-limit in the time and space dimensions to generate an over-limit correlation matrix; integrate the over-limit feature vector and the over-limit correlation matrix, and output a voltage over-limit feature set.

[0091] By adopting a method based on time-window segmentation and voltage similarity calculation, combined with the topological information of the distribution network, the accurate extraction of voltage over-limit characteristics is achieved. Through the sliding time-window technology, the dynamic characteristics of voltage changes can be captured, rather than simple static analysis. The construction of the time-series similarity matrix takes into account the voltage change correlation between monitoring points, enabling the system to identify regions with similar change patterns. At the same time, by introducing the network connection matrix and node influence matrix, the physical topological structure and voltage characteristics are organically combined, and the mutual influence relationship between nodes can be accurately reflected. This analysis method combining time-series characteristics and topological characteristics can not only identify the time characteristics of voltage over-limit, but also analyze the propagation law of over-limit events in space. By extracting the key node set, the system can identify the nodes that have the most significant impact on voltage over-limit, providing an important basis for subsequent early warning and control. This multi-dimensional feature extraction method significantly improves the accuracy and interpretability of over-limit feature recognition.

[0092] According to one aspect of the present application, step S3 is specifically as follows:

[0093] Step S31: Read the photovoltaic data in the standardized multi-source time-series dataset, analyze its periodic change characteristics, and generate a periodic feature set; extract the fluctuation information of photovoltaic output, and generate a fluctuation feature set; analyze the random change characteristics of photovoltaic output, and output a random feature set.

[0094] Step S32: Read the photovoltaic output data and voltage change data, establish the corresponding relationship between the two, and generate a coupling relationship matrix; calculate the influence degree of photovoltaic output change on voltage, and form a sensitivity vector; analyze the influence differences in different time periods, and output a time-period influence matrix.

[0095] Step S33: Read the coupling relationship matrix and the sensitivity vector, construct an influence evaluation model; quantitatively analyze the influence degree of photovoltaic power generation on voltage, and generate an influence evaluation matrix; integrate all evaluation results, and output a photovoltaic influence feature set.

[0096] Through multi-dimensional photovoltaic feature analysis and coupling relationship modeling, the precise quantification of the impact of photovoltaic output on voltage is achieved. The system analyzes photovoltaic features from three dimensions: periodicity, volatility, and randomness, constructs a complete combination of photovoltaic features, and comprehensively depicts the variation characteristics of photovoltaic output. By establishing a coupling relationship matrix between photovoltaic output and voltage change, the system can accurately reflect the impact degree on voltage under different photovoltaic output levels. The introduction of the sensitivity vector enables the system to quantify the differences in photovoltaic impacts at different times and positions, providing a scientific basis for photovoltaic management. This method based on multi-dimensional feature analysis and coupling relationship modeling significantly improves the accuracy and timeliness of photovoltaic impact assessment. Especially when considering the impact differences at different times, it can more accurately reflect the impact of the dynamic characteristics of photovoltaic power generation on the power grid, providing reliable technical support for the assessment of photovoltaic access capacity.

[0097] According to one aspect of the present application, step S4 is specifically as follows:

[0098] Step S41: Read the voltage over-limit feature set, construct a multi-dimensional early warning index system, and generate an early warning index set; read the photovoltaic impact feature set, calculate the dynamic early warning threshold, and form a threshold matrix; determine the division criteria of the early warning level, and output the early warning level set.

[0099] Step S42: Read the standardized multi-source time series data set, the early warning index set, and the threshold matrix, and construct a deep learning early warning model; optimize the model training process in combination with the data quality assessment matrix, and generate a model parameter set.

[0100] Step S43: Read the real-time monitoring data and the model evaluation index, and analyze using the trained early warning model; correct the credibility of the early warning result based on the model evaluation index, calculate the risk degree of voltage over-limit, and generate a risk assessment vector; according to the risk assessment result, output the over-limit early warning signal.

[0101] By establishing a multi-dimensional early warning index system and a deep learning early warning model, the precise early warning of voltage over-limit is achieved. The construction of the early warning index system takes into account the voltage over-limit feature and the photovoltaic impact feature, and improves the adaptability of the early warning through dynamic threshold design. The introduction of the deep learning early warning model enables the system to adaptively learn the complex features of voltage over-limit. Especially after optimizing the model training process in combination with the data quality assessment matrix, the robustness and accuracy of the model are significantly improved. The early warning result credibility correction mechanism based on the model evaluation index ensures the reliability of the early warning signal. This early warning method combining dynamic threshold and deep learning can not only accurately identify potential over-limit risks, but also give a quantitative assessment of the risk degree, providing timely and accurate early warning information for operation and maintenance personnel.

[0102] According to one aspect of the present application, step S5 is specifically as follows:

[0103] Step S51: Read the distribution network parameter information, establish a power flow analysis model, and generate a network model; calculate the theoretical maximum carrying capacity considering the voltage margin constraint to form a theoretical capacity matrix; analyze the network structure characteristics and output a network feature set.

[0104] Step S52: Read the photovoltaic impact feature set, analyze the impact of output fluctuations, and generate a dynamic impact factor matrix; read the over-limit warning signals, analyze the historical over-limit situations, and form a historical over-limit feature set; analyze the impact of load characteristics and output a load impact matrix.

[0105] Step S53: Read the theoretical capacity matrix, dynamic impact factor matrix, historical over-limit feature set, network feature set, and data quality assessment matrix, construct a multi-dimensional evaluation index system; evaluate the rationality of the existing photovoltaic access capacity to generate a capacity evaluation result; analyze the system optimization space based on the capacity evaluation result to generate an optimization recommendation set; integrate the capacity evaluation result and the optimization recommendation set and output a photovoltaic bearing capacity evaluation report.

[0106] By combining theoretical power flow analysis and dynamic impact assessment, an accurate assessment of the photovoltaic bearing capacity is achieved. The system first establishes a theoretical capacity benchmark considering the voltage margin constraint through power flow analysis, and then generates a dynamic impact factor matrix by analyzing the impact of photovoltaic output fluctuations, making the evaluation result more in line with the actual operating conditions. The establishment of the multi-dimensional evaluation index system combines theoretical analysis and actual operating data, making the evaluation result more comprehensive and reliable. Especially when considering the data quality assessment matrix, the system can quantify the credibility of the evaluation result and provide more reliable optimization recommendations. This evaluation method that comprehensively considers static capacity and dynamic characteristics can not only accurately evaluate the rationality of the existing photovoltaic access capacity, but also provide specific technical solutions for system optimization.

[0107] According to one aspect of the present application, step S11 is specifically:

[0108] Step S111: Read the original voltage monitoring data, extract the timestamp information to generate a time index set; detect the continuity of the timestamps, mark the time interval abnormal points, and generate a time anomaly mark set; perform time-dimensional preprocessing on the original voltage monitoring data based on the time anomaly mark set to generate a preprocessed voltage data set.

[0109] Step S112: Read the photovoltaic power generation data, match its timestamp with the time index set to generate a photovoltaic time mapping table; extract the installed capacity and actual power generation information, and reconstruct the data sequence in combination with the photovoltaic time mapping table to generate a reconstructed photovoltaic data set.

[0110] Step S113: Read meteorological data, establish a time correspondence relationship using the time index set, and generate a meteorological time mapping table; extract irradiance, temperature, and cloud cover information, and reconstruct the data sequence in combination with the meteorological time mapping table to generate a reconstructed meteorological data set.

