An intelligent configuration method for power utilization information collection

Abstract

The application discloses a kind of electric information collection intelligent configuration method, it is related to power distribution automation technical field, including, convergence equipment account, topology, historical operation and configuration record are standardized, form data set and candidate template;Extract operation, topology, equipment and alarm features, build comprehensive feature vector;With comprehensive features and candidate template respectively by SVM, random forest and CNN calculate matching probability, form probability matrix;With odds ratio, consistency coefficient and historical success rate fusion calibration, get robust matching probability;Sensitivity detection is carried out to core operation index, together with matching probability determines optimal template, and according to three levels segmentation, in turn, under the parameter of roll-back is sent out.This application determines optimal template by the ingenious combination of the probability calculated by three models, ensures the reliability and accuracy of optimal template.

Description

Technical Field

[0001] This invention relates to the field of power distribution automation technology, and in particular to an intelligent configuration method for collecting electricity consumption information. Background Technology

[0002] Electricity information acquisition systems focus on parameter configuration and template selection for distribution terminals and master stations. This typically relies on equipment ledgers, distribution topology, historical operation records, and historical configuration records. After cleaning and alignment, standardized data is obtained, and features such as operating status, topology, equipment ledger data, historical configurations, and alarms are extracted to construct a comprehensive operating feature vector. A common approach is to combine candidate configuration template sets with support vector machine (SVM), random forest (RFS), and convolutional neural network (CNN) classification models for scoring, forming a probability matrix for reference when developing configuration schemes.

[0003] In terms of decision integration, single model results or arithmetic averages are often combined, resulting in a rough representation of dominance relationships and consistency, as well as their connection with historical execution records, making it difficult to form a unified confidence metric. In terms of parameter execution organization, whole package distribution or rule triggering are often the main methods, and the formal definition of hierarchical structure and backtracking path is insufficient. Summary of the Invention

[0004] In view of the aforementioned existing problems, the present invention is proposed.

[0005] Therefore, this invention provides an intelligent configuration method for electricity information collection to solve the problems of lack of dominant relationships and consistency fusion in multi-model results, unified confidence modeling, and lack of hierarchical parameter distribution.

[0006] To solve the above-mentioned technical problems, the present invention provides the following technical solution:

[0007] This invention provides an intelligent configuration method for collecting electricity information, which includes collecting equipment ledger data, power distribution topology data, historical operation data and historical configuration records, performing data processing to obtain a standardized dataset and a set of candidate configuration templates;

[0008] Based on standardized datasets, operational status features, topological features, equipment ledger features, historical configuration and alarm features are extracted to construct a comprehensive operational feature vector.

[0009] Using the comprehensive feature vector and candidate configuration template set as input, the candidate matching probability values ​​are calculated separately through support vector machine classification, random forest classification, and convolutional neural network, and a probability matrix is ​​constructed.

[0010] The odds ratio, consistency coefficient, and historical execution success rate are combined and calculated for the elements in the probability matrix to obtain the matching probability;

[0011] Sensitivity testing is performed on the candidate configuration template set based on core operational metrics from historical operational data, and the final configuration template is determined by combining the matching probability.

[0012] The final configuration template will be distributed according to the three-tier segmented execution parameters.

[0013] As a preferred embodiment of the intelligent configuration method for collecting electricity information according to the present invention, the equipment ledger data includes the terminal model, manufacturer, communication protocol, and installation location.

[0014] The power distribution topology data includes the connection relationships between nodes, path lengths, line types, and rated capacities.

[0015] The historical operation data is retrieved from the main station for each terminal, including historical operation data such as power consumption, voltage, current, acquisition rate, communication delay, and link packet loss rate;

[0016] The historical configuration records include configuration parameters, execution results, and runtime alarms.

[0017] As a preferred embodiment of the intelligent configuration method for electricity information collection described in this invention, the data processing includes: in a standardized dataset, for continuously running data, using linear interpolation, calculating intermediate values ​​proportionally according to the values ​​of adjacent time points before and after the missing point, based on the time interval.

[0018] For discrete event data, the forward imputation method is used to fill in the missing values ​​with the most recent non-missing values.

[0019] As a preferred embodiment of the intelligent configuration method for electricity information collection described in this invention, the construction of the comprehensive operation feature vector includes calculating the mean, standard deviation, maximum value, minimum value and fluctuation range of continuous operation data to construct operation status features;

[0020] For the graph structure of the distribution network topology, extract the degree, shortest path length, branch capacity ratio, and line type code of each terminal in the distribution network topology graph to construct topology features;

[0021] Feature extraction is performed on the equipment ledger. Model, manufacturer, and communication protocol are extracted using one-hot encoding. Installation location is converted into latitude and longitude coordinate difference to construct equipment ledger features.

