Inverter manufacturer original state and standard state automatic mapping model and management method

Abstract

The application discloses an inverter manufacturer original state and standard state automatic mapping model and a management method, relates to the technical field of data mapping, and comprises a multi-decision tree classification layer, a rule engine matching layer, a confidence degree arbitration layer and a dynamic updating layer. The method establishes or imports an inverter standard state system, acquires original state data of inverters of multiple manufacturers, and constructs a state feature vector containing code values, texts, alarms, measuring points and time sequence features; the multi-decision tree classification layer is used to call a candidate standard state set and a candidate rule set; multi-evidence matching is performed through the rule engine matching layer; the confidence degree arbitration layer is used to determine a final standard state in combination with a comprehensive confidence degree, a state priority and a state transition constraint; when the confidence degree is insufficient or a rule is not hit, the dynamic updating layer is used to generate a candidate updating rule and update a rule library and a sample library. The application can improve the accuracy, expandability and traceability of standardization mapping of state data of inverters of multiple manufacturers.

Description

Technical Field

[0001] This invention relates to the field of data mapping technology, specifically to an automatic mapping model and management method for the original state and standard state of inverter manufacturers. Background Technology

[0002] With the large-scale construction of photovoltaic power plants, energy storage systems, and integrated energy stations, new energy asset management platforms, power plant monitoring platforms, and centralized operation and maintenance platforms need to collect, display, and analyze the operating status of a large number of inverter devices in a unified manner. As a key device in new energy power generation systems, the status data of inverters directly affects fault early warning, operation and maintenance scheduling, and asset management.

[0003] In the existing technology, inverters from different manufacturers, models, and firmware versions typically use different status code definitions, status texts, alarm codes, and measurement point naming rules. The same operating state may correspond to different original status codes in different manufacturers' devices, and the same original status code may also have different meanings in different models or firmware versions, making it difficult to standardize the status data of inverters from multiple manufacturers in a unified platform.

[0004] The common practice is to manually maintain mapping tables, protocol parsing scripts, or configure fixed rules to convert the manufacturer's original status codes into the platform's standard status codes. However, this method relies on manual configuration, and rules need to be re-maintained when adding manufacturers, device models, firmware versions, or status codes, which is inefficient and prone to problems such as omissions, misconfigurations, and inconsistencies.

[0005] Furthermore, existing mapping methods typically rely solely on the original status code or status text for judgment, lacking joint analysis of alarm information, associated measurement point data, and status time-series changes. Incorrect mappings can easily occur when the status code is inconsistent with measurement point information such as actual power, grid connection flag, power limit flag, or communication quality. Simultaneously, existing rule bases lack confidence arbitration, state transition constraints, and version tracking mechanisms, making it difficult to handle multi-rule conflicts, abnormal state transitions, and the consistency of historical data after rule updates.

[0006] Therefore, it is necessary to provide an automatic mapping model and management method for the original state and standard state of inverter manufacturers, so as to realize the automatic mapping, dynamic updating and traceable management of the original state to the standard state of inverters from multiple manufacturers. Summary of the Invention

[0007] To address the aforementioned technical issues, this technical solution provides an automatic mapping model and management method for the original and standard states of inverter manufacturers. This solution resolves the problem that existing multi-manufacturer inverter original state data is difficult to automatically, accurately, scalably, and traceably map to a unified standard state.

[0008] To achieve the above objectives, the technical solution adopted by the present invention is as follows:

[0009] An automatic mapping model and management method for the original state and standard state of inverter manufacturers, wherein the automatic mapping model includes a multi-decision tree classification layer, a rule engine matching layer, a confidence level arbitration layer, and a dynamic update layer; the management method includes:

[0010] S1. Establish or import the inverter standard state system, which includes standard state codes, standard state names, state priorities, and state transition constraints.

[0011] S2. Obtain original status data of inverter manufacturers from multiple manufacturers. The data includes manufacturer identification, model version, original status code, status text, alarm information, associated measurement point data, and timestamp.

