A power marketing volume fee data intelligent early warning method and system and a storage medium

Abstract

The application discloses a kind of electric power marketing quantity fee data intelligent early warning method, system and storage medium, belong to electric power marketing big data intelligent inspection technical field.Method includes to the standardization preprocessing of multiple source original business data, constructs quantification multivariate group model, converts business rule into quantification constraint logic, calculates business index deviation characteristics and is mapped to constraint conflict intensity and carries out index fusion, generates constraint conflict intensity and rule violation degree after fusion;The index and time series fluctuation factor, historical abnormal factor input risk inference model, calculate risk quantization value;Get artificial review result and construct feedback sample set, process and generate update rule set after hot update to dynamic rule base and backflow reuse;Continuous acquisition performance index reverse tuning core parameter, form quantification optimization closed loop.The application effectively solves the problems, such as traditional early warning rule solidification and high false alarm rate, realizes the whole process self-adapting iteration of marketing quantity fee data early warning.

Description

Technical Field

[0001] This invention relates to an intelligent early warning method, system, and storage medium for electricity marketing data, belonging to the field of intelligent auditing technology for big data in electricity marketing. Background Technology

[0002] Electricity marketing data constitutes a typical multi-dimensional, time-series, strongly constrained data system, encompassing heterogeneous data from multiple sources, including user static profiles, time-series meter readings, time-of-use electricity data, electricity pricing policy parameters, electricity billing results, and distribution area line loss indicators. These data exhibit strict business coupling constraints and numerical calculation relationships. Existing traditional marketing data early warning technologies suffer from significant structural defects and cannot adapt to the needs of intelligent auditing. Specific technical shortcomings are as follows:

[0003] First, the early warning rules are static and fixed, lacking dynamic quantitative adaptation capabilities. Traditional early warning systems rely entirely on manually setting fixed thresholds for single-value comparisons, without a quantitative adaptive mechanism. This makes them unable to adapt to dynamic scenarios such as seasonal production fluctuations, iterative updates to electricity pricing policies, and user capacity increases / cancellations / class changes. Consequently, the false alarm rate and missed alarm rate remain consistently high, resulting in extremely poor robustness of early warning systems.

[0004] Second, it lacks the ability to recognize and model business rules, and only performs superficial numerical discrimination. Existing technologies only perform threshold judgments on individual electricity and electricity bill data, and cannot model the coupling constraint relationship between electricity consumption, electricity price, data records, and line loss through mathematical models. They also cannot quantitatively distinguish between normal business fluctuations and real volume and bill anomalies, resulting in extremely low accuracy in intelligent identification.

[0005] Third, there is no autonomous evolutionary iterative mathematical mechanism, and rule updates rely entirely on manual intervention. Traditional rule bases lack sample feedback loops and quantitative optimization algorithms. The accumulation of abnormal scenarios, rule threshold deviations, and new business scenarios all require manual sorting, configuration, and updates. The iteration cycle is long and the lag is strong, making it unsuitable for real-time early warning scenarios with massive user big data.

[0006] In summary, existing technologies lack a unified modeling method for business constraint cognitive space, an abnormal risk potential field reasoning mechanism, and a rule credibility convergence and self-evolution mechanism. They suffer from fundamental defects, low intelligence, and poor early warning accuracy. There is an urgent need for a new intelligent early warning method with quantitative cognition, accurate reasoning, and autonomous iterative upgrading capabilities. Summary of the Invention

[0007] The purpose of this invention is to overcome the shortcomings of the prior art and provide an intelligent early warning method, system and storage medium for electricity marketing volume and fee data. It effectively solves the technical problems of rigid early warning rules, insufficient business cognition ability and high false alarm rate, and realizes full-process adaptive iteration and accurate identification of marketing volume and fee data early warning.

[0008] To achieve the above objectives, the present invention is implemented using the following technical solution:

[0009] In a first aspect, the present invention provides an intelligent early warning method for electricity marketing volume and fee data, comprising:

[0010] Preprocess the acquired multi-source raw marketing volume and cost data to generate a standardized business dataset;

[0011] Based on the standardized business dataset, a quantitative plural group model is constructed to transform business rules into quantitative constraint logic, calculate business indicator deviation features, and map the business indicator deviation features to the pre-generated constraint conflict intensity for indicator fusion, generating the fused constraint conflict intensity and simultaneously generating the rule violation degree.

[0012] The fused constraint conflict strength and rule violation degree, along with the introduced time-series fluctuation factor and historical anomaly factor, are input into a pre-built risk potential field inference model to calculate the risk quantification value.

[0013] Anomaly classification is determined based on the risk quantification value, and anomaly source is traced based on the entity association relationship in the quantified plural group model. Early warning information with built-in feature vectors is generated, and the early warning information is output as the early warning result.

[0014] Obtain the verification results of the warning information, label the feature vector of each warning information with a classification label, and construct a feedback sample set;

[0015] Based on the feedback sample set, rule optimization is performed to generate an optimized updated rule set;

[0016] Load the updated rule set, update the dynamic rule base, and reuse the updated rule system back to the construction stage of the quantitative plural model, so that the business indicator deviation calculation and constraint judgment can be re-executed for newly connected standardized business datasets.