[0022] Parameter frequency statistics are used to analyze historical configuration parameters, extract success rate and alarm rate, and construct historical configuration and alarm features.

[0023] As a preferred embodiment of the intelligent configuration method for electricity information collection described in this invention, the step of taking the comprehensive running feature vector and the candidate configuration template set as input and calculating the candidate matching probability value separately through support vector machine classification, random forest classification and convolutional neural network includes training the historical running feature vector and corresponding template labels through support vector machine classification, using radial basis function kernel as kernel function, and selecting the penalty factor and kernel function parameters through cross-validation to output the SVM candidate matching probability value;

[0024] Modeling is performed on the same training data using random forest classification, with CART decision tree as the base classifier. The number and maximum depth of decision trees are determined by minimizing out-of-bag error. The RF candidate matching probability value is obtained based on the voting ratio of all decision trees in the forest.

[0025] The comprehensive running feature vector is rearranged into a one-dimensional feature matrix and input into the convolutional layer through a convolutional neural network. Supervised training is performed using historical running datasets, the optimization function is the cross-entropy loss function, and the softmax output is used as the CNN candidate matching probability value.

[0026] As a preferred embodiment of the intelligent configuration method for electricity information collection described in this invention, the step of performing the fusion calculation of the advantage ratio, consistency coefficient and historical execution success rate of the elements in the probability matrix includes converting the elements in the probability matrix into advantage ratios.

[0027] Calculate the standard deviation of the candidate matching probability values ​​of each candidate configuration template in the three algorithms, and map the standard deviation to the consistency coefficient through a monotonically decreasing function;

[0028] The historical execution success rate is calculated based on the execution results issued according to historical configurations.

[0029] As a preferred embodiment of the intelligent configuration method for electricity information collection described in this invention, the method for obtaining the matching probability includes calculating the matching probability by constraining the advantage ratio of three algorithms through equal-weighted geometric average and by multiplicative aggregation.

[0030] As a preferred embodiment of the intelligent configuration method for electricity information collection described in this invention, the sensitivity detection of the candidate configuration template set based on the core operating indicators of historical operating data includes determining the core operating indicators as electricity consumption, voltage, current, collection rate, communication delay and link packet loss rate based on historical operating data, and defining the performance indicator benchmark value according to the historical average of each core operating indicator.

[0031] Each core operational indicator in the historical operational data is perturbed, and the corresponding performance indicator perturbation value is calculated.

[0032] The perturbation sensitivity coefficient is calculated based on the performance index baseline value and the performance index perturbation value.

[0033] As a preferred embodiment of the intelligent configuration method for electricity information collection described in this invention, the step of determining the final configuration template includes fusing the perturbation sensitivity coefficient with the matching probability to obtain the optimal matching probability.

[0034] Simultaneously, based on the natural fluctuations of core operating indicators in historical operating data, a sensitivity threshold is determined, and candidate configuration templates with perturbation sensitivity coefficients higher than the sensitivity threshold are eliminated. Among the remaining candidate configuration templates, they are sorted according to the best matching probability, and the final configuration template is determined.

[0035] As a preferred embodiment of the intelligent configuration method for electricity information collection described in this invention, the step of distributing the final configuration template in three-level segmented execution parameters includes, after determining the final configuration template, dividing the configuration scheme into a core operating parameter set, a key optimization parameter set, and a performance enhancement parameter set, and distributing the execution parameters in three levels in sequence.

[0036] The beneficial effects of this invention are as follows: By fusing the odds ratio, consistency coefficient, and historical execution success rate of the elements in the probability matrix, the matching probability is obtained, achieving same-dimensional mapping of evidence and unified scale measurement, and performing sensitivity detection, thus making the prediction results more accurate. Three prediction models are computed in parallel to ensure the reliability of the calculation results. The final configuration template is distributed according to three-level segmented execution parameters, making configuration more flexible and enabling faster error detection when some parameters are incorrect, reducing maintenance time. Attached Figure Description

[0037] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the following description of the embodiments will be briefly introduced. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0038] Figure 1 A flowchart for an intelligent configuration method for electricity consumption information collection.

[0039] Figure 2 This is a flowchart for calculating the independent probabilities of the three models.

[0040] Figure 3 This is a flowchart of the decision-making process after probability fusion.

[0041] Figure 4 The flowchart for issuing and executing parameters. Detailed Implementation

[0042] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings.