[0012] S3. Standardize the format and handle anomalies of the manufacturer's original status data, and construct a status feature vector that includes code value, text, alarm, measurement point and time sequence features;

[0013] S4. Input the state feature vector into the automatic mapping model and perform multi-decision tree classification, rule engine matching, confidence arbitration and dynamic update in the order of coarse screening, fine matching, arbitration and evolution.

[0014] S5. The multi-decision tree determines the leaf node or classification label based on the manufacturer type, model version and status type, and retrieves the candidate standard status set and candidate rule set bound to it.

[0015] S6. Perform multi-evidence matching on the candidate rule set to generate candidate mapping results. Determine the final standard state according to the comprehensive confidence level, state priority and state transition constraints. Reduce the weight or mark the results that violate the state transition constraints as abnormal.

[0016] S7. When the overall confidence level is lower than the threshold or no candidate rule set is hit, candidate update rules are generated based on unknown state records and similar historical mapping samples. After confirmation, the rule base and sample base are updated, and standardized state data containing the final standard state, rule number, confidence level, occurrence time and version number are output.

[0017] Preferably, in step S1, the inverter standard state system includes:

[0018] Standard state classification field, standard state code field, standard state name field, state priority field, allowed preceding state field, allowed subsequent state field, state mutual exclusion constraint field, and state determination description field;

[0019] The standard states include at least normal operation, standby, alarm operation, fault shutdown, power-limited operation, communication interruption, and maintenance shutdown.

[0020] Preferably, in step S2, the associated measurement point data includes:

[0021] At least one of the following: DC voltage, DC current, AC output power, grid connection indicator, power limit indicator, inverter temperature, and communication quality indicator;

[0022] The associated measurement point data and the original status code are associated according to the same inverter number and the same sampling time window.

[0023] Preferably, in step S3, constructing the state feature vector includes:

[0024] The original status code is parsed to generate code value features;

[0025] Keyword extraction is performed on status text and alarm information to generate text features and alarm features;

[0026] The associated measurement point data is normalized to generate measurement point features;

[0027] And generate timing features based on the previous standard state, state duration, and state transition interval;

[0028] When there is no previous standard state, the timing features are generated with a preset initial state.

[0029] Preferably, in step S4, the multi-decision tree classification layer includes:

[0030] Cascaded decision trees for manufacturer identification, model / version identification, and status type identification;

[0031] Among them, the manufacturer identification decision tree is used to determine the manufacturer type, the model version identification decision tree is used to determine the inverter model version under the manufacturer type constraint, and the status type identification decision tree is used to determine the status type corresponding to the manufacturer's original status data under the manufacturer type and model version constraints.

[0032] Preferably, in step S5, the leaf node or classification label is pre-bound to the candidate standard state set and candidate rule set number in the rule index table; the multi-decision tree classification layer queries the rule index table according to the leaf node or classification label to retrieve the corresponding candidate standard state set and candidate rule set.

[0033] Preferably, in step S6, the candidate rule set includes at least one of the following: code value direct mapping rule, status keyword matching rule, alarm information matching rule, status code and associated measurement point joint judgment rule, and status code, alarm information and associated measurement point joint judgment rule.

[0034] Preferably, in step S6, the multi-evidence matching includes:

[0035] The code value matching score, text matching score, alarm matching score, measurement point consistency score, and state transition consistency score of the candidate mapping results are calculated respectively, and a weighted calculation is performed based on each score and its corresponding weight to generate the comprehensive confidence of the candidate mapping results.

[0036] The corresponding weights are determined based on the manufacturer type, status type, or historical mapping accuracy.

[0037] Preferably, in step S6, the state transition constraint is represented by a state transition matrix or a pre- and post-order state relationship table; when the state transition corresponding to the candidate mapping result does not satisfy the state transition constraint, the overall confidence of the candidate mapping result is reduced, or the candidate mapping result is marked as an abnormal mapping result.