[0017] The performance indicators of the continuously collected early warning information are used to adjust the core parameters of the risk potential field inference model and the rule optimization based on the performance indicators, thereby achieving closed-loop adaptive iteration.

[0018] Furthermore, the multi-source raw marketing data includes user profiles, time-series meter readings, electricity billing, and transformer area line loss data; the preprocessing includes missing value repair, noise suppression, time-series alignment, unit normalization, and business semantic mapping.

[0019] Furthermore, based on the standardized business dataset, a quantitative plural model is constructed to transform business rules into quantitative constraint logic, calculate business indicator deviation features, and map these features to pre-generated constraint conflict strengths for indicator fusion, generating a fused constraint conflict strength. Simultaneously, the degree of rule violation is generated, including:

[0020] Based on the standardized business dataset, extract business entities. Entity Relationship Topology Rigidity and flexibility verification rules and threshold under constraints and the upper threshold of the constraint Construct a quantitative plural group model:

[0021] ;

[0022] Based on the aforementioned quantitative plural model, all text-based electricity marketing business rules are transformed into quantitative constraint logic that can be numerically calculated.

[0023] The total periodic settlement electricity in the standardized business dataset Standard unit price and power factor adjustment coefficient Substitute the data into the data usage calculation formula to calculate the theoretical electricity cost. The formula is as follows:

[0024] ;

[0025] Combined with the actual settlement electricity charges in the standardized business dataset Calculate the deviation characteristics of business indicators The formula is as follows:

[0026] ;

[0027] The quantitative tuple model was compared item by item. The Middle Threshold under the constraints of business items and threshold constraints The deviation from the actual business indicator value generates the first Constraint Conflict Strength ;

[0028] Traversing the The Middle Each rule assigns a value based on whether the current business data meets the rule logic and the severity of the violation, generating a rule violation level. ;

[0029] The deviation characteristics of the business indicators Mapped to the constraint conflict strength By fusing the indicators, the combined constraint conflict strength is obtained. .

[0030] Furthermore, the fused constraint conflict strength and rule violation degree, along with the introduced time-series fluctuation factor and historical anomaly factor, are input into a pre-constructed risk potential field inference model to calculate the risk quantification value, as shown in the formula:

[0031] ;

[0032] in, The aforementioned risk quantification value, The strength of the constraint conflict after fusion. As to the degree of rule violation, For time series fluctuation factor, As a historical outlier, For the first item The weighting coefficients, For the first Degree of violation of the rule The weighting coefficients, Time series fluctuation factor The weighting coefficients, Historical outlier The weighting coefficients, It is the total number of constraint conflict strengths. It is the total number of rule violations;

[0033] Anomaly classification is determined based on the risk quantification value, and anomaly tracing is performed based on the entity association relationships in the quantified tuple model. Early warning information with built-in feature vectors is generated, and the early warning information is output as the warning result, including:

[0034] Based on the preset three risk thresholds , , According to the aforementioned risk quantification value The warning levels are divided into intervals:

[0035] when ≤ < At that time, it was determined to be a general warning;

[0036] when ≤ < At that time, it was determined to be an important warning;

[0037] when ≥ At that time, it was determined to be an emergency warning;

[0038] According to the quantized tuple model Entity association topology Tracing the abnormal propagation path in reverse, locating the abnormal source entity, conflict constraints, and scope of impact;

[0039] The warning level and the deviation characteristics of the business indicators Risk quantification value Entity coding, time-series volatility factor F, and historical anomaly factor Together they form a feature vector, where the entity encoding is a business constraint tuple. The unique identification code is used to generate a standardized early warning work order by incorporating the feature vector, and the early warning work order is output as the early warning result.

[0040] Furthermore, the verification results of the warning information are obtained, and the feature vector of each warning information is labeled with a classification label to construct a feedback sample set, including:

[0041] Obtain the verification result of the warning work order after manual review;

[0042] The feature vector of each early warning work order Labeling with binary categories Among them, positive samples =1 indicates that the verification result is a true anomaly, a negative sample. =0 indicates that the verification result is a normal fluctuation;

[0043] Summarize all marked early warning work orders and construct a standardized feedback sample set. ,in, The total number of samples;

[0044] The feature vector Fully inherits the standardized metrics and the quantitative plural group model from the standardized business dataset. The constraint features and the risk quantification value And the entity code obtained from anomaly tracing.

[0045] Furthermore, based on the feedback sample set, rule optimization is performed to generate an optimized update rule set, including:

[0046] With the standardized feedback sample set As input, the existing rule correction link and the incremental rule mining link are executed synchronously:

[0047] The existing rule correction link includes: constructing a loss function L that converges to global credibility, and using the binary classification labels labeled in the standardized feedback sample set. Compared with model early warning prediction output value Minimizing the error is the optimization objective.

[0048] ;

[0049] The gradient descent algorithm is used to quantize the tuple model. The original constraint threshold in Perform iterative updates:

[0050] ;

[0051] in The threshold values ​​for the original stock rules to be optimized. For the updated threshold, For learning rate, For loss function, For existing rules threshold, For loss function Threshold for existing rules gradient, To determine the sign of the partial derivative, the loop iterates until the stopping condition is met, thus obtaining the adaptively corrected stock rule.