[0043] Many specific details are set forth in the following description in order to provide a full understanding of the invention. However, the invention may also be practiced in other ways different from those described herein, and those skilled in the art can make similar extensions without departing from the spirit of the invention. Therefore, the invention is not limited to the specific embodiments disclosed below.

[0044] Secondly, the term "one embodiment" or "embodiment" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The phrase "in one embodiment" appearing in different places in this specification does not necessarily refer to the same embodiment, nor is it a single or selective embodiment that is mutually exclusive with other embodiments.

[0045] Reference Figures 1-4 As an embodiment of the present invention, this embodiment provides an intelligent configuration method for electricity consumption information collection, including the following steps:

[0046] S1: Collect equipment ledger data, power distribution topology data, historical operation data, and historical configuration records, process the data, and obtain a standardized dataset and a set of candidate configuration templates.

[0047] Furthermore, the collected data is divided into four main categories: equipment ledger data, power distribution topology data, historical operation data, and historical configuration records.

[0048] The equipment ledger data includes the terminal's model, manufacturer, communication protocol, and installation location.

[0049] The topology data is obtained from the distribution network and stored using a graph data structure. Nodes represent equipment and substations, and edges represent line connections. The distribution topology data includes the connection relationships between nodes, path lengths, line types, and rated capacities.

[0050] Historical operating data is retrieved from the main station for each terminal, including historical operating data such as power consumption, voltage, current, acquisition rate, communication latency, and link packet loss rate.

[0051] Historical configuration records include configuration parameters, execution results, and runtime alarms.

[0052] For data processing, linear interpolation is used for continuously running data (such as voltage and current). Intermediate values ​​are calculated proportionally to the time intervals between the values ​​of adjacent time points before and after the missing point.

[0053] For discrete event data (such as configuration parameters and execution results), the forward imputation method is used to fill in the missing values ​​with the most recent non-missing values.

[0054] Finally, the processed data is standardized to obtain a standardized dataset, and the data collected from different terminal data sources is uniformly mapped to the IDs in the device ledger table.

[0055] After completing the collection of four main categories of data, a candidate configuration template set is constructed based on historical configuration profiles and new configuration templates that improve upon historical configuration profiles using expert experience. K represents the total number of configuration templates.

[0056] S2: Extract operating status features, topology features, equipment ledger features, historical configuration and alarm features from standardized datasets to construct a comprehensive operating feature vector.

[0057] Specifically, constructing a comprehensive operational feature vector includes calculating the mean, standard deviation, maximum value, minimum value, and fluctuation range of continuously running data to construct operational status features;

[0058] For the graph structure of the distribution network topology, extract the degree, shortest path length, branch capacity ratio, and line type code of each terminal in the distribution network topology graph to construct topology features;

[0059] Feature extraction is performed on the equipment ledger. Model, manufacturer, and communication protocol are extracted using one-hot encoding. Installation location is converted into latitude and longitude coordinate difference to construct equipment ledger features.

[0060] Parameter frequency statistics are used to analyze historical configuration parameters, extract success rate and alarm rate, and construct historical configuration and alarm features.

[0061] Furthermore, the standardized dataset is transformed into a vector form that can be processed by pattern recognition algorithms.

[0062] First, continuous numerical operation features are extracted. For continuous operation data such as voltage, current, and energy, the mean, standard deviation, maximum value, minimum value, and fluctuation range are calculated. These five statistical features are used to construct an operation status feature that includes the mean, standard deviation, maximum value, minimum value, and fluctuation range of continuous operation data. .

[0063] Then, feature extraction is performed on the topology. Utilizing the graph structure of the distribution network topology, the degree (number of connections), shortest path length, branch capacity ratio, and line type code of each terminal in the distribution network topology graph are extracted. For example, if the shortest path length between terminal A and the substation is 3 edges, and the sum of the rated capacities of its branches is 10 MVA, then the terminal's path capacity ratio feature is... Represented as:

[0064] ;

[0065] in, This refers to the rated power requirement of the terminal.

[0066] Construct topology features including degree, shortest path length, branch capacity ratio, and line type encoding. .

[0067] Topological features in this embodiment Includes node degree, shortest path length, and branch capacity ratio. And the line type code (e.g., overhead line = 1, cable line = 0).

[0068] For example, if the terminal node degree is 2, the shortest path length is 3, the branch capacity ratio is 0.75, and the line type is overhead line, then:

[0069] ;

[0070] Secondly, feature extraction is performed on the equipment ledger. Model, manufacturer, and communication protocol are extracted using one-hot encoding. Installation location is converted into latitude and longitude coordinate differences to reflect the potential impact of geographical distance on communication. A feature set for the equipment ledger is constructed, including one-hot encoding of manufacturer, model, and communication protocol, as well as latitude and longitude differences of installation location. .