[0038] Preferably, in step S7, the unknown state record includes:

[0039] Manufacturer identification, model version, original status code, status text, alarm information, associated measurement point data, candidate standard status, and generation time;

[0040] The dynamic update layer generates candidate update rules based on the similarity between unknown state records and historical mapping samples;

[0041] When the similarity is higher than the preset similarity threshold and the associated measurement point data meets the judgment conditions of the candidate standard state, or after manual confirmation, the candidate update rule is written into the rule base and the corresponding unknown state record is written into the training sample base.

[0042] The rule records in the rule base include rule number, applicable manufacturer, applicable model version, matching conditions, target standard status, effective time, and rule version number.

[0043] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0044] This invention establishes or imports a standard inverter state system and combines it with a multi-decision tree classification layer, a rule engine matching layer, a confidence arbitration layer, and a dynamic update layer to form a "coarse screening—fine matching—arbitration—evolution" state mapping process. This solves the problems of inconsistent state mapping standards, high maintenance costs, and weak scalability caused by the lack of uniformity in state codes, state texts, alarm information, and measurement point rules among inverters from different manufacturers. By performing multi-evidence matching on the original state codes, state texts, alarm information, associated measurement point data, and time-series features, and combining comprehensive confidence, state priority, and state transition constraints to determine the final standard state, this invention solves the problems of easy misjudgment in single state code mapping, conflicting results of multiple rule hits, and difficulty in handling abnormal state transitions. Through the update mechanism of unknown state records, candidate update rules, rule base, and training sample base, it can adapt to new manufacturers, new model versions, and new state codes. By using mapping rule numbers and mapping version numbers to achieve mapping result traceability, it improves the accuracy, interpretability, and maintainability of standardized processing of state data from inverters from multiple manufacturers. Attached Figure Description

[0045] Figure 1 This is a flowchart of the method of the present invention. Detailed Implementation

[0046] The following description is intended to disclose the invention and enable those skilled in the art to implement it. The preferred embodiments described below are merely examples, and other obvious variations will occur to those skilled in the art.

[0047] This invention provides an automatic mapping model and management method for the original and standard states of inverters from different manufacturers. It is applicable to the standardized processing of status data from inverters from multiple manufacturers in photovoltaic power plants, wind farm energy storage systems, new energy asset management platforms, power plant monitoring platforms, or centralized operation and maintenance platforms. The method automatically maps the original status codes, status texts, alarm information, and associated measurement point data output by inverters from different manufacturers, models, and firmware versions into unified standard status data.

[0048] In this invention, the automatic mapping model includes a multi-decision tree classification layer, a rule engine matching layer, a confidence arbitration layer, and a dynamic update layer. The multi-decision tree classification layer is used for coarse screening based on manufacturer type, model version, and status type, narrowing down the rule matching range. The rule engine matching layer is used for precise matching of multiple pieces of evidence within the candidate rule set. The confidence arbitration layer is used to determine the final standard status among multiple candidate mapping results. The dynamic update layer is used to handle unknown status codes and update the rule base and training sample base. These layers are executed in the order of coarse screening, fine matching, arbitration, and evolution, forming a continuous data processing pipeline, rather than a stack of independent functions.

[0049] Reference Figure 1As shown, an automatic mapping model and management method for the original state and standard state of inverter manufacturers are presented. The automatic mapping model includes a multi-decision tree classification layer, a rule engine matching layer, a confidence level arbitration layer, and a dynamic update layer. The management method includes:

[0050] S1. Establish or import the inverter standard state system, which includes standard state codes, standard state names, state priorities, and state transition constraints.

[0051] S2. Obtain original status data of inverter manufacturers from multiple manufacturers. The data includes manufacturer identification, model version, original status code, status text, alarm information, associated measurement point data, and timestamp.