[0052] The incremental rule mining process includes: traversing the standardized feedback sample set. All positive samples in For the quantized tuple model Perform frequent association mining to screen new quantized tuples that meet the support and confidence thresholds. After validity verification, new early warning rules are generated;

[0053] The corrected existing rules output from the existing rule correction link and the newly added early warning rules output from the incremental rule mining link are merged to generate a complete update rule set.

[0054] Furthermore, the updated rule set is loaded to update the dynamic rule base, and the updated rule system is recycled back to the construction stage of the quantized tuple model, so that subsequent newly added standardized business datasets can re-execute business indicator deviation calculation and constraint determination, including:

[0055] Load the complete update rule set and replace the constrained threshold values ​​in the quantized tuple model T. and threshold constraints and rigid and flexible verification rules Supplementing the newly added quantitative multivariate groups ;

[0056] A non-disruptive hot update mechanism is adopted to synchronously push the updated complete rules to the risk inference engine;

[0057] The updated complete quantitative tuple cognitive space is directly returned and reused for the construction of the quantitative tuple model. For the subsequently accessed standardized service data set, the theoretical calculated electricity fee is recalculated by the electricity quantity fee calculation formula according to the complete updated rule set after iterative optimization, and the service index deviation characteristics are calculated and constraint determination.

[0058] Further, continuously collecting performance indicators of the early warning information, adjusting core parameters of the risk potential field reasoning model and rule optimization according to the performance indicators to realize closed-loop adaptive iteration includes:

[0059] continuously collecting the early warning accuracy, false alarm rate, missing alarm rate and rule credibility of the early warning work order as the performance indicators;

[0060] when any of the performance indicators exceeds a preset tolerance interval, a tuning instruction is automatically output, and two-way linkage tuning is performed:

[0061] tuning the weight coefficients of the risk potential field reasoning model , , , , correcting the calculation logic of the risk quantization value;

[0062] tuning the learning rate of the stock rule correction link and the frequency of rule iterative update to control the self-evolution convergence speed;

[0063] the tuned weight coefficients and learning rate parameters are automatically written back to the risk potential field reasoning model and the stock rule correction link, and take effect in the next round of the whole process to form a permanent quantitative optimization closed loop.

[0064] In a second aspect, the present invention provides an intelligent early warning system for electric power marketing quantity and fee data, which is used to implement the intelligent early warning method for electric power marketing quantity and fee data according to any preceding claim, and comprises:

[0065] a preprocessing module, configured to preprocess acquired multi-source original marketing quantity and fee data to generate a standardized service data set;

[0066] a processing module, configured to construct a quantitative tuple model based on the standardized service data set, convert service rules into quantitative constraint logic, calculate service index deviation characteristics, map the service index deviation characteristics to pre-generated constraint conflict intensity for index fusion, generate fused constraint conflict intensity, and generate rule violation degree at the same time;

[0067] The calculation module is used to input the fused constraint conflict strength and rule violation degree, as well as the introduced time-series fluctuation factor and historical anomaly factor, into the pre-built risk potential field inference model to calculate the risk quantification value.

[0068] The early warning module is used to determine the anomaly classification based on the risk quantification value, trace the source of the anomaly based on the entity association relationship in the quantified plural model, generate early warning information with built-in feature vectors, and output the early warning information as the early warning result.

[0069] The annotation module is used to obtain the verification results of the early warning information, annotate the feature vector of each early warning information with a classification label, and construct a feedback sample set;

[0070] The rule optimization module is used to perform rule optimization based on the feedback sample set and generate an optimized updated rule set;

[0071] The update module is used to load the update rule set, update the dynamic rule base, and return the updated rule system to the construction stage of the quantitative plural model for subsequent newly connected standardized business datasets to re-execute business indicator deviation calculation and constraint judgment.

[0072] The performance indicators of the continuously collected early warning information are used to adjust the core parameters of the risk potential field inference model and the rule optimization based on the performance indicators, thereby achieving closed-loop adaptive iteration.

[0073] Thirdly, the present invention provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of any of the methods described above.

[0074] Fourthly, the present invention provides an electronic device, comprising:

[0075] Memory, used to store computer programs / instructions;

[0076] A processor for executing the computer program / instructions to implement the steps of any of the methods described above.

[0077] Fifthly, the present invention provides a computer program product, including a computer program / instructions that, when executed by a processor, implement the steps of any of the methods described above.

[0078] Compared with the prior art, the beneficial effects achieved by the present invention are as follows:

[0079] 1. This invention provides an intelligent early warning method, system, and storage medium for electricity marketing volume and fee data. By standardizing and preprocessing multi-source raw marketing volume and fee data, a unified standardized business dataset is constructed, providing a unique data foundation for all subsequent modeling, reasoning, and evolution steps. On this basis, a business constraint cognitive space is constructed. Through a quantitative plural model, the coupling constraint relationships of business entities such as user profiles, metering points, transformer areas, electricity pricing policies, and accounting models are mathematically expressed. All text-based electricity marketing business rules are transformed into numerically computable quantitative constraint logic. At the same time, core quantitative formulas such as volume and fee accounting and load fluctuations are embedded, enabling the system to have the computational cognitive ability to understand complex electricity marketing business logic. This fundamentally solves the principle defect of traditional technologies that lack the ability to cognitively model business rules.