[0071] Equipment ledger features in this embodiment The model number, manufacturer, and communication protocol in the code use one-hot encoding (e.g., communication protocols DL / T 645 and IEC 60870-5 are mapped to...). ; Convert installation location to latitude and longitude coordinate difference .

[0072] For example, if the manufacturer code is The specification code is The position difference is ,but:

[0073] ;

[0074] Finally, historical configuration and alarm features were extracted. The parameters issued in the historical configuration were statistically analyzed using parameter frequency, such as the proportion of the last 100 configurations with a collection interval of 60 seconds.

[0075] The execution results are expressed as a success rate, calculated as the ratio of successful attempts to the total number of attempts. Alarm data is expressed as an alarm rate, calculated as the ratio of alarms per unit time to total runtime. Historical configuration and alarm features, including parameter frequency, configuration success rate, and alarm rate, are constructed. .

[0076] For example, if the percentage of parameter frequency vector acquisition intervals is 0.8 for 60s and 0.2 for 30s, the configuration success rate is 0.9, and the alarm rate is 0.05, then:

[0077] ;

[0078] Max-min normalization is performed on all features with numerical values.

[0079] Construct a comprehensive operational feature vector by arranging the data in the order of their sources. , is represented as:

[0080] ;

[0081] in: This indicates a concatenation operation on four vectors.

[0082] S3: Using the integrated feature vector and candidate configuration template set as input, the candidate matching probability values ​​are calculated separately through support vector machine classification, random forest classification and convolutional neural network, and a probability matrix is ​​constructed.

[0083] Furthermore, to avoid bias caused by relying on a single algorithm, this embodiment uses three existing mature algorithms to independently calculate the candidate matching probability value.

[0084] First, candidate matching probability values ​​are calculated using Support Vector Machine (SVM) classification.

[0085] The training data uses historical running feature vectors With the corresponding template tags .by Using the device ID and time window as the main body, it is associated with historical configuration records to obtain the template ID that is effective in the current period, which serves as the template tag for this sample. .

[0086] A one-vs-rest strategy is adopted, training a binary classification SVM for each template in the candidate configuration template set, with the selected template as the positive class and the remaining templates as the negative class. The kernel function is a radial basis function kernel.

[0087] Penalty factors were selected through cross-validation. and kernel function parameters When the category distribution is unbalanced, class weight balancing is enabled to reduce the risk of misjudgment by long-tail templates.

[0088] Each binary classifier is fitted on the training set. During cross-validation, the macro-average and log-loss are used as the main evaluation metrics. When the macro-averages are the same, the combination with better probability quality (log-loss) is prioritized. The original decision values ​​of SVM are not probabilities. To ensure that the output is on the same scale as other models, the decision scores of each binary classifier are first calibrated using latt sigmoid monotonic probability on the validation set. For each sample, the calibrated scores of all binary classifiers are calculated separately and used as the SVM candidate matching probability values.

[0089] Next, the candidate matching probability value is calculated by Random Forest (RF) classification.

[0090] Using standardized historical operational data as a sample, historical configuration records are associated with device identifiers and time windows, and configuration templates that are actually effective and successfully executed within the time window are taken as template tags.

[0091] The training, validation, and test sets are divided chronologically. The base learner is a CART decision tree classifier. Multiple sample subsets are generated from the training data using sampling with replacement, and multiple trees are trained on each subset. At each node split, a subset of features is randomly selected to participate in the candidate set, forming a decorrelated ensemble of trees. Out-of-bag (OOB) evaluation is enabled to monitor generalization and convergence without using the validation set. When the class distribution is imbalanced, sample weights are automatically adjusted according to class frequency to reduce the risk of misclassification of long-tailed templates. During training, the number of trees is gradually increased, and different depth limits and minimum leaf node sample settings are compared. A robust configuration is selected based on a combination of OOB metrics and validation set performance.

[0092] The number of decision trees is determined by minimizing the out-of-bag error. Maximum depth The matching probability value of RF candidates is obtained by using the voting ratio of all decision trees in the forest. , is represented as:

[0093] ;

[0094] in, The number of decision trees in the random forest. For the first Each decision tree is used to process the input feature vector. The classification results For indicator functions, when The value is 1 when it is true, and 0 otherwise.

[0095] Finally, a convolutional neural network (CNN) is used to calculate the candidate matching probability value.

[0096] Will The features are rearranged into a one-dimensional feature matrix and input to the convolutional layer.