[0111] Step S114: Read load data, establish a time correspondence relationship using the time index set, and generate a load time mapping table; extract active power and reactive power information, and reconstruct the data sequence in combination with the load time mapping table to generate a reconstructed load data set.

[0112] Step S115: Read the preprocessed voltage data set, the reconstructed photovoltaic data set, the reconstructed meteorological data set, and the reconstructed load data set, construct a data alignment matrix with a unified time scale, and generate an initial alignment matrix;

[0113] Among them, the alignment matrix AM(t) = W(t)·[D1(t), D2(t),..., Dn(t)]; where: Di(t) = [di(t-k),..., di(t),..., di(t+k)] is the time window sequence of the i-th data source; W(t) = diag[w1(t), w2(t),..., wn(t)] is the data source weight matrix; wi(t) = exp(-|fi-f0| / f0), wi(t) is the dynamic weight of the i-th data source, fi is the sampling frequency of the data source, f0 is the standard sampling frequency; k is the half-width of the time window; n is the number of data sources; the aligned data value vi(t) = ∑(wj·dj(t)) / ∑wj, j belongs to the set of valid data points within the time window.

[0114] Detect the data alignment quality in the initial alignment matrix to generate an alignment quality index set; optimize the initial alignment matrix based on the alignment quality index set and output a time-aligned data set.

[0115] Among them, the process of detecting data alignment quality is specifically: the alignment quality index AQ(t) = β1·TC(t) + β2·DC(t) + β3·MC(t); where: TC(t) = 1 - max(|ti - t|) / Δt, TC(t) is the time consistency index, ti is the time stamp of the i-th data source, t is the standard time stamp, Δt is the sampling period; DC(t) = 1 - σ(t) / σmax, DC(t) is the data dispersion index, σ(t) is the standard deviation at the alignment moment, σmax is the maximum allowable standard deviation; MC(t) = min(ni(t)) / N, MC(t) is the data completeness index, ni(t) is the number of valid data of the i-th data source at time t, N is the total number of data sources; β1, β2, β3 are weight coefficients and satisfy β1 + β2 + β3 = 1; the value range of AQ(t) is [0,1], and when AQ(t) is less than the preset threshold, it is marked as an abnormal alignment point.

[0116] Through refined time - dimension data processing and multi - source data alignment mechanism, high - quality data foundation construction is achieved. Specifically, first, by establishing a time index set and a time anomaly marker set, the system can accurately identify and process timestamp anomalies, ensuring the time continuity of data. When processing photovoltaic power generation data, through the construction of a photovoltaic time mapping table, not only time alignment is achieved, but also the corresponding relationship between installed capacity and actual power generation is retained, which is crucial for subsequent analysis of photovoltaic output characteristics. For meteorological data, the system reconstructs irradiance, temperature, and cloud cover information through a meteorological time mapping table, ensuring the time consistency of meteorological data with other data and providing accurate meteorological input for photovoltaic power generation prediction. During the reconstruction of load data, active power and reactive power information are uniformly processed through a load time mapping table, making the analysis of load characteristics more accurate. Finally, by constructing a data alignment matrix with a unified time scale and introducing a set of alignment quality indicators for optimization, not only the problem of inconsistent time scales of multi - source data is solved, but also the reliability of data alignment is ensured through quality assessment. This multi - level data alignment processing mechanism establishes a high - quality data foundation for subsequent feature extraction and analysis, significantly improving the credibility of analysis results.

[0117] According to one aspect of the present application, step S13 is specifically as follows:

[0118] Step S131: Read the voltage data in the cleaned dataset, calculate the maximum value, minimum value, and mean value to generate a set of basic statistics; use the set of basic statistics to calculate the standard deviation and variance to generate a set of fluctuation statistics; read the voltage data in the cleaned dataset, calculate the difference sequence and change rate of the data to generate a set of change characteristics; read the voltage data in the cleaned dataset, count the distribution parameters and quantiles of the data to generate a set of distribution characteristics; combine the set of basic statistics, the set of fluctuation statistics, the set of change characteristics, and the set of distribution characteristics, and output a set of voltage characteristics.

[0119] In another embodiment of the present application, the voltage feature vector VF(t) = [VM(t), VD(t), VR(t), VP(t)]; where: VM(t) = max(|v(τ)-vn|), τ ∈ [t - T, t], VM(t) is the maximum deviation feature; VD(t) = ∑|v(τ + 1)-v(τ)| / (T - 1), VD(t) is the volatility feature; VR(t) = (vmax - vmin) / vn, VR(t) is the fluctuation range feature; VP(t) = FFT(v(τ)), VP(t) is the spectrum feature vector; v(τ) is the voltage sampling value; vn is the rated voltage; T is the feature extraction time window; FFT is the fast Fourier transform.

[0120] Step S132: Read the photovoltaic data in the cleaned dataset, calculate the power difference sequence at adjacent moments to generate a power change sequence; perform statistical analysis on the power change sequence to generate a change statistic set; read the photovoltaic data in the cleaned dataset, calculate the fluctuation components at different time scales to generate a multi-scale fluctuation set; read the multi-scale fluctuation set, calculate the fluctuation amplitude and frequency characteristics to generate a fluctuation feature set; combine the change statistic set, the multi-scale fluctuation set, and the fluctuation feature set, and output the photovoltaic feature set.

[0121] Step S133: Read the meteorological data in the cleaned dataset, extract irradiance, temperature, and cloud cover data to generate a meteorological element set; use the temperature data in the meteorological element set to calculate the temperature coefficient of the photovoltaic panel to generate a temperature coefficient set; read the cloud cover data in the meteorological element set to calculate the irradiance attenuation coefficient to generate an attenuation coefficient set; read the meteorological element set, the temperature coefficient set, and the attenuation coefficient set, and combine with the standard parameters of the photovoltaic panel to calculate the predicted power generation value, and output the photovoltaic prediction set.

[0122] Step S134: Read the voltage feature set, the photovoltaic feature set, and the photovoltaic prediction set, calculate the correlation coefficients between the features to generate a correlation coefficient matrix; use the correlation coefficient matrix to screen features to generate a feature screening set; read the feature screening set, construct feature combination items to generate a combined feature set; integrate the feature screening set and the combined feature set, and output the fused feature set.

[0123] Through a multi-dimensional feature extraction and fusion mechanism, a comprehensive characterization of voltage and photovoltaic features is achieved. In terms of voltage feature extraction, through the combination of the basic statistic set, the fluctuation statistic set, the change feature set, and the distribution feature set, not only the static distribution characteristics of the voltage are reflected, but also the dynamic change characteristics are captured. In particular, the calculation of the difference sequence and the change rate enables the system to accurately identify the rapid change characteristics of the voltage. In terms of photovoltaic feature extraction, the system comprehensively characterizes the dynamic characteristics of photovoltaic output through the power change sequence and multi-scale fluctuation analysis. Especially the analysis of the fluctuation characteristics at different time scales provides an important basis for evaluating the impact of photovoltaic on voltage. Through the comprehensive analysis of meteorological elements, the system establishes a photovoltaic power generation prediction model including the temperature coefficient and the irradiance attenuation coefficient, improving the prediction accuracy. Finally, through the correlation coefficient matrix and the feature screening mechanism, an optimal combination of features is achieved, ensuring the effectiveness and representativeness of the fused features. This multi-level feature extraction and fusion method not only improves the expression ability of the features, but also reduces data redundancy, providing a reliable feature basis for subsequent early warning analysis.