[0052] S3. Standardize the format and handle anomalies of the manufacturer's original status data, and construct a status feature vector that includes code value, text, alarm, measurement point and time sequence features;

[0053] S4. Input the state feature vector into the automatic mapping model and perform multi-decision tree classification, rule engine matching, confidence arbitration and dynamic update in the order of coarse screening, fine matching, arbitration and evolution.

[0054] The four processing layers described above have a clear data flow relationship. The multi-decision tree classification layer takes the state feature vector as input and outputs leaf nodes or classification labels; the rule engine matching layer retrieves the candidate rule set based on the leaf nodes or classification labels and outputs the candidate mapping results; the confidence arbitration layer resolves conflicts in the candidate mapping results and outputs the final standard state; the dynamic update layer generates candidate update rules when the confidence is insufficient or the rule is not hit, which is used to expand the rule base and training sample base.

[0055] S5. The multi-decision tree determines the leaf node or classification label based on the manufacturer type, model version and status type, and retrieves the candidate standard status set and candidate rule set bound to it.

[0056] S6. Perform multi-evidence matching on the candidate rule set to generate candidate mapping results. Determine the final standard state according to the comprehensive confidence level, state priority and state transition constraints. Reduce the weight or mark the results that violate the state transition constraints as abnormal.

[0057] S7. When the overall confidence level is lower than the threshold or no candidate rule set is hit, candidate update rules are generated based on unknown state records and similar historical mapping samples. After confirmation, the rule base and sample base are updated, and standardized state data containing the final standard state, rule number, confidence level, occurrence time and version number are output.

[0058] Through the above steps, this invention forms a continuous processing flow of "coarse screening—fine matching—arbitration—evolution". Specifically, the multi-decision tree classification layer is used to narrow down the rule matching range, the rule engine matching layer is used to improve mapping accuracy, the confidence arbitration layer is used to resolve conflicts between multiple candidate mapping results, and the dynamic update layer is used to adapt to new manufacturers, new model versions, or new status codes.

[0059] In step S1, the inverter standard state system includes:

[0060] Standard state classification field, standard state code field, standard state name field, state priority field, allowed preceding state field, allowed subsequent state field, state mutual exclusion constraint field, and state determination description field;

[0061] Among them, the standard state classification field is used to indicate the category to which the standard state belongs; the standard state code field is used to indicate the unified code; the standard state name field is used to indicate the unified state name; the state priority field is used to indicate the processing order when a conflict occurs; the allowed preceding state field and the allowed subsequent state field are used to limit the transition relationship between standard states; the state mutual exclusion constraint field is used to limit the states that cannot be established at the same time window; and the state judgment description field is used to record the judgment basis of the standard state.

[0062] The standard states include at least normal operation, standby, alarm operation, fault shutdown, power-limited operation, communication interruption, and maintenance shutdown. For example, the criteria for normal operation can be: the inverter's grid connection flag is valid, the AC output power is greater than the preset operating power threshold, and there are no high-priority fault alarms. The criteria for standby can be: the inverter is not connected to the grid or the AC output power is lower than the preset operating power threshold, and there are no fault shutdown alarms. The criteria for power-limited operation can be: the inverter's grid connection flag is valid, a power-limiting flag is present, or the AC output power is lower than the current available power threshold. The criteria for communication interruption can be: the communication quality flag is abnormal or no valid data is received for several consecutive sampling periods.

[0063] In a preferred embodiment, the status priorities, from highest to lowest, are: fault shutdown, communication interruption, maintenance shutdown, alarm operation, power-limited operation, standby, and normal operation. This priority can be adjusted according to project management requirements.

[0064] Furthermore, state transition constraints can be represented by a state transition matrix or a table of preceding and following states. The state transition matrix can be denoted as M, where M(i,j)=1 indicates that standard state i is allowed to transition to standard state j, and M(i,j)=0 indicates that standard state i is not allowed to directly transition to standard state j. For example, normal operation can transition to alarm operation, power-limited operation, or fault shutdown; maintenance shutdown usually does not directly transition to fault shutdown, but first transitions to standby or normal operation. State transition constraints can prevent unreasonable standard state jumps caused by a single abnormal sampling.