[0097] Supervised learning is employed, with cross-entropy as the objective function for training. The optimizer uses a mainstream adaptive method, with a gradually decaying learning rate. During training, convergence is monitored using either grouped cross-validation or time-deferred validation. Early stopping is enabled, meaning the training stops and rolls back to the optimal weights when the loss on the validation set no longer improves or overfitting occurs. When class distribution is imbalanced, class weights or sample resampling are used to reduce the risk of long-tail templates being ignored. The final model version is determined by combining the probability quality of the validation set with the overall hit performance. The softmax output is used as the candidate matching probability value. , is represented as:

[0098] ;

[0099] in, Candidate configuration templates for fully connected layers The output value.

[0100] Calculate the three candidate matching probability values ​​for each candidate configuration template, and construct a 3x3 K-column probability matrix. The element in the v-th row and k-th column of the probability matrix is... , where v is the v-th algorithm.

[0101] S4: Perform a combined calculation of the odds ratio, consistency coefficient, and historical execution success rate on the elements in the probability matrix to obtain the matching probability.

[0102] Specifically, the calculation of the odds ratio, consistency coefficient and historical execution success rate of the elements in the probability matrix includes converting the elements in the probability matrix into odds ratios;

[0103] Calculate the standard deviation of the candidate matching probability values ​​of each candidate configuration template in three independent algorithms, and map the standard deviation to a consistency coefficient through a monotonically decreasing function;

[0104] The historical execution success rate is calculated based on the execution results issued according to historical configurations.

[0105] Furthermore, in existing probabilistic fusion methods, directly using candidate matching probability values ​​for weighted summation often masks significant differences between different algorithms. To enhance discriminative power, candidate matching probability values ​​are first converted into advantage ratios. Advantage ratios clearly reflect the relative probability of an event occurring versus not occurring. In this scenario, if the candidate matching probability value of a candidate configuration template is close to 1, the advantage ratio increases exponentially, highlighting the template's advantage over other templates and thus avoiding the dilution of high-scoring templates in simple probability weighting. Advantage Ratio Represented as:

[0106] ;

[0107] in, For the first Algorithms for candidate configuration templates The advantage ratio, For the first Templates obtained by various algorithms Matching probability.

[0108] when The closer it is to 1, the greater the advantage ratio, reflecting the strong advantage of the candidate configuration template under the algorithm.

[0109] In the process of fusing the probability results of multiple algorithms, the first issue to address is the potential for significant differences between the outputs of different algorithms. Simply averaging or weighting these probabilities directly can be susceptible to interference from outliers, thus reducing the reliability of the final selection. Therefore, we first consider using a consistency coefficient to quantify the degree of consistency among the three algorithms' results for the same candidate configuration template.

[0110] The most commonly used statistical method for measuring consistency is standard deviation, which reflects the degree of dispersion between numerical values. The smaller the standard deviation, the closer the results are and the more consistent the algorithms are; the larger the standard deviation, the more significant the differences are and the more divergent the algorithms are. However, directly using standard deviation for calculations has two problems: firstly, the numerical range is not easy to combine with other coefficients; secondly, when the standard deviation is zero, it may lead to unreasonable calculation results.

[0111] Therefore, the standard deviation is assigned a monotonically decreasing trend, and the consistency coefficient is expressed as:

[0112] ;

[0113] in, template Consistency coefficient Three algorithms for template The standard deviation of the candidate matching probability values;

[0114] When the standard deviation is small (low discrepancy), When the standard deviation is close to 1; when the standard deviation is large (high discrepancy), Close to 0.

[0115] Relying solely on probability and consistency coefficients calculated by algorithms may still introduce some bias. For example, a configuration template might perform well in terms of data characteristics, but frequently fail to deploy in practice. Therefore, historical execution success rate is used as a supplementary metric. Historical execution success rate reflects the template's stability in past applications and can bridge the gap between algorithm predictions and actual execution. For entirely new configuration templates, a value of 1 is used as the historical execution success rate. Determined based on the execution result.

[0116] Specifically, obtaining the matching probability includes constraining the advantage ratio of the three algorithms through equal-weighted geometric average and calculating the matching probability through multiplicative aggregation.

[0117] A matching probability is constructed based on the odds ratio, consistency coefficient, and historical execution success rate. Represented as:

[0118] ;

[0119] Arithmetic averaging and weighted summation may lead to the result of a single algorithm becoming overly dominant. Therefore, the fusion method used in this embodiment can avoid the extreme cases of a certain algorithm and ensure overall robustness.