[0080] 2. This invention uses a risk potential field inference model to comprehensively quantify information such as the intensity of multi-dimensional business constraint conflicts and the degree of rule violation into a unique risk potential value. While relying on rigid rule weights to ensure the accuracy of hard error identification, it also uses soft inference weights to mine hidden anomalies, realizing the quantitative identification of normal business fluctuations and real volume and cost anomalies. Based on a three-stage risk threshold, it realizes standardized hierarchical judgment of general warning, important warning and emergency warning, avoiding the subjective error of human experience judgment. At the same time, it combines the business constraint topology network to reverse track the anomaly propagation path and locate the anomaly source, which greatly improves the accuracy of anomaly identification and the efficiency of source tracing.

[0081] 3. This invention relies on manually reviewed and quantified feedback sample sets to construct a rule credibility convergence loss function. It combines gradient descent algorithm to adaptively correct the threshold of existing rules. At the same time, it automatically mines high-dimensional constraint combinations that frequently co-occur based on positive samples as incremental business rules, realizing fully automatic iterative upgrade of the early warning rule system without human intervention. This completely solves the technical problems of traditional technical rule updates relying entirely on manual labor, long iteration cycles, and strong lag. Attached Figure Description

[0082] Figure 1 This is a flowchart of the intelligent early warning method for electricity marketing volume and fee data provided in an embodiment of the present invention;

[0083] Figure 2 This is a flowchart of a loss function-driven rule self-evolution algorithm provided in an embodiment of the present invention. Detailed Implementation

[0084] The present invention will be further described below with reference to the accompanying drawings. The following embodiments are only used to more clearly illustrate the technical solution of the present invention, and should not be used to limit the scope of protection of the present invention.

[0085] Example 1:

[0086] This embodiment introduces an intelligent early warning method for electricity marketing data, including:

[0087] Preprocess the acquired multi-source raw marketing volume and cost data to generate a standardized business dataset;

[0088] Based on the standardized business dataset, a quantitative plural group model is constructed to transform business rules into quantitative constraint logic, calculate business indicator deviation features, and map the business indicator deviation features to the pre-generated constraint conflict intensity for indicator fusion, generating the fused constraint conflict intensity and simultaneously generating the rule violation degree.

[0089] The fused constraint conflict strength and rule violation degree, along with the introduced time-series fluctuation factor and historical anomaly factor, are input into a pre-built risk potential field inference model to calculate the risk quantification value.

[0090] Anomaly classification is determined based on the risk quantification value, and anomaly source is traced based on the entity association relationship in the quantified plural group model. Early warning information with built-in feature vectors is generated, and the early warning information is output as the early warning result.

[0091] Obtain the verification results of the warning information, label the feature vector of each warning information with a classification label, and construct a feedback sample set;

[0092] Based on the feedback sample set, rule optimization is performed to generate an optimized updated rule set;

[0093] Load the updated rule set, update the dynamic rule base, and reuse the updated rule system back to the construction stage of the quantitative plural model, so that the business indicator deviation calculation and constraint judgment can be re-executed for newly connected standardized business datasets.

[0094] The performance indicators of the continuously collected early warning information are used to adjust the core parameters of the risk potential field inference model and the rule optimization based on the performance indicators, thereby achieving closed-loop adaptive iteration.

[0095] like Figure 1 As shown in the figure, the intelligent early warning method for electricity marketing data provided in this embodiment involves the following steps in its application process:

[0096] S1: Multi-source data cleaning and standardized preprocessing for data volume and cost, outputting standardized business datasets;

[0097] This embodiment aggregates raw business data from multiple sources, including user profiles, time-series meter readings, electricity billing calculations, and transformer area line loss data, to construct a raw business data set D. Each data entry contains fields such as user ID, electricity address, industry classification, supply voltage, approved capacity, electricity price tier, power factor assessment standard, meter number, and transformer area affiliation code. Missing value repair, noise suppression, time-series alignment, unit normalization, and business semantic mapping are uniformly performed on D to eliminate cross-system data heterogeneity and semantic conflicts, generating a standardized business dataset. This standardized business dataset uniformly encapsulates the numerical, time-series, and archival indicators of all business entities, including users, meters, transformer areas, electricity prices, and bills, encompassing actual settled electricity charges. Total power Standard electricity price Power factor This serves as the sole input data source for all subsequent modeling, reasoning, and evolutionary steps.

[0098] S2: Based on standardized business datasets A quantitative multivariate model is constructed to transform business rules into quantitative constraint logic, calculate the deviation characteristics of business indicators, and map the deviation characteristics of business indicators to the pre-generated constraint conflict strength for indicator fusion, generating the fused constraint conflict strength, and generating the rule violation degree at the same time.

[0099] This embodiment is based on a standardized business dataset. Extract business entities Entity Relationship Topology Rigidity and flexibility verification rules and threshold under constraints and the upper threshold of the constraint Constructing a quantitative multivariate model Based on this quantitative plural model, all text-based electricity marketing business rules are transformed into quantitative constraint logic that can be numerically calculated.