[0065] Furthermore, in step S2, the manufacturer's original status data can be obtained from the inverter communication interface, the power plant SCADA system, the energy management system, the data acquisition device, the edge gateway, or the new energy asset management platform.

[0066] The manufacturer's original status data includes manufacturer identifier, model version, original status code, status text, alarm information, associated measurement point data, and timestamp. The manufacturer identifier distinguishes different inverter manufacturers; the model version distinguishes different device models or firmware versions within the same manufacturer; the original status code can be decimal, hexadecimal, character encoding, or a combination of these; the status text can include text such as "Running," "Standby," "Fault," "Alarm," "Power Limit," and "Communication Anomaly"; the alarm information can include alarm code, alarm name, alarm level, and alarm occurrence time.

[0067] In step S2, the associated measurement point data includes:

[0068] At least one of the following: DC voltage, DC current, AC output power, grid connection indicator, power limit indicator, inverter temperature, and communication quality indicator;

[0069] The associated measurement point data and the original status code are associated according to the same inverter number and the same sampling time window. The sampling time window can be set from 1 minute to 10 minutes, preferably 5 minutes. When multiple measurement point data exist within the same sampling time window, the average, maximum, minimum, or last value can be used as the measurement point characteristic value of that window. For example, for AC output power, the average power within the sampling window is used; for alarm information, the highest alarm level within the sampling window is used; for grid connection flags and power limit flags, the flag value that appears most frequently within the sampling window is used.

[0070] Furthermore, in step S3, the original status data of the manufacturer is formatted and anomalies are handled.

[0071] The format standardization includes: converting the original status codes from different manufacturers into a preset encoding format; converting the status text into a preset character encoding; converting the timestamps into the same time format; and converting the associated measurement point data into a unified unit and a unified dimension.

[0072] Anomaly handling includes missing data handling, duplicate data handling, and outlier handling. For missing data, if the missing time is less than a preset missing threshold, forward padding, linear interpolation, or neighboring values ​​within the same window are used for compensation; if the missing time is greater than or equal to the preset missing threshold, the data is marked as low-confidence data. For duplicate data, deduplication is performed according to the inverter number, timestamp, and original status code. For outliers, if the measurement point value exceeds the physical range of the equipment or the rate of change exceeds a preset change threshold, the measurement point is marked as an outlier, and its influence weight is reduced in subsequent measurement point consistency scoring.

[0073] In step S3, constructing the state feature vector includes:

[0074] The original status code is parsed to generate code value features;

[0075] Keyword extraction is performed on status text and alarm information to generate text features and alarm features;

[0076] The associated measurement point data is normalized to generate measurement point features;

[0077] And generate timing features based on the previous standard state, state duration, and state transition interval;

[0078] When there is no previous standard state, the timing features are generated with a preset initial state.

[0079] In a preferred embodiment, the state feature vector is represented as:

[0080]

[0081] in, Represents the state feature vector; Indicates code value characteristics; Representing text features; Indicates alarm characteristics; Indicates the characteristics of the measuring point; It represents the temporal characteristics.

[0082] Code value features This can include the original status code value, status code bit range, status code length, status code radix, and whether it belongs to a known code range. Text features This can include keywords such as running, standby, alarm, fault, power limit, and communication from the status text. Alarm characteristics. This can include alarm level, alarm type, and alarm keywords. Measurement point characteristics. This can include the ratio of AC output power to rated power, whether the DC voltage has reached the start-up threshold, whether the grid connection flag is valid, whether the power limit flag is valid, and whether the communication quality is normal. Timing characteristics. It can include the previous standard state, the duration of the current state, and the time interval between two state transitions.