[0120] S5: Based on the core operating indicators of historical operating data, perform sensitivity testing on the candidate configuration template set, and combine the matching probability to determine the final configuration template.

[0121] Furthermore, before proceeding to sensitivity testing, this step requires clearly defining the parameters used and how they are obtained:

[0122] The core operating indicators are six items from historical operating data: power consumption, voltage, current, acquisition rate, communication latency, and link packet loss rate.

[0123] Historical operational data is retrieved, and the average value is calculated for each core operational indicator to define a candidate matching template. Performance under undisturbed conditions, serving as a benchmark for performance indicators. , where j is one of the core operating indicators.

[0124] Perturbations are applied to the core performance indicators of historical operating data, and the results of each performance indicator are recalculated after the perturbations, serving as the perturbation values ​​for the performance indicators. .

[0125] The sensitivity coefficient is used to measure the degree of change in performance indicators of a candidate configuration template when the operating parameters are slightly disturbed, reflecting the candidate configuration template's ability to resist interference from environmental changes.

[0126] The perturbation sensitivity coefficient of the candidate configuration template is calculated based on six core operational indicators. , is represented as:

[0127] ;

[0128] If it is a new configuration template, then Take 0.

[0129] Perturbation sensitivity coefficient and matching probability Combine to obtain the best matching probability , is represented as:

[0130] ;

[0131] The sensitivity threshold is obtained by statistically analyzing the core operating indicators of historical operating data (by the mean and variance range of the natural volatility of the core operating indicators). Candidate configuration templates with perturbation sensitivity coefficients higher than the sensitivity threshold are eliminated, and the candidate configuration template with the highest best matching probability is selected as the final configuration template.

[0132] S6: Distribute the final configuration template according to the three-level segmented execution parameters.

[0133] Specifically, the final configuration template will be distributed in three segments according to the three levels. After the final configuration template is determined, the configuration scheme will be divided into a core running parameter set, a key optimization parameter set, and a performance enhancement parameter set, and the execution parameters will be distributed in segments according to the three levels.

[0134] Furthermore, after determining the final configuration template, the configuration scheme is divided into three categories of data, as shown in Table 1.

[0135] Table 1 Configuration Scheme:

[0136] Parameter classification Field Name Example values Range of values Function Description Core operating parameter set Data collection interval 60 15~300 The terminal's data acquisition frequency is determined by its range; too short a frequency increases the load, while too long a frequency reduces real-time performance. Core operating parameter set Communication protocol type DL / T645-2007 Various rules and agreements Determine the communication protocol between the terminal and the master station; errors will cause communication interruption. Core operating parameter set Data compression methods Differential compression Differential / Run / Uncompressed Determines data storage and transmission efficiency Key optimization parameter set Cache size 512 128~2048 Affects data buffering capacity and latency performance Key optimization parameter set Number of retransmissions 3 1~5 Excessive retransmission attempts during communication failures may lead to congestion. Performance Enhancement Parameter Set Compression level 2 1~5 Compression level: the higher the value, the higher the compression ratio, but the greater the computational cost. Performance Enhancement Parameter Set Data prefetching strategy Pre-fetch on demand Prefetch all / Prefetch on demand Preloading data impacts caching efficiency. Performance Enhancement Parameter Set Load balancing strategy Least connections Polling / Least Connections Scheduling strategies in multi-link environments

[0137] The parameter distribution is carried out in three stages to avoid distributing all parameters at once. If a parameter is not compatible, the entire terminal will be paralyzed. In addition, there are too many parameters to check, and the time to restore normal operation is too long.

[0138] Therefore, it is necessary to first verify whether the minimum functional state of sending the core operating parameter set to the terminal can be run, and then optimize and adjust it in turn.

[0139] The significance of layered execution is to first verify the stability of core operating parameters to ensure that the terminal is in a "runnable" minimum functional state, and then gradually issue optimized and enhanced parameters.

[0140] When the optimal matching probability is obtained At that time, the first level of parameters is distributed, and the core operating parameter set of the final configuration template (collection interval time, communication protocol type, data compression method) is distributed; one operating cycle of the terminal is used as the monitoring time, and real-time performance curves (collection rate curve, latency curve, path change curve) are constructed during the monitoring process and plotted through periodic collection and interpolation methods;

[0141] If any core abnormal events, including data acquisition interruption, communication interruption, and excessive latency, are triggered during monitoring, subsequent execution will be immediately suspended and the system will revert to the previous configuration template.

[0142] If the monitored data is running normally, then the second level of parameters will be distributed, including a set of key optimization parameters.