[0100] This embodiment will standardize the business dataset. Total electricity consumption during periodic settlement Standard electricity price Power factor Substitute into the quantity charge calculation formula Calculate the theoretical electricity cost Combined with standardized business datasets Actual settlement electricity cost According to the formula Calculate the deviation characteristics of business indicators .

[0101] This embodiment compares each element in a quantitative plural model. Threshold under the constraint of the i-th business constraint and threshold constraints The deviation from the actual business indicator value generates the constraint conflict strength of the i-th term. Traversing rigid and flexible verification rules For rule j, based on whether the current business data meets the rule logic and the severity of the violation, a rule violation degree is generated. This includes the quantitative results of violations across all verification rules, such as electricity price matching, transformer area line loss, load fluctuation, and file consistency. It also includes the characteristics of deviations in business indicators. Mapping to constraint conflict strength By fusing the indicators, the combined constraint conflict strength is obtained. Each business constraint within a multivariate group is associated with a set of initial upper and lower threshold values. , The original initial thresholds in this group that have not undergone evolutionary optimization are uniformly labeled as the original constraint thresholds. The parameters are stored in the quantized tuple model T and used as parameters to be updated in subsequent gradient iteration optimization.

[0102] S3: Input the fused constraint conflict strength and rule violation degree, as well as the introduced time-series fluctuation factor and historical anomaly factor, into the risk inference model to calculate the risk quantification value;

[0103] This embodiment reads the constraint conflict strength after fusion of all elements in the quantized tuple model T. and the degree of rule violation Simultaneously, a multi-layered weighted fusion risk inference model is constructed by introducing a time-series volatility factor F and a historical anomaly factor H derived from entity correlation topology:

[0104] ;

[0105] in, The aforementioned risk quantification value, The strength of the constraint conflict after fusion. As to the degree of rule violation, For time series fluctuation factor, As a historical outlier, For the first item The weighting coefficients, For the first Degree of violation of the rule The weighting coefficients, Time series fluctuation factor The weighting coefficients, Historical outlier The weighting coefficients, It is the total number of constraint conflict strengths. It represents the total number of rule violations; each weight coefficient is based on the rigidity and flexibility verification rules in the quantized plural model T. The attributes are assigned adaptively. The final output is a unique risk quantification value for each data point. This serves as the sole basis for subsequent anomaly classification determination.

[0106] S4: Based on the risk quantification value, anomaly classification is determined, and the anomaly source is traced according to the entity relationship in the quantified multivariate model. The warning information with built-in feature vectors is generated and the warning information is output as the warning result.

[0107] This embodiment is based on three preset risk thresholds. , , According to risk quantification value The warning levels are divided into intervals:

[0108] when ≤ < At that time, it was determined to be a general warning;

[0109] when ≤ < At that time, it was determined to be an important warning;

[0110] when ≥ At that time, it was determined to be an emergency warning.

[0111] Simultaneously based on the quantitative plural model Entity association topology Tracing the anomaly propagation path in reverse reveals the source entity, conflict constraints, and scope of impact. This includes analyzing early warning levels and deviations in business metrics. Risk quantification value Entity coding, time-series volatility factor F, and historical anomaly factor Together, they form a feature vector, where the entity code is the unique identity code of the business constraint tuple T. The built-in feature vector generates a standardized early warning work order, which is then pushed to the manual review stage. This step outputs an early warning work order with complete multi-dimensional features. All work order data flows into subsequent stages to construct a quantitative feedback sample set for manual review.

[0112] S5: Obtain the verification results of the early warning information, label the feature vector of each early warning information with a classification label, and construct a feedback sample set;

[0113] This embodiment receives all early warning work orders. After offline review by inspectors, the feature vector of each early warning work order is analyzed. Labeling with binary categories Among them, positive samples =1 indicates that the verification result is a true anomaly, a negative sample. =0 indicates that the verification result is within the normal range.

[0114] Summarize all marked early warning work orders and construct a standardized feedback sample set. ,in, The total number of samples.

[0115] The feature vector Fully inherits standardized metrics and quantitative plural models from standardized business datasets. Constraint characteristics and risk quantification values ​​in And entity codes obtained from anomaly tracing. This sample set provides all the training and mining data sources for subsequent rule optimization using a dual-parallel link.

[0116] S6: Based on the feedback sample set, perform rule optimization and generate an optimized update rule set.

[0117] This embodiment uses a standardized feedback sample set. As input, the existing rule correction link and the incremental rule mining link are executed simultaneously, and the two links share the sample set. Data support:

[0118] In the existing rule correction process, a global credibility convergence loss function L is constructed to standardize the binary classification labels labeled in the feedback sample set. Compared with model early warning prediction output value Minimizing the error is the optimization objective.

[0119] ;

[0120] Gradient descent algorithm is used to quantize the tuple model. The original constraint threshold in Perform iterative updates:

[0121] ;

[0122] in The threshold values ​​for the original stock rules to be optimized. For the updated threshold, For learning rate, For loss function, For existing rules threshold, For loss function Threshold for existing rules gradient, To determine the sign of the partial derivative, the process iterates until the stopping condition is met, resulting in the adaptively corrected stock rule.