[0124] According to one aspect of the present application, step S21 is specifically as follows:

[0125] Step S211: Read the standardized multi-source time series dataset, calculate the time interval features of the data to generate a time feature sequence; determine the optimal window length using the time feature sequence to generate a window parameter set; segment the data according to the window parameter set to generate an initial segmented sequence; optimize the boundaries of the initial segmented sequence and output a time window sequence.

[0126] Step S212: Read the voltage data in the time window sequence, calculate the numerical distribution features of each monitoring point to generate a distribution feature matrix; calculate the Euclidean distance between the monitoring points using the distribution feature matrix to generate a distance matrix; read the time window sequence, calculate the dynamic time warping distance between the monitoring points to generate a dynamic distance matrix; perform weighted fusion on the distance matrix and the dynamic distance matrix and output a time series similarity matrix.

[0127] In another embodiment of the present application, the calculation method of similarity can be: similarity index S(i,j) = γ1·ST(i,j) + γ2·SP(i,j) + γ3·SF(i,j); where: ST(i,j) = 1 - DTW(Vi,Vj) / L, ST(i,j) is the time domain similarity, DTW is the dynamic time warping distance, L is the sequence length; SP(i,j) = cos<Pi,Pj>, SP(i,j) is the pattern similarity, Pi and Pj are voltage feature vectors; SF(i,j) = exp(-|Fi - Fj| / σf), SF(i,j) is the frequency domain similarity, Fi and Fj are spectral features; γ1, γ2, and γ3 are adaptive weight coefficients, dynamically adjusted by pattern importance.

[0128] Step S213: Read the time window sequence and the time series similarity matrix, identify typical voltage change segments to generate a typical sequence set; perform clustering analysis on the typical sequence set to generate a clustering center set; read the typical sequence set and the clustering center set, extract pattern feature parameters to generate a pattern parameter set; perform pattern matching on all sequences according to the pattern parameter set and output a time series pattern set.

[0129] Through adaptive time window partitioning and multi-dimensional similarity calculation, the accurate recognition of voltage time series patterns is achieved. First, the system determines the optimal window length through time feature sequence analysis, enabling the window partitioning to capture the complete voltage change features while maintaining computational efficiency. Through boundary optimization processing, the problem of feature truncation that may be caused by traditional fixed windows is solved. In terms of similarity calculation, the Euclidean distance based on distribution features and the distance metric based on dynamic time warping are innovatively combined, considering both the similarity of numerical distributions and the dynamic characteristics of time series changes. In particular, the introduction of the dynamic time warping distance enables the system to handle time series patterns with different lengths and phase differences. Through the weighted fusion mechanism, the system can adjust the weights of different distance metrics according to actual needs, improving the flexibility and accuracy of similarity calculation. In the pattern recognition stage, through the extraction of typical sequences and clustering analysis, the system can automatically identify and summarize the main voltage change patterns, providing a reliable pattern basis for subsequent early warning analysis. This analysis method combining adaptive windows and multi-dimensional similarity significantly improves the accuracy and efficiency of voltage time series pattern recognition.

[0130] According to one aspect of the present application, step S23 is specifically as follows:

[0131] Step S231: Read the time series pattern set, count the over-limit amplitude information of each monitoring point, and generate an over-limit amplitude sequence; read the node influence matrix, extract the influence weights between nodes, and generate a node weight vector; combine the over-limit amplitude sequence and the node weight vector for feature calculation to generate a basic feature matrix; perform dimensionality reduction processing on the basic feature matrix and output an over-limit feature vector.

[0132] Step S232: Read the time series pattern set, extract the timestamp information of over-limit events, and generate a time mark sequence; read the node influence matrix, construct the node connection relationship, and generate a connection relationship graph; calculate the propagation time series based on the time mark sequence and the connection relationship graph to generate a propagation sequence set; perform statistical analysis on the propagation sequence set and output an over-limit correlation matrix.

[0133] The calculation process of propagation feature extraction can be: propagation feature vector PF(t) = [PS(t), PV(t), PR(t)]; where: PS(t) = ∑(di,j·δi,j(t)) / N, PS(t) is the spatial propagation feature, di,j is the electrical distance between nodes, and δi,j(t) is the over-limit state indicator function; PV(t) = ∑(vi(t) - vi(t - 1))·μi / Δt, PV(t) is the propagation speed feature, μi is the node weight; PR(t) = Nv(t) / N, PR(t) is the propagation range feature, Nv(t) is the number of over-limit nodes, and N is the total number of nodes.

[0134] Step S233: Read the out-of-limit feature vectors, calculate the combined relationships between the features, and generate a feature combination set; read the out-of-limit correlation matrix, extract the key correlation patterns, and generate a correlation pattern set; perform feature fusion on the feature combination set and the correlation pattern set to generate a fused feature set; perform standardization processing on the fused feature set, and output the voltage out-of-limit feature set.

[0135] Through a multi-level feature extraction and correlation analysis mechanism, the accurate characterization of voltage out-of-limit features and the mining of correlation laws are realized. Specifically, the system first constructs an out-of-limit amplitude analysis method considering node importance by combining the time series pattern set and the node weight vector. This weight-based feature extraction method enables the system to highlight the out-of-limit features of important nodes and improves the representativeness of the features. In terms of propagation feature analysis, by combining the time-tagged sequence and the connection relationship graph, the system can track the propagation path and time series relationship of out-of-limit events in the network. In particular, the construction of the propagation sequence set not only reflects the spatial spread law of out-of-limit events but also depicts the propagation speed and influence range, which is of great significance for predicting the development trend of out-of-limit events. In the feature fusion stage, the system establishes a multi-dimensional out-of-limit feature expression through the integration of the feature combination set and the correlation pattern set, which contains both the information of individual features and the correlation relationships between features. Through standardization processing, the comparability of features in different dimensions is ensured. This method based on multi-level feature extraction and correlation analysis significantly improves the accuracy and integrity of voltage out-of-limit feature recognition and provides comprehensive feature support for subsequent early warning analysis.

[0136] According to one aspect of the present application, step S32 is specifically as follows:

[0137] Step S321: Read the photovoltaic output data, calculate the output change rate, and generate an output change sequence; read the voltage change data, calculate the voltage deviation value, and generate a voltage deviation sequence; align the output change sequence and the voltage deviation sequence in time to generate a time-aligned matrix; perform data matching on the time-aligned matrix, and output a coupling relationship matrix.

[0138] In another embodiment of the present application, the construction process of the photovoltaic-voltage coupling matrix can be: coupling matrix CM(t) = λ1·TM(t) + λ2·SM(t) + λ3·FM(t); where: TM(t) is the time-domain coupling matrix, vi is the node voltage, and pj is the photovoltaic output; SM(t) = [ρij·σi / σj], SM(t) is the statistical coupling matrix, ρij is the correlation coefficient, and σi, σj are the standard deviations; FM(t) = [Hij(ω)], FM(t) is the frequency-domain coupling matrix, and Hij(ω) is the transfer function; λ1, λ2, λ3 are dynamic weight coefficients.