[0143] Establish a performance improvement curve and use a moving average function to smooth out indicators such as acquisition rate and latency to determine the trend of performance changes;

[0144] If the overall performance improvement curve is lower than the performance curve of the previous configuration template, select a configuration template with the same key optimization parameter set and performance enhancement parameter set as the final configuration template, replace the core running parameter set, and reissue the first-level parameters.

[0145] If the overall performance improvement curve is higher than the performance curve of the previous configuration template, then the third-level parameters are issued; the performance enhancement parameter set is issued.

[0146] Start long-term monitoring to monitor the acquisition rate curve, latency curve, and network load curve.

[0147] If a long-term performance degradation event occurs during long-term monitoring (the acquisition rate curve, latency curve, and network load curve are all lower than the average acquisition rate curve, average latency curve, and average network load curve constructed from historical data of the terminal), the system will automatically roll back to the second-level parameter set and trigger a first-level re-evaluation.

[0148] If no degradation event occurs, the final configuration template will be retained.

[0149] This embodiment also provides a computer device applicable to the centralized heating network pressure monitoring method based on artificial intelligence, including: a memory and a processor; the memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions to realize the centralized heating network pressure monitoring method based on artificial intelligence as proposed in the above embodiment.

[0150] The computer device can be a terminal, comprising a processor, memory, communication interface, display screen, and input devices connected via a system bus. The processor provides computing and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system and computer programs. The internal memory provides an environment for the operation of the operating system and computer programs stored in the non-volatile storage media. The communication interface is used for wired or wireless communication with external terminals; wireless communication can be achieved through Wi-Fi, carrier networks, NFC (Near Field Communication), or other technologies. The display screen can be an LCD screen or an e-ink screen. The input devices can be a touch layer covering the display screen, buttons, a trackball, or a touchpad on the computer device's casing, or an external keyboard, touchpad, or mouse.

[0151] This embodiment also provides a storage medium storing a computer program. When executed by a processor, the program implements the artificial intelligence-based centralized heating network pressure monitoring method proposed in the above embodiments. The storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as Static Random Access Memory (SRAM), Electrically Erasable Programmable Read-Only Memory (EEPROM), Erasable Programmable Read Only Memory (EPROM), Programmable Red-Only Memory (PROM), Read-Only Memory (ROM), magnetic storage, flash memory, magnetic disk, or optical disk.

[0152] In summary, this invention achieves more accurate predictions by: fusing the odds ratio, consistency coefficient, and historical execution success rate of elements in the probability matrix to obtain the matching probability; realizing same-dimensional mapping of evidence and unified scale measurement; and performing sensitivity detection. Parallel computation of three prediction models ensures the reliability of the calculation results. The final configuration template is distributed according to three-tiered segmented execution parameters, allowing for more flexible configuration and faster error detection when some parameters are incorrect, thus reducing maintenance time.

[0153] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A method for intelligent configuration of electricity consumption information collection, characterized in that: include, Collect equipment ledger data, power distribution topology data, historical operation data, and historical configuration records, process the data, and obtain a standardized dataset and a set of candidate configuration templates. Based on standardized datasets, operational status features, topological features, equipment ledger features, historical configuration and alarm features are extracted to construct a comprehensive operational feature vector. Using the comprehensive feature vector and candidate configuration template set as input, the candidate matching probability values ​​are calculated separately through support vector machine classification, random forest classification, and convolutional neural network, and a probability matrix is ​​constructed. The odds ratio, consistency coefficient, and historical execution success rate are combined and calculated for the elements in the probability matrix to obtain the matching probability; Sensitivity testing is performed on the candidate configuration template set based on core operational metrics from historical operational data, and the final configuration template is determined by combining the matching probability. The final configuration template will be distributed according to the three-tier segmented execution parameters. The step of taking the comprehensive running feature vector and the candidate configuration template set as input and calculating the candidate matching probability value separately through support vector machine classification, random forest classification and convolutional neural network includes training the historical comprehensive running feature vector and corresponding template label through support vector machine classification, using radial basis function kernel as kernel function, and selecting penalty factor and kernel function parameters through cross-validation to output SVM candidate matching probability value; Modeling is performed on the same training data using random forest classification, with CART decision tree as the base classifier. The number and maximum depth of decision trees are determined by minimizing out-of-bag error. The RF candidate matching probability value is obtained based on the voting ratio of all decision trees in the forest. The comprehensive running feature vector is rearranged into a one-dimensional feature matrix by a convolutional neural network and input into the convolutional layer. Supervised training is performed using historical running data. The optimization function is the cross-entropy loss function, and the softmax output is used as the CNN candidate matching probability value. The calculation of the fusion of the odds ratio, consistency coefficient and historical execution success rate of the elements in the probability matrix includes converting the elements in the probability matrix into odds ratios; Calculate the standard deviation of the candidate matching probability values ​​of each candidate configuration template in the three algorithms, and map the standard deviation to the consistency coefficient through a monotonically decreasing function; Calculate the historical execution success rate based on the execution results issued by the historical configuration; When the optimal matching probability is obtained, the first level of parameters is distributed, and the core set of runtime parameters for the final configuration template is distributed. The monitoring time is taken as one operating cycle of the terminal. During the monitoring process, a real-time performance curve is constructed and plotted using periodic data collection and interpolation methods. If any core abnormal events, including data acquisition interruption, communication interruption, and excessive latency, are triggered during monitoring, subsequent execution will be immediately suspended and the system will revert to the previous configuration template. If the monitored data is running normally, then the second level of parameters will be distributed, including a set of key optimization parameters. Establish a performance improvement curve and use a moving average function to smooth out indicators such as acquisition rate and latency to determine the trend of performance changes; If the overall performance improvement curve is lower than the performance curve of the previous configuration template, select a configuration template with the same key optimization parameter set and performance enhancement parameter set as the final configuration template, change the core running parameter set, and reissue the first-level parameters. If the overall performance improvement curve is higher than the performance curve of the previous configuration template, then the third-level parameters will be issued. Issue the performance enhancement parameter set; Start long-term monitoring to monitor the acquisition rate curve, latency curve, and network load curve; If a long-term performance degradation event occurs during long-term monitoring, the system will automatically roll back to the second-level parameter set and trigger a first-level re-evaluation. If no degradation event occurs, the final configuration template will be retained.