[0123] Traversing the standardized feedback sample set in the incremental rule mining process All positive samples in For the quantized tuple model Perform frequent association mining to screen new quantized tuples that meet the support and confidence thresholds. After validity verification, new early warning rules are generated.

[0124] The corrected existing rules output from the existing rule correction link and the newly added early warning rules output from the incremental rule mining link are merged to generate a complete update rule set.

[0125] S7: Load and update the rule set, update the dynamic rule base, and return the updated rule system to the construction stage of the quantitative plural model for subsequent new standardized business datasets to re-execute business indicator deviation calculation and constraint judgment;

[0126] This embodiment loads the complete update rule set and replaces the constrained threshold values ​​within the quantized tuple model T. and threshold constraints and rigid and flexible verification rules Supplementing the newly added quantitative multivariate groups .

[0127] A non-disruptive hot update mechanism is adopted to synchronously push the updated complete rules to the risk inference engine. The updated complete quantitative versine cognitive space is directly recycled back to the quantitative versine model construction stage. Subsequently added standardized business datasets will be re-executed based on the iteratively optimized complete updated rule set to calculate theoretical electricity costs using the quantity and cost accounting formula and to calculate business indicator deviation characteristics. And constraint determination, forming a rule iterative loop, such as Figure 2 As shown.

[0128] S8: Continuously collect performance indicators of early warning information, and adjust the core parameters of risk inference model and rule optimization based on performance indicators to achieve closed-loop adaptive iteration;

[0129] This embodiment continuously collects four quantitative performance indicators from early warning work orders: early warning accuracy, false alarm rate, missed alarm rate, and rule reliability. When any performance indicator exceeds the preset tolerance range, an optimization command is automatically output for bidirectional optimization: optimizing the weight coefficients of the risk inference model. , , , Correcting the risk quantification value The computational logic is optimized; the learning rate η and rule iteration update frequency of the existing rule correction link are tuned to control the self-evolution convergence speed.

[0130] The optimized weight coefficients and learning rate parameters are automatically written back to the risk inference model and the existing rule correction loop, taking effect in the next round of the entire process. This step optimizes the core computational parameters of the risk inference model and rule optimization through performance feedback, allowing the entire early warning process to continuously adapt and iterate, forming a permanent quantitative optimization closed loop.

[0131] Example 2:

[0132] This embodiment provides an intelligent early warning system for electricity marketing data, including:

[0133] The preprocessing module is used to preprocess the acquired multi-source raw marketing volume and cost data to generate standardized business datasets;

[0134] The processing module is used to construct a quantitative plural model based on the standardized business dataset, transform business rules into quantitative constraint logic, calculate business indicator deviation features, map the business indicator deviation features to the pre-generated constraint conflict intensity for indicator fusion, generate the fused constraint conflict intensity, and generate the rule violation degree.

[0135] The calculation module is used to input the fused constraint conflict strength and rule violation degree, as well as the introduced time-series fluctuation factor and historical anomaly factor, into the pre-built risk potential field inference model to calculate the risk quantification value.

[0136] The early warning module is used to determine the anomaly classification based on the risk quantification value, trace the source of the anomaly based on the entity association relationship in the quantified plural model, generate early warning information with built-in feature vectors, and output the early warning information as the early warning result.

[0137] The annotation module is used to obtain the verification results of the early warning information, annotate the feature vector of each early warning information with a classification label, and construct a feedback sample set;

[0138] The rule optimization module is used to perform rule optimization based on the feedback sample set and generate an optimized updated rule set;

[0139] The update module is used to load the update rule set, update the dynamic rule base, and return the updated rule system to the construction stage of the quantitative plural model for subsequent newly connected standardized business datasets to re-execute business indicator deviation calculation and constraint judgment.

[0140] The performance indicators of the continuously collected early warning information are used to adjust the core parameters of the risk potential field inference model and the rule optimization based on the performance indicators, thereby achieving closed-loop adaptive iteration.

[0141] The specific functions of each module described above are explained in the relevant content of the method in Embodiment 1, and will not be repeated here.

[0142] Example 3:

[0143] This embodiment provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the method described in Embodiment 1.

[0144] Example 4:

[0145] This embodiment provides an electronic device, including:

[0146] Memory, used to store computer programs / instructions;

[0147] A processor for executing the computer program / instructions to implement the steps of the method described in Embodiment 1.

[0148] Example 5:

[0149] This embodiment provides a computer program product, including a computer program / instructions, which, when executed by a processor, implement the steps of the method described in Embodiment 1.

[0150] The above description is only a preferred embodiment of the present invention. It should be noted that for those skilled in the art, several improvements and modifications can be made without departing from the technical principles of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention.

[0151] Those skilled in the art will understand that embodiments of this disclosure can be provided as methods, systems, or computer program products. Therefore, this disclosure can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this disclosure can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0152] This disclosure is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this disclosure. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, create a machine for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0153] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0154] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0155] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of this disclosure and not to limit its protection scope. Although this disclosure has been described in detail with reference to the above embodiments, those skilled in the art should understand that after reading this disclosure, they can still make various changes, modifications or equivalent substitutions to the specific implementation of the invention, but these changes, modifications or equivalent substitutions are all within the protection scope of the pending claims.