[0139] Step S322: Read the coupling relationship matrix, extract key change points, and generate a key point sequence; read the coupling relationship matrix, calculate the change trend features, and generate a trend feature set; combine the key point sequence and the trend feature set to calculate the response characteristics, and generate a response feature matrix; perform statistical analysis on the response feature matrix and output a sensitivity vector.

[0140] Among them, the calculation formula for sensitivity can be: SV(i) = η1·ST(i) + η2·SD(i) + η3·SF(i); where: ST(i) is the static sensitivity; SD(i) is the dynamic sensitivity, and wp(t) is the photovoltaic output weight; SF(i) = ∫|H(ω)|·G(ω)dω, SF(i) is the frequency sensitivity, H(ω) is the frequency response function, G(ω) is the photovoltaic fluctuation spectrum; η1, η2, and η3 are characteristic weight coefficients.

[0141] Step S323: Read the coupling relationship matrix, divide the data by time period, and generate a time-sharing data set; extract features for each time period of the time-sharing data set to generate a time period feature set; read the sensitivity vector, calculate the sensitivity distribution of each time period, and generate a time period distribution matrix; fuse the time period feature set and the time period distribution matrix and output a time period impact matrix.

[0142] Through the refined coupling relationship analysis and sensitivity calculation mechanism, the accurate quantification of the impact of photovoltaic output on voltage is realized. First, the system establishes a direct correspondence between the change of photovoltaic output and voltage response through the time alignment analysis of the output change sequence and voltage deviation sequence. This time alignment-based analysis method can accurately capture the immediate impact of photovoltaic output change on voltage. In terms of sensitivity analysis, by extracting the key point sequence and trend feature set, the system not only identifies the critical moments with significant impacts but also depicts the persistence and cumulative effect of the impacts through trend features. The construction of the response feature matrix comprehensively considers multiple dimensions such as the amplitude, speed, and duration of the impacts, making the sensitivity assessment more comprehensive. Especially in the time period impact analysis, the system accurately depicts the impact characteristics of different time periods through the feature extraction of the time-sharing data set and the construction of the time period distribution matrix, which has important guiding significance for formulating differentiated control strategies. This method combining immediate response analysis and long-term impact assessment not only improves the accuracy of photovoltaic impact assessment but also provides a reliable technical basis for the optimization of photovoltaic access capacity.

[0143] According to one aspect of the present application, step S42 is specifically:

[0144] Step S421: Read the standardized multi-source time-series data set, divide the training set and the validation set in chronological order to generate a data set division matrix; read the early warning index set, construct sample labels to generate a label sequence; read the data quality evaluation matrix, calculate sample weights to generate a sample weight vector; combine the data set division matrix, the label sequence and the sample weight vector, and output the training data set.

[0145] Step S422: Read the training data set and the threshold matrix, construct the input layer structure to generate an input feature matrix; design the network hierarchy according to the dimension of the input feature matrix to generate network structure parameters; read the sample weight vector, configure the loss function parameters to generate an optimization parameter set; integrate the network structure parameters and the optimization parameter set, and output the initial model parameters.

[0146] Step S423: Read the training data set and the initial model parameters, perform model training to generate a training process matrix; read the training process matrix, adjust the learning parameters to generate a parameter adjustment sequence; update the model parameters according to the parameter adjustment sequence to generate an updated parameter set; perform an optimization selection on the updated parameter set, and output the model parameter set.

[0147] Step S424: Read the training data set and the model parameter set, calculate the prediction results to generate a prediction result sequence; compare the prediction result sequence with the label sequence to generate an error statistics matrix; read the error statistics matrix, calculate the evaluation indicators to generate an indicator calculation result; perform a summary analysis on the indicator calculation result, and output the model evaluation indicators.

[0148] Among them, the model parameter optimization method can be: the optimization objective function \(L(θ)=\sum(α_i·LE(i)+β_i·LP(i)+μ_i·LR(i))\); where: \(LE(i)=|y_i - f(x_i,θ)|^2\), \(LE(i)\) is the prediction error loss; \(LP(i)=||θ||^2\), \(LP(i)\) is the parameter regularization term; \(LR(i)=max(0,ε - y_i·f(x_i,θ))\), \(LR(i)\) is the robustness loss; \(α_i\), \(β_i\), \(μ_i\) are sample weights; \(θ\) is the model parameter; \(f(x,θ)\) is the early warning model function.

[0149] Through the optimized design of the deep learning model and the quality control of the training process, high-precision voltage over-limit warning is achieved. First, the system ensures the temporal integrity and representativeness of the training data by constructing a dataset division matrix. The introduction of the sample weight vector enables the system to dynamically adjust the importance of different samples according to data quality, improving the reliability of model training. In terms of network structure design, the system adaptively determines the network hierarchy through the dimensional analysis of the input feature matrix, avoiding the problems of overly complex or simple network structures. Especially in the configuration of the loss function, through the integration of sample weights, key learning of high-quality samples is achieved. During the model training process, the system dynamically adjusts the learning parameters through the analysis of the training process matrix, effectively preventing overfitting and underfitting problems. The introduction of model evaluation metrics not only provides a quantitative evaluation of model performance but also guides the optimization direction of the model through the analysis of the error statistics matrix. This training method combining quality control and adaptive optimization significantly improves the accuracy and generalization ability of the warning model.

[0150] According to one aspect of the present application, step S52 is specifically as follows:

[0151] Step S521: Read the photovoltaic impact feature set, extract the fluctuation feature sequence, and generate a fluctuation sequence matrix; perform frequency decomposition on the fluctuation sequence matrix to generate a frequency component set; read the frequency component set, calculate the impact weights of each frequency component, and generate a frequency weight vector; perform weighted combination on the frequency component set and the frequency weight vector, and output a dynamic impact factor matrix.

[0152] In another embodiment of the present application, the calculation formula of the dynamic impact factor can be: IF(t) = κ1·WF(t) + κ2·CF(t) + κ3·RF(t); where: WF(t) = ∑|p(τ + 1) - p(τ)|·w(τ) / T, WF(t) is the fluctuation intensity factor, p(τ) is the photovoltaic output, w(τ) is the time weight; CF(t) = ∑(vi(t) - vi,ref)2 / N, CF(t) is the voltage deviation factor, vi,ref is the reference voltage; RF(t) = 1 - min(Pi,actual / Pi,limit), RF(t) is the risk factor, Pi,actual is the actual power, Pi,limit is the limit value; κ1, κ2, κ3 are comprehensive weight coefficients.

[0153] Step S522: Read the over-limit warning signal, count the time distribution of over-limit events, and generate a time distribution sequence; perform pattern recognition on the time distribution sequence to generate a pattern feature set; read the pattern feature set, extract key feature parameters, and generate a feature parameter matrix; perform feature integration on the feature parameter matrix, and output a historical over-limit feature set.

[0154] Step S523: Read the historical over-limit feature set, extract the load change information, and generate a load change sequence; perform a correlation analysis on the load change sequence to generate a correlation coefficient matrix; read the correlation coefficient matrix, calculate the influence degree index, and generate an influence index set; convert the influence index set into an influence feature matrix, and output the load influence matrix.