[0083] The ratio of AC output power to rated power is calculated as follows:

[0084] :

[0085] in, This indicates the percentage of AC output power to rated power. This indicates the AC output power within the sampling time window; This indicates the rated power of the inverter.

[0086] when When the power output exceeds the first power threshold and the grid connection flag is valid, it indicates that the inverter is in a grid-connected power generation state; when When the power threshold is less than the second power threshold and the grid connection flag is invalid, it indicates that the inverter may be in a standby or shutdown state. The first power threshold can be set to 0.05 to 0.20, preferably 0.10; the second power threshold can be set to 0 to 0.05, preferably 0.02.

[0087] In step S4, the multi-decision tree classification layer includes:

[0088] Cascaded decision trees for manufacturer identification, model / version identification, and status type identification;

[0089] The manufacturer identification decision tree determines the manufacturer type based on manufacturer identifier, communication protocol source, status code format, and status text language characteristics. The model version identification decision tree, under manufacturer type constraints, determines the inverter model version based on the model version field, firmware version field, and status code distribution characteristics. The status type identification decision tree, under manufacturer type and model version constraints, determines the status type corresponding to the original status data based on code value characteristics, text characteristics, alarm characteristics, measurement point characteristics, and timing characteristics.

[0090] The status types include at least one of the following: running, standby, alarm, fault, power limit, communication anomaly, and maintenance. The output of the multi-decision tree classification layer is not the final standard state, but rather a leaf node or classification label. This leaf node or classification label is used to retrieve the candidate standard state set and candidate rule set in subsequent steps.

[0091] In step S5, the leaf node or classification label is pre-bound to the candidate standard state set and candidate rule set number in the rule index table; the multi-decision tree classification layer queries the rule index table according to the leaf node or classification label to retrieve the corresponding candidate standard state set and candidate rule set.

[0092] In a preferred embodiment, the rule index table includes an index number, manufacturer type, model version, status type, leaf node number, candidate standard status set, candidate rule set number, and rule priority field.

[0093] For example, if the manufacturer identification decision tree outputs manufacturer A, the model version identification decision tree outputs model B-V1.2, and the status type identification decision tree outputs power limit related status, then by querying the rule index table for "manufacturer A, model B-V1.2, power limit related status", a candidate standard status set and a candidate rule set can be obtained. The candidate standard status set can include normal operation, alarm operation, and power limit operation; the candidate rule set can include joint rules for status codes and power limit flags, AC output power ratio judgment rules, alarm information joint judgment rules, etc.

[0094] In this way, the multi-decision tree classification layer does not directly replace the rule engine, but provides a range of candidate rules for the rule engine, reduces the matching of all rules, improves mapping efficiency, and preserves the interpretability of rule matching results.

[0095] In step S6, the candidate rule set includes at least one of the following: code value direct mapping rule, status keyword matching rule, alarm information matching rule, status code and associated measurement point joint judgment rule, and status code, alarm information and associated measurement point joint judgment rule.

[0096] The direct mapping rule for code values ​​is used to handle cases where the original status code is well-defined. For example, if the original status code "00" in a certain model from a certain manufacturer is fixed to represent normal operation, then this rule will directly generate candidate mapping results for normal operation.

[0097] Status keyword matching rules are used to make judgments based on keywords in the status text. For example, if the status text contains "standby" or "standby", a candidate mapping result for standby is generated; if the status text contains "fault" or "fault", a candidate mapping result for fault shutdown is generated.

[0098] Alarm information matching rules are used to make judgments based on alarm codes, alarm levels, or alarm text. For example, when the alarm level is a critical alarm and the alarm text contains keywords such as "overtemperature," "insulation fault," or "DC overvoltage," candidate mapping results for fault shutdown or alarm operation are generated.