2. The intelligent configuration method for electricity information collection as described in claim 1, characterized in that: The equipment ledger data includes the terminal's model, manufacturer, communication protocol, and installation location; The power distribution topology data includes the connection relationships between nodes, path lengths, line types, and rated capacities. The historical operating data refers to the historical operating data of each terminal retrieved from the main station, including power consumption, voltage, current, acquisition rate, communication delay, and link packet loss rate. The historical configuration records include configuration parameters, execution results, and runtime alarms.

3. The intelligent configuration method for electricity information collection as described in claim 2, characterized in that: The data processing includes, in a standardized dataset, using linear interpolation for continuously running data, calculating intermediate values ​​proportionally to the time intervals between the values ​​of adjacent time points before and after the missing point. For discrete event data, the forward imputation method is used to fill in the missing values ​​with the most recent non-missing values.

4. The intelligent configuration method for electricity information collection as described in claim 3, characterized in that: The construction of the comprehensive operational feature vector includes calculating the mean, standard deviation, maximum value, minimum value, and fluctuation range of continuous operational data to construct operational status features; For the graph structure of the distribution network topology, extract the degree, shortest path length, branch capacity ratio, and line type code of each terminal in the distribution network topology graph to construct topology features; Feature extraction is performed on the equipment ledger. Model, manufacturer, and communication protocol are extracted using one-hot encoding. Installation location is converted into latitude and longitude coordinate differences to construct equipment ledger features. Parameter frequency statistics are used to analyze historical configuration parameters, extract success rate and alarm rate, and construct historical configuration and alarm features.

5. The intelligent configuration method for electricity information collection as described in claim 4, characterized in that: The process of obtaining the matching probability includes constraining the advantage ratio of the three algorithms by equal-weighted geometric average and calculating the matching probability by multiplicative aggregation.

6. The intelligent configuration method for electricity information collection as described in claim 5, characterized in that: The sensitivity detection of the candidate configuration template set based on the core operating indicators based on historical operating data includes determining the core operating indicators as power consumption, voltage, current, acquisition rate, communication delay and link packet loss rate based on historical operating data, and defining the performance indicator benchmark value according to the historical average of each core operating indicator. Each core operational indicator in the historical operational data is perturbed, and the corresponding performance indicator perturbation value is calculated. The perturbation sensitivity coefficient is calculated based on the performance index baseline value and the performance index perturbation value.

7. The intelligent configuration method for electricity information collection as described in claim 6, characterized in that: The process of determining the final configuration template includes fusing the perturbation sensitivity coefficient with the matching probability to obtain the optimal matching probability. Simultaneously, based on the natural fluctuations of core operating indicators in historical operating data, a sensitivity threshold is determined, and candidate configuration templates with perturbation sensitivity coefficients higher than the sensitivity threshold are eliminated. Among the remaining candidate configuration templates, they are sorted according to the best matching probability, and the final configuration template is determined.

Images