[0155] Through multi-dimensional photovoltaic fluctuation analysis and historical data mining, the accurate construction of photovoltaic impact factors is achieved. The system first identifies the influence characteristics of different frequency components through the frequency decomposition of the fluctuation sequence matrix. This frequency-domain-based analysis method can accurately characterize the periodic and random characteristics of photovoltaic output fluctuations. Through the design of the frequency weight vector, the system realizes the differential evaluation of the influence of different frequency components. In terms of historical over-limit analysis, the system extracts the typical characteristics and occurrence rules of over-limit events through the pattern recognition of the time distribution sequence. The construction of the characteristic parameter matrix comprehensively considers multiple dimensions such as the frequency, degree, and duration of over-limit events. In the load impact analysis, through correlation analysis and influence degree quantification, the system accurately evaluates the regulatory effect of load changes on photovoltaic impact. This method combining fluctuation characteristic analysis and historical data mining not only improves the accuracy of impact factors but also enhances the interpretability of evaluation results.

[0156] According to one aspect of the present application, step S53 is specifically as follows:

[0157] Step S531: Read the theoretical capacity matrix and the network feature set, extract the capacity constraint conditions, and generate a constraint condition set; read the dynamic influence factor matrix, calculate the dynamic adjustment coefficient, and generate an adjustment coefficient matrix; read the historical over-limit feature set, extract the over-limit influence parameters, and generate an influence parameter set; perform index conversion on the constraint condition set, the adjustment coefficient matrix, and the influence parameter set, and output the evaluation index system.

[0158] Step S532: Read the evaluation index system, calculate the weight coefficients of each index, and generate a weight coefficient vector; read the theoretical capacity matrix, perform capacity analysis and calculation, and generate a capacity analysis matrix; perform weighted calculation on the capacity analysis matrix according to the weight coefficient vector to generate a weighted result set; perform a threshold judgment on the weighted result set, and output the capacity evaluation result.

[0159] Step S533: Read the capacity evaluation result and the evaluation index system, analyze the index differences, and generate a difference feature set; perform clustering analysis on the difference feature set to generate a clustering result matrix; read the clustering result matrix, identify the optimization direction, and generate an optimization direction set; convert the optimization direction set into specific suggestions, and output the optimization suggestion set.

[0160] Step S534: Read the capacity evaluation results, extract the key evaluation information, and generate an evaluation summary set; read the optimization suggestion set, organize the optimization measures, and generate a measure list set; format the evaluation summary set and the measure list set to generate a report structure set; fill in the content of the report structure set and output the photovoltaic carrying capacity evaluation report.

[0161] In another embodiment of the present application, the bearing capacity index can be set as: CC = min{CC1, CC2, CC3}; where: CC1 = Pmax·(1 - ∑ωi·Vi / Vmax), CC1 is the voltage constraint capacity, Vi is the node voltage deviation, Vmax is the maximum allowable deviation; CC2 = Sn·(1 - max(Sij / Sij,max)), CC2 is the equipment constraint capacity, Sij is the line power, Sij,max is the rated capacity; CC3 = Pb·(1 - ∑φi·IFi / IFmax), CC3 is the fluctuation constraint capacity, IFi is the influence factor, IFmax is the allowable maximum influence; ωi, φi are the node weight coefficients; Pmax, Sn, Pb are the reference capacities.

[0162] In summary, through the construction of a comprehensive evaluation system and multi-dimensional optimization analysis, the scientific evaluation of photovoltaic carrying capacity is realized. The system first establishes a capacity constraint framework considering network characteristics by extracting the constraint condition set. The introduction of the dynamic adjustment coefficient matrix enables the system to dynamically adjust the evaluation criteria according to the real-time operating state. In the construction of the evaluation index system, the system realizes the differential evaluation of different indexes through the design of the weight coefficient vector. Especially in the process of capacity analysis, the system can comprehensively consider the constraint conditions of multiple dimensions through the construction of the weighted result set. In terms of optimization analysis, the system identifies the key directions for system optimization through the difference feature set and cluster analysis. In the process of generating optimization suggestions, the system not only considers the technical feasibility but also provides a specific implementation plan in the form of a measure list set. The automatic generation mechanism of the evaluation report ensures the standardization and readability of the evaluation results. This method combining multi-dimensional evaluation and optimization analysis significantly improves the scientificity and practicality of the bearing capacity evaluation.

[0163] Embodiment 2, the data processing flow can also be briefly described as follows:

[0164] S1: Multi-source data collection and preprocessing

[0165] S11: Data collection and time alignment

[0166] Read the original voltage monitoring data of the distribution transformer area (including timestamp, measurement point ID, voltage value) to generate the initial voltage dataset V1; read the photovoltaic power generation data (including timestamp, installed capacity, actual power generation) to generate the initial photovoltaic dataset P1; read the meteorological data (including timestamp, irradiance, temperature, cloud cover) to generate the initial meteorological dataset W1; read the load data (including timestamp, active power, reactive power) to generate the initial load dataset L1; use the sliding time window method to align all data to a unified time scale to generate the time-aligned dataset D1;

[0167] S12: Data quality assessment and correction

[0168] Perform data integrity check on the time-aligned dataset D1, and calculate the missing rate index M1 of each type of data; when the missing rate index M1 is less than the threshold, use the tensor completion algorithm for data repair to generate the completed dataset D2; use the denoising method based on wavelet transform to process the outliers in the completed dataset D2 to generate the cleaned dataset D3; calculate the data credibility index to generate the data quality assessment matrix Q1;

[0169] S13: Feature extraction and data fusion

[0170] Extract statistical features from the voltage data in the cleaned dataset D3 to generate the voltage feature set V2; extract the fluctuation features from the photovoltaic data to generate the photovoltaic feature set P2; calculate the expected value of photovoltaic power generation in combination with the meteorological data to generate the photovoltaic prediction set P3; use the adaptive weight fusion algorithm to fuse the voltage feature set V2, the photovoltaic feature set P2, and the photovoltaic prediction set P3 to generate the fusion feature set F1;

[0171] S14: Data standardization and storage

[0172] Perform normalization processing on the fusion feature set F1 to generate the standardized feature set F2; construct a spatio-temporal correlation matrix to associate the standardized feature set F2 with the distribution network topology information to generate the associated feature set F3; finally output the standardized multi-source time series dataset D4;

[0173] S2: Voltage over-limit feature extraction and analysis

[0174] S21: Time series feature analysis

[0175] Perform time window segmentation on the standardized multi-source time series dataset D4 to generate the time window sequence T1; calculate the similarity of the voltage sequences at each monitoring point to generate the time series similarity matrix R1; extract the time series patterns of voltage changes to generate the time series pattern set P4;

[0176] S22: Voltage feature space analysis

[0177] Combined with the distribution network topology information, construct the network connection matrix G1; analyze the voltage influence relationship between nodes, generate the node influence matrix I1; identify the key monitoring nodes, and generate the key node set N1;

[0178] S23: Overlimit feature extraction

[0179] Based on the time series pattern set P4 and the node influence matrix I1, extract the voltage overlimit features, generate the overlimit feature vector V3; analyze the spatio-temporal correlation of overlimit occurrences, generate the overlimit correlation matrix M2; output the voltage overlimit feature set E2;

[0180] S3: Photovoltaic power generation influence feature analysis

[0181] S31: Photovoltaic output feature analysis

[0182] Conduct periodic analysis on the photovoltaic data in the standardized multi-source time series data set D4 to generate the periodic feature set C1;