Claims

1. A method for intelligent early warning of electricity marketing volume and fee data, characterized in that, include: Preprocess the acquired multi-source raw marketing volume and cost data to generate a standardized business dataset; Based on the standardized business dataset, a quantitative plural group model is constructed to transform business rules into quantitative constraint logic, calculate business indicator deviation features, and map the business indicator deviation features to the pre-generated constraint conflict intensity for indicator fusion, generating the fused constraint conflict intensity and simultaneously generating the rule violation degree. The fused constraint conflict strength and rule violation degree, along with the introduced time-series fluctuation factor and historical anomaly factor, are input into a pre-built risk potential field inference model to calculate the risk quantification value. Anomaly classification is determined based on the risk quantification value, and anomaly source is traced based on the entity association relationship in the quantified plural group model. Early warning information with built-in feature vectors is generated, and the early warning information is output as the early warning result. Obtain the verification results of the warning information, label the feature vector of each warning information with a classification label, and construct a feedback sample set; Based on the feedback sample set, rule optimization is performed to generate an optimized updated rule set; Load the updated rule set, update the dynamic rule base, and reuse the updated rule system back to the construction stage of the quantitative plural model, so that the business indicator deviation calculation and constraint judgment can be re-executed for newly connected standardized business datasets. The performance indicators of the continuously collected early warning information are used to adjust the core parameters of the risk potential field inference model and the rule optimization based on the performance indicators, thereby achieving closed-loop adaptive iteration.

2. The intelligent early warning method for electricity marketing volume and fee data according to claim 1, characterized in that, The multi-source raw marketing data includes user profiles, time-series meter readings, electricity billing, and transformer area line loss data; the preprocessing includes missing value repair, noise suppression, time-series alignment, unit normalization, and business semantic mapping.

3. The intelligent early warning method for electricity marketing volume and fee data according to claim 1, characterized in that, Based on the standardized business dataset, a quantitative tuple model is constructed to transform business rules into quantitative constraint logic. Business indicator deviation features are calculated, and these features are mapped to pre-generated constraint conflict strengths for indicator fusion, generating a fused constraint conflict strength. Simultaneously, the degree of rule violation is generated, including: Based on the standardized business dataset, extract business entities. Entity Relationship Topology Rigidity and flexibility verification rules and threshold under constraints and the upper threshold of the constraint Construct a quantitative plural group model: ； Based on the aforementioned quantitative plural model, all text-based electricity marketing business rules are transformed into quantitative constraint logic that can be numerically calculated. The total periodic settlement electricity in the standardized business dataset Standard unit price and power factor adjustment coefficient Substitute the data into the data usage calculation formula to calculate the theoretical electricity cost. The formula is as follows: ； Combined with the actual settlement electricity charges in the standardized business dataset Calculate the deviation characteristics of business indicators The formula is as follows: ； The quantitative tuple model was compared item by item. The Middle Threshold under the constraints of business items and threshold constraints The deviation from the actual business indicator value generates the first Constraint Conflict Strength ; Traversing the The Middle Each rule assigns a value based on whether the current business data meets the rule logic and the severity of the violation, generating a rule violation level. ; The deviation characteristics of the business indicators Mapped to the constraint conflict strength By fusing the indicators, the combined constraint conflict strength is obtained. .

4. The intelligent early warning method for electricity marketing volume and fee data according to claim 3, characterized in that, The fused constraint conflict strength and rule violation degree, along with the introduced time-series fluctuation factor and historical anomaly factor, are input into a pre-constructed risk potential field inference model to calculate the risk quantification value. The calculation formula of the risk potential field inference model is as follows: ； in, The aforementioned risk quantification value, The strength of the constraint conflict after fusion. As to the degree of rule violation, For time series fluctuation factor, As a historical outlier, For the first item The weighting coefficients, For the first Degree of violation of the rule The weighting coefficients, Time series fluctuation factor The weighting coefficients, Historical outlier The weighting coefficients, It is the total number of constraint conflict strengths. It is the total number of rule violations; Anomaly classification is determined based on the risk quantification value, and anomaly tracing is performed based on the entity association relationships in the quantified tuple model. Early warning information with built-in feature vectors is generated, and the early warning information is output as the warning result, including: Based on the preset three risk thresholds , , According to the aforementioned risk quantification value The warning levels are divided into intervals: when ≤ < At that time, it was determined to be a general warning; when ≤ < At that time, it was determined to be an important warning; when ≥ At that time, it was determined to be an emergency warning; According to the quantized tuple model Entity association topology Tracing the abnormal propagation path in reverse, locating the abnormal source entity, conflict constraints, and scope of impact; The warning level and the deviation characteristics of the business indicators Risk quantification value Entity coding, time-series volatility factor F, and historical anomaly factor Together they form a feature vector, where the entity encoding is a business constraint tuple. The unique identification code is used to generate a standardized early warning work order by incorporating the feature vector, and the early warning work order is output as the early warning result.