[0183] Extract the fluctuation features of photovoltaic output to generate the fluctuation feature set W2;

[0184] Analyze the randomness features of photovoltaic output to generate the random feature set R2;

[0185] S32: Photovoltaic-voltage coupling analysis

[0186] Establish the mapping relationship between photovoltaic output and voltage change to generate the coupling relationship matrix K1; calculate the sensitivity index of photovoltaic output to voltage to generate the sensitivity vector S1; analyze the influence degree in different time periods to generate the time period influence matrix T2;

[0187] S33: Comprehensive influence assessment

[0188] Based on the coupling relationship matrix K1 and the sensitivity vector S1, establish an influence assessment model; quantify the influence degree of photovoltaic power generation on voltage to generate the influence assessment matrix I2; output the photovoltaic influence feature set P5;

[0189] S4: Voltage overlimit warning model construction

[0190] S41: Warning index construction

[0191] Based on the voltage overlimit feature set E2, construct a multi-dimensional warning index system to generate the warning index set W3; combined with the photovoltaic influence feature set P5, design a dynamic warning threshold to generate the threshold matrix T3; establish a warning level classification standard to generate the warning level set L2;

[0192] S42: Warning model training

[0193] Build a warning model based on deep learning and train it using the standardized multi-source time-series dataset D4; introduce an attention mechanism to optimize the model performance and generate the model parameter set P6; conduct model verification and optimization to generate the model evaluation index E3;

[0194] S43: Warning result generation

[0195] Conduct warning analysis on the real-time monitoring data D5; calculate the overlimit risk degree to generate the risk assessment vector R3; output the overlimit warning signal A1;

[0196] S5: Photovoltaic carrying capacity assessment

[0197] S51: Static carrying capacity benchmark calculation

[0198] Establish a power flow analysis model based on the distribution network parameters to generate the network model M3; consider the voltage margin constraint, calculate the theoretical maximum carrying capacity to generate the theoretical capacity matrix C2; analyze the network structure characteristics to generate the network feature set N2;

[0199] S52: Dynamic influencing factor analysis

[0200] Based on the photovoltaic influence feature set P5, analyze the influence of output power fluctuations to generate the dynamic influence factor matrix D6; evaluate the historical overlimit situation and generate the historical overlimit feature set H1 based on the overlimit warning signal A1; analyze the influence of load characteristics to generate the load influence matrix L3;

[0201] S53: Comprehensive carrying capacity assessment

[0202] Establish a multi-dimensional evaluation index system, including dimensions such as voltage quality, equipment capacity, and operation margin; construct a comprehensive evaluation model considering the theoretical capacity matrix C2, the dynamic influence factor matrix D6, and the historical overlimit feature set H1; evaluate the rationality of the existing photovoltaic access capacity to generate the capacity evaluation result E4; analyze the system's optimizable space to generate the optimization suggestion set O1; output the photovoltaic carrying capacity assessment report R4 (including the capacity evaluation result E4 and the optimization suggestion set O1).

[0203] The preferred embodiments of the present invention have been described in detail above. However, the present invention is not limited to the specific details in the above embodiments. Within the scope of the technical concept of the present invention, various equivalent transformations can be made to the technical solutions of the present invention, and these equivalent transformations all fall within the protection scope of the present invention.

Claims

1. Active area voltage over-limit warning and photovoltaic carrying capacity assessment method, characterized in that: The steps include: Step S1: Obtain the original voltage monitoring data, photovoltaic power generation data, meteorological data and load data of the distribution station area, and generate a time-aligned data set through time alignment processing; perform data quality assessment and correction on the time-aligned data set to generate a cleaned data set; Extract voltage and photovoltaic features from the cleaned data set and perform data fusion to form a fused feature set; The fused feature set is standardized and associated with the distribution network topology information to output a standardized multi-source time series data set; Step S2: read the standardized multi-source time series data set, obtain the time window sequence by time window segmentation processing, calculate the voltage similarity to obtain the time series similarity matrix; Construct a network connection matrix based on the distribution network topology information, analyze the node influence relationship and generate a node influence matrix; Extract voltage over-limit features based on the time series similarity matrix and node influence matrix, and output voltage over-limit feature set; Step S3: Read the photovoltaic data in the standardized multi-source time series data set, analyze its periodicity, volatility and randomness characteristics, and generate a photovoltaic feature combination; establish the correlation between photovoltaic output and voltage change, form a coupling relationship matrix and a sensitivity vector; construct an impact assessment model based on the coupling relationship matrix and the sensitivity vector, and output a photovoltaic impact feature set; Step S4: read the voltage over-limit feature set, construct an early warning indicator system to obtain an early warning indicator set, design an early warning threshold value in combination with the photovoltaic impact feature set to obtain a threshold matrix; train an early warning model based on a standardized multi-source time series data set to generate a model parameter set; perform early warning analysis on real-time monitoring data, and output an over-limit warning signal; Step S5: reading the distribution network parameters to perform power flow analysis and generate a theoretical capacity matrix; Analyze the impact of photovoltaic output fluctuations and obtain a dynamic impact factor matrix; A multi-dimensional evaluation index system is established to conduct carrying capacity assessment based on the theoretical capacity matrix and dynamic impact factor matrix, and a photovoltaic carrying capacity assessment report is output.

2. The method for voltage over-limit warning and photovoltaic carrying capacity assessment in active area according to claim 1 is characterized in that: Step S1 is specifically as follows: Step S11: receiving the original voltage monitoring data of the distribution station area, extracting the timestamp, measurement point identification and voltage value information therein; receiving photovoltaic power generation data, extracting the timestamp, installed capacity and actual power generation information therein; receiving meteorological data, extracting the timestamp, irradiance, temperature and cloud cover information therein; receiving load data, extracting the timestamp, active power and reactive power information therein; aligning the above data according to a unified time scale, and outputting a time-aligned data set; Step S12: reading the time-aligned data set, calculating the missing degree of each type of data, and obtaining a missing rate index; completing the data according to the missing rate index, and generating a completed data set; Process outliers on the completed data set and output the cleaned data set; Quantitatively calculate the credibility of the cleaned data set and generate a data quality assessment matrix; Step S13: reading the voltage data in the cleaned data set, extracting its statistical features, and generating a voltage feature set; processing the photovoltaic data in the cleaned data set, extracting its fluctuation features, and generating a photovoltaic feature set; calculating the expected value of photovoltaic power generation in combination with the meteorological data in the cleaned data set, and generating a photovoltaic prediction set; performing feature fusion on the voltage feature set, the photovoltaic feature set, and the photovoltaic prediction set, and outputting a fused feature set; Step S14: read the fused feature set, perform numerical normalization processing, and generate a standardized feature set; read the distribution network topology information, associate the standardized feature set with the topology information, and generate an associated feature set; integrate the standardized feature set, the associated feature set, and the data quality assessment matrix, and output a standardized multi-source time series data set.

3. The method for voltage over-limit warning and photovoltaic carrying capacity assessment in active area according to claim 1 is characterized in that: Step S2 is specifically as follows: Step S21: read the standardized multi-source time series data set, segment the data using a sliding window method, and generate a time window sequence; calculate the similarity between the voltage data sequences of each monitoring point to form a time series similarity matrix; Extract the timing characteristics of voltage changes from the timing similarity matrix and output a timing pattern set; Step S22: read the distribution network topology information, construct a network connection matrix that represents the node connection relationship; analyze the voltage influence degree between each node, and generate a node influence matrix; Identify nodes with significant impact on voltage over-limit from the node influence matrix and output the key node set; Step S23: reading the timing pattern set and the node influence matrix, extracting the characteristic information of voltage exceeding the limit, and generating an exceeding-limit characteristic vector; The correlation characteristics of voltage over-limit in time and space dimensions are analyzed to generate an over-limit correlation matrix; the over-limit feature vector and the over-limit correlation matrix are integrated to output the voltage over-limit feature set.

4. The method for voltage over-limit warning and photovoltaic carrying capacity assessment in active area according to claim 1 is characterized in that: Step S3 is specifically as follows: Step S31: reading photovoltaic data in the standardized multi-source time series data set, analyzing its periodic variation characteristics, and generating a periodic feature set; extracting the fluctuation information of photovoltaic output, generating a fluctuation feature set; analyzing the random variation characteristics of photovoltaic output, and outputting a random feature set; Step S32: reading photovoltaic output data and voltage change data, establishing a corresponding relationship between the two, and generating a coupling relationship matrix; Calculate the impact of photovoltaic output changes on voltage and form a sensitivity vector; analyze the impact differences in different time periods and output the time period impact matrix; Step S33: read the coupling relationship matrix and sensitivity vector, and construct an impact assessment model; quantitatively analyze the impact of photovoltaic power generation on voltage, and generate an impact assessment matrix; integrate all assessment results and output a photovoltaic impact feature set.

5. The method for voltage over-limit warning and photovoltaic carrying capacity assessment in active area according to claim 1 is characterized in that: Step S4 is specifically as follows: Step S41: reading the voltage over-limit feature set, constructing a multi-dimensional early warning indicator system, and generating an early warning indicator set; reading the photovoltaic impact feature set, calculating the dynamic early warning threshold, and forming a threshold matrix; Determine the classification criteria for warning levels and output the warning level set; Step S42: read the standardized multi-source time series data set, the warning indicator set and the threshold matrix, and construct a deep learning warning model; optimize the model training process in combination with the data quality assessment matrix, and generate a model parameter set; Step S43: read the real-time monitoring data and model evaluation indicators, and use the trained early warning model for analysis; Based on the model evaluation index, the credibility of the warning result is corrected, the voltage over-limit risk degree is calculated, and the risk assessment vector is generated; Based on the risk assessment results, an out-of-limit warning signal is output.

6. The method for voltage over-limit warning and photovoltaic carrying capacity assessment in active area according to claim 5 is characterized in that: Step S5 is specifically as follows: Step S51: reading the distribution network parameter information, establishing a power flow analysis model, and generating a network model; calculating the theoretical maximum carrying capacity considering the voltage margin constraint, and forming a theoretical capacity matrix; analyzing the network structure characteristics, and outputting a network feature set; Step S52: reading the photovoltaic impact feature set, analyzing the impact of output fluctuations, and generating a dynamic impact factor matrix; Read the over-limit warning signal, analyze the historical over-limit situation, and form a historical over-limit feature set; analyze the impact of load characteristics and output the load impact matrix; Step S53: reading the theoretical capacity matrix, the dynamic impact factor matrix, the historical over-limit feature set, the network feature set and the data quality assessment matrix, and constructing a multi-dimensional assessment index system; Evaluate the rationality of existing photovoltaic access capacity and generate capacity assessment results; analyze system optimization space based on capacity assessment results and generate a set of optimization suggestions; integrate capacity assessment results and optimization suggestions and output a photovoltaic carrying capacity assessment report.

7. The method for voltage over-limit warning and photovoltaic carrying capacity assessment in active area according to claim 2 is characterized in that: Step S11 is specifically as follows: Step S111: read the original voltage monitoring data, extract the timestamp information to generate a time index set; detect the continuity of the timestamp, mark the time interval abnormal points, and generate a time abnormality mark set; perform time dimension preprocessing on the original voltage monitoring data based on the time abnormality mark set to generate a preprocessed voltage data set; Step S112: Read photovoltaic power generation data, match its timestamp with the time index set, and generate a photovoltaic time mapping table; extract installed capacity and actual power generation information, reconstruct the data sequence in combination with the photovoltaic time mapping table, and generate a reconstructed photovoltaic data set; Step S113: Read meteorological data, establish time correspondence using the time index set, and generate a meteorological time mapping table; extract irradiance, temperature and cloud cover information, reconstruct the data sequence in combination with the meteorological time mapping table, and generate a reconstructed meteorological data set; Step S114: read the load data, establish a time correspondence using the time index set, and generate a load time mapping table; extract active power and reactive power information, reconstruct the data sequence in combination with the load time mapping table, and generate a reconstructed load data set; Step S115: reading the preprocessed voltage data set, the reconstructed photovoltaic data set, the reconstructed meteorological data set and the reconstructed load data set, constructing a data alignment matrix of a unified time scale, and generating an initial alignment matrix; Detecting the data alignment quality in the initial alignment matrix and generating a set of alignment quality indicators; Optimize the initial alignment matrix based on the alignment quality indicator set and output the time-aligned dataset; Step S13 is specifically as follows: Step S131: read the voltage data in the cleaned data set, calculate the maximum value, minimum value and mean value, and generate a basic statistical set; use the basic statistical set to calculate the standard deviation and variance to generate a fluctuation statistical set; read the voltage data in the cleaned data set, calculate the differential sequence and change rate of the data, and generate a change feature set; Read the voltage data in the cleaned data set, calculate the distribution parameters and quantiles of the statistics, and generate a distribution feature set; The basic statistics set, the fluctuation statistics set, the variation feature set and the distribution feature set are combined to output a voltage feature set; Step S132: reading the photovoltaic data in the cleaned data set, calculating the power difference sequence at adjacent moments, and generating a power change sequence; performing statistical analysis on the power change sequence, and generating a change statistics set; reading the photovoltaic data in the cleaned data set, calculating the fluctuation components at different time scales, and generating a multi-scale fluctuation set; Read the multi-scale fluctuation set, calculate the fluctuation amplitude and frequency characteristics, and generate the fluctuation characteristic quantity set; combine the variation statistics set, the multi-scale fluctuation set and the fluctuation characteristic quantity set, and output the photovoltaic characteristic set; Step S133: Read the meteorological data in the cleaned data set, extract the irradiance, temperature and cloud cover data, and generate a meteorological element set; use the temperature data in the meteorological element set to calculate the temperature coefficient of the photovoltaic panel and generate a temperature coefficient set; read the cloud cover data in the meteorological element set, calculate the irradiance attenuation coefficient, and generate an attenuation coefficient set; read the meteorological element set, the temperature coefficient set and the attenuation coefficient set, calculate the power generation prediction value in combination with the standard parameters of the photovoltaic panel, and output the photovoltaic prediction set; Step S134: reading the voltage feature set, the photovoltaic feature set and the photovoltaic prediction set, calculating the correlation coefficients between the features, and generating a correlation coefficient matrix; using the correlation coefficient matrix to screen features, and generating a feature screening set; Read the feature screening set, construct feature combination items, and generate a combined feature set; Integrate the feature screening set and the combined feature set to output the fused feature set.