5. The intelligent early warning method for electricity marketing volume and fee data according to claim 4, characterized in that, Obtain the verification results of the warning information, label the feature vector of each warning information with a classification label, and construct a feedback sample set, including: Obtain the verification result of the warning work order after manual review; The feature vector of each early warning work order Labeling with binary categories Among them, positive samples =1 indicates that the verification result is a true anomaly, a negative sample. =0 indicates that the verification result is a normal fluctuation; Summarize all marked early warning work orders and construct a standardized feedback sample set. ,in, The total number of samples; The feature vector Fully inherits the standardized metrics and the quantitative plural group model from the standardized business dataset. The constraint features and the risk quantification value And the entity code obtained from anomaly tracing.

6. The intelligent early warning method for electricity marketing volume and fee data according to claim 5, characterized in that, Based on the feedback sample set, rule optimization is performed to generate an optimized update rule set, including: With the standardized feedback sample set As input, the existing rule correction link and the incremental rule mining link are executed synchronously: The existing rule correction link includes: constructing a loss function L that converges to global credibility, and using the binary classification labels labeled in the standardized feedback sample set. Compared with model early warning prediction output value Minimizing the error is the optimization objective. ； The gradient descent algorithm is used to quantize the tuple model. The original constraint threshold in Perform iterative updates: ； in The threshold values ​​for the original stock rules to be optimized. For the updated threshold, For learning rate, For loss function, For existing rules threshold, For loss function Threshold for existing rules gradient, To determine the sign of the partial derivative, the loop iterates until the stopping condition is met, thus obtaining the adaptively corrected stock rule. The incremental rule mining process includes: traversing the standardized feedback sample set. All positive samples in For the quantized tuple model Perform frequent association mining to screen new quantized tuples that meet the support and confidence thresholds. After validity verification, new early warning rules are generated; merging the corrected existing rules output by the existing rule correction link and the new early warning rules output by the incremental rule mining link to generate a complete updated rule set.

7. The intelligent early warning method for electricity marketing volume and fee data according to claim 6, characterized in that, loading the updated rule set, updating a dynamic rule base, and回流 reusing the updated rule system to the construction link of the quantitative tuple model for a subsequently accessed standardized service data set to re-execute service index deviation calculation and constraint judgment, comprising: Load the complete update rule set and replace the constrained threshold values ​​in the quantized tuple model T. and threshold constraints and rigid and flexible verification rules Supplementing the newly added quantitative multivariate groups ; adopting a non-stop hot update mechanism to synchronously push the updated complete rules to a risk reasoning engine; The updated complete quantitative tuple cognitive space is directly recycled back to the construction stage of the quantitative tuple model. Subsequently added standardized business datasets will be re-executed based on the iteratively optimized complete updated rule set to calculate the theoretical electricity cost using the quantity-based billing formula and to calculate the deviation characteristics of business indicators. and constraint determination.

8. The intelligent early warning method for electricity marketing volume and fee data according to claim 7, characterized in that, continuously collecting performance indicators of the early warning information, adjusting core parameters of the risk potential field reasoning model and the rule optimization according to the performance indicators, so as to realize closed-loop adaptive iteration, comprising: continuously collecting early warning accuracy, false alarm rate, missing alarm rate and rule credibility of early warning work orders as the performance indicators; when any of the performance indicators exceeds a preset tolerance interval, automatically outputting a tuning instruction to perform two-way linkage tuning: Optimize the weight coefficients of the risk potential field inference model. , , , Correct the aforementioned risk quantification value The computational logic; Optimize the learning rate of the existing rule correction link. The frequency of rule iteration updates controls the self-evolution convergence speed; the tuned weight coefficient and learning rate parameter are automatically written back to the risk potential field reasoning model and the existing rule correction link, and take effect in the next round of the whole process, forming a permanent quantitative optimization closed loop.

9. An intelligent early warning system for electricity marketing volume and fee data, characterized in that, comprising: a preprocessing module, configured to preprocess acquired multi-source original marketing expense data to generate a standardized service data set; a processing module, configured to construct a quantitative tuple model based on the standardized service data set, convert service rules into quantitative constraint logic, calculate service index deviation characteristics, map the service index deviation characteristics to a pre-generated constraint conflict intensity for index fusion to generate fused constraint conflict intensity, and generate a rule violation degree at the same time; a calculation module, configured to input the fused constraint conflict intensity and the rule violation degree, as well as an introduced time sequence fluctuation factor and a historical anomaly factor into a pre-constructed risk potential field reasoning model to calculate a quantitative risk value; an early warning module, configured to perform anomaly grading judgment according to the quantitative risk value, perform anomaly tracing according to entity association relationships in the quantitative tuple model, generate early warning information with built-in feature vectors, and output the early warning information as an early warning result; an annotation module, configured to acquire a verification result of the early warning information, annotate classification labels on the feature vectors of each piece of early warning information, and construct a feedback sample set; a rule optimization module, configured to perform rule optimization based on the feedback sample set to generate an optimized updated rule set; an update module, configured to load the updated rule set, update a dynamic rule base, and回流 reuse the updated rule system to the construction link of the quantitative tuple model for a subsequently accessed standardized service data set to re-execute service index deviation calculation and constraint judgment; continuously collecting performance indicators of the early warning information, adjusting core parameters of the risk potential field reasoning model and the rule optimization according to the performance indicators, and realizing closed-loop adaptive iteration.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: when the program is executed by a processor, the steps of the method according to any one of claims 1-8 are implemented.

Images