8. The method for voltage over-limit warning and photovoltaic carrying capacity assessment in active area according to claim 3 is characterized in that: Step S21 is specifically as follows: Step S211: read the standardized multi-source time series data set, calculate the time interval characteristics of the data, and generate a time feature sequence; use the time feature sequence to determine the optimal window length and generate a window parameter set; segment the data according to the window parameter set to generate an initial segment sequence; optimize the boundaries of the initial segment sequence and output a time window sequence; Step S212: reading the voltage data in the time window sequence, calculating the numerical distribution characteristics of each monitoring point, and generating a distribution characteristic matrix; using the distribution characteristic matrix to calculate the Euclidean distance between the monitoring points, and generating a distance matrix; Read the time window sequence, calculate the dynamic time-warped distance between monitoring points, and generate a dynamic distance matrix; perform weighted fusion of the distance matrix and the dynamic distance matrix, and output a time series similarity matrix; Step S213: reading the time window sequence and the time series similarity matrix, identifying typical voltage change segments, and generating a typical sequence set; Perform cluster analysis on the typical sequence set to generate a cluster center set; read the typical sequence set and the cluster center set, extract the pattern feature parameters, and generate a pattern parameter set; perform pattern matching on all sequences according to the pattern parameter set, and output the time series pattern set; Step S23 is specifically as follows: Step S231: read the time series pattern set, count the over-limit amplitude information of each monitoring point, and generate an over-limit amplitude sequence; read the node influence matrix, extract the influence weights between nodes, and generate a node weight vector; combine the over-limit amplitude sequence and the node weight vector to perform feature calculation and generate a basic feature matrix; perform dimensionality reduction processing on the basic feature matrix and output the over-limit feature vector; Step S232: read the time series pattern set, extract the timestamp information of the limit crossing event, and generate a time mark sequence; read the node influence matrix, construct the node connection relationship, and generate a connection relationship graph; calculate the propagation time series based on the time mark sequence and the connection relationship graph, and generate a propagation sequence set; perform statistical analysis on the propagation sequence set, and output the limit crossing association matrix; Step S233: read the limit-crossing feature vector, calculate the combination relationship between the features, and generate a feature combination set; read the limit-crossing association matrix, extract the key association pattern, and generate an association pattern set; Perform feature fusion on the feature combination set and the associated pattern set to generate a fused feature set; The fused feature set is standardized to output the voltage-out-of-limit feature set.

9. The method for voltage over-limit warning and photovoltaic carrying capacity assessment in active area according to claim 4 is characterized in that: The step S32 is specifically as follows: Step S321: reading photovoltaic output data, calculating the output change rate, and generating an output change sequence; reading voltage change data, calculating the voltage deviation value, and generating a voltage deviation sequence; aligning the output change sequence and the voltage deviation sequence in time to generate a time alignment matrix; Perform data matching on the time alignment matrix and output the coupling relationship matrix; Step S322: reading the coupling relationship matrix, extracting key change points, and generating a key point sequence; Read the coupling relationship matrix, calculate the change trend characteristics, and generate a trend feature set; Calculate the response characteristics by combining the key point sequence and the trend feature set to generate a response feature matrix; Perform statistical analysis on the response feature matrix and output the sensitivity vector; Step S323: read the coupling relationship matrix, divide the data by time period, and generate a time-sharing data set; extract features for each time period of the time-sharing data set to generate a time period feature set; Read the sensitivity vector, calculate the sensitivity distribution of each time period, and generate the time period distribution matrix; The time period feature set and the time period distribution matrix are fused to output the time period impact matrix.

10. The method for voltage over-limit warning and photovoltaic carrying capacity assessment in active area according to claim 6, characterized in that: Step S42 is specifically as follows: Step S421: read the standardized multi-source time series data set, divide the training set and the validation set in chronological order, and generate a data set partition matrix; read the warning indicator set, construct sample labels, and generate a label sequence; read the data quality assessment matrix, calculate the sample weight, and generate a sample weight vector; Combine the data set partition matrix, label sequence and sample weight vector to output the training data set; Step S422: read the training data set and the threshold matrix, construct the input layer structure, and generate the input feature matrix; Design the network hierarchy according to the dimension of the input feature matrix and generate network structure parameters; Read the sample weight vector, configure the loss function parameters, generate the optimization parameter set; integrate the network structure parameters and the optimization parameter set, and output the initial model parameters; Step S423: reading the training data set and initial model parameters, performing model training, and generating a training process matrix; Read the training process matrix, adjust the learning parameters, and generate the parameter adjustment sequence; Update the model parameters according to the parameter adjustment sequence to generate an updated parameter set; Optimize and select the update parameter set and output the model parameter set; Step S424: read the training data set and the model parameter set, calculate the prediction results, and generate a prediction result sequence; compare the prediction result sequence with the label sequence to generate an error statistical matrix; read the error statistical matrix, calculate the evaluation index, and generate the index calculation result; summarize and analyze the index calculation results, and output the model evaluation index; Step S52 is specifically as follows: Step S521: read the photovoltaic impact feature set, extract the fluctuation feature sequence, and generate a fluctuation sequence matrix; perform frequency decomposition on the fluctuation sequence matrix to generate a frequency component set; read the frequency component set, calculate the influence weight of each frequency component, and generate a frequency weight vector; Perform weighted combination of the frequency component set and the frequency weight vector, and output a dynamic impact factor matrix; Step S522: read the limit-crossing warning signal, count the time distribution of the limit-crossing event, and generate a time distribution sequence; perform pattern recognition on the time distribution sequence to generate a pattern feature set; read the pattern feature set, extract key feature parameters, and generate a feature parameter matrix; Integrate the feature parameter matrix and output the historical limit-crossing feature set; Step S523: read the historical over-limit feature set, extract the load change information, and generate a load change sequence; perform correlation analysis on the load change sequence and generate a correlation coefficient matrix; Read the correlation coefficient matrix, calculate the impact index, and generate an impact index set; convert the impact index set into an impact feature matrix and output the load impact matrix; Step S53 is specifically as follows: Step S531: read the theoretical capacity matrix and the network feature set, extract the capacity constraint conditions, and generate the constraint condition set; read the dynamic impact factor matrix, calculate the dynamic adjustment coefficient, and generate the adjustment coefficient matrix; read the historical over-limit feature set, extract the over-limit impact parameters, and generate the impact parameter set; The constraint condition set, adjustment coefficient matrix and influencing parameter set are converted into indicators to output the evaluation indicator system; Step S532: read the evaluation index system, calculate the weight coefficient of each index, and generate a weight coefficient vector; read the theoretical capacity matrix, perform capacity analysis calculation, and generate a capacity analysis matrix; Perform weighted calculation on the capacity analysis matrix according to the weight coefficient vector to generate a weighted result set; Perform threshold judgment on the weighted result set and output the capacity assessment result; Step S533: reading the capacity assessment results and the assessment index system, analyzing the index differences, and generating a difference feature set; Perform cluster analysis on the difference feature set and generate a clustering result matrix; Read the clustering result matrix, identify the optimization direction, and generate an optimization direction set; convert the optimization direction set into specific suggestions and output the optimization suggestion set; Step S534: read the capacity assessment result, extract key assessment information, and generate an assessment summary set; read the optimization suggestion set, organize the optimization measures, and generate a measure list set; format the assessment summary set and the measure list set to generate a report structure set; Fill the report structure set with content and output the photovoltaic carrying capacity assessment report.

Images