[0099] The joint judgment rule of status code and associated measurement point is used to solve the problem that the same status code has different meanings under different measurement point conditions. For example, a certain original status code may indicate standby when the power is low and not connected to the grid, but indicates power-limited operation when the grid connection is valid and the power limit flag is valid.

[0100] The joint judgment rule for status codes, alarm information, and associated measurement points is used to handle situations where there is a combined judgment relationship between the original status code, alarm information, and measurement point data. For example, if the original status code indicates an operating state, but there is a high-level alarm and the inverter temperature exceeds the temperature threshold, candidate mapping results for alarm operation or fault shutdown can be generated.

[0101] In step S6, the multi-evidence matching includes:

[0102] The code value matching score, text matching score, alarm matching score, measurement point consistency score, and state transition consistency score of the candidate mapping results are calculated respectively, and a weighted calculation is performed based on each score and its corresponding weight to generate the comprehensive confidence of the candidate mapping results.

[0103] The corresponding weights are determined based on the manufacturer type, status type, or historical mapping accuracy.

[0104] The overall confidence level is calculated as follows:

[0105]

[0106] Where C represents the overall confidence level of the candidate mapping results; Indicates the score for code value matching; Indicates the text matching score; This indicates the alarm matching score; Indicates the consistency score of the measurement points; Indicates the state transition consistency score; , , , and Indicates the corresponding weight, and .

[0107] The corresponding weights can be determined based on the manufacturer type, status type, or historical mapping accuracy. For example, for manufacturers with stable status code definitions, the weight of code value matching score is increased; for manufacturers with unstable status text but complete measurement point data, the weight of measurement point consistency score is increased; for fault-type statuses, the weights of alarm matching score and status transition consistency score are increased.

[0108] In a preferred embodiment, code value matching score The text matching score is determined based on whether the original status code matches the code value condition in the candidate rule. The alarm matching score is determined based on the degree of matching between the status text keywords and the candidate standard status keywords. The consistency score is determined based on the consistency between the alarm code, alarm level, and alarm text and the candidate rule conditions; [This is followed by a seemingly unrelated sentence about a test point consistency score.] The state transition consistency score is determined based on whether the associated measurement point data meets the candidate standard state determination criteria. The state is determined based on whether the transition relationship between the current candidate standard state and the previous standard state meets the state transition constraints.

[0109] In step S6, the state transition constraint is represented by a state transition matrix or a sequential state relationship table; when the state transition corresponding to the candidate mapping result does not satisfy the state transition constraint, the overall confidence of the candidate mapping result is reduced, or the candidate mapping result is marked as an abnormal mapping result.

[0110] For example, if the previous standard state was maintenance shutdown, and the current candidate mapping result is fault shutdown, and if maintenance shutdown in the state transition matrix does not allow a direct transition to fault shutdown, then the overall confidence of this candidate mapping result is reduced, or it is marked as an abnormal mapping result. When both standby and normal operation candidate mapping results exist and satisfy the state transition constraints, the candidate mapping result with higher overall confidence and a legal state transition is selected as the final standard state.

[0111] In a preferred embodiment, if the difference in the overall confidence level between two candidate mapping results is less than a preset difference threshold, the state priorities are further compared. For high-priority states, such as fault shutdown, communication interruption, or alarm operation, high-priority states can be output first, provided that the state transition constraints are met, to avoid missing critical states.

[0112] In step S7, the unknown state record includes:

[0113] Manufacturer identification, model version, original status code, status text, alarm information, associated measurement point data, candidate standard status, and generation time;

[0114] The dynamic update layer generates candidate update rules based on the similarity between unknown state records and historical mapping samples;

[0115] Similarity is calculated as follows:

[0116]

[0117] Wherein, Sim represents the overall similarity between the unknown state record and the historical mapping sample; Indicates code value similarity; Indicates text similarity; Indicates the similarity of alarms; The similarity of the measurement points is represented by α, β, γ and δ, which represent the corresponding weights, and α + β + γ + δ = 1.

[0118] Among them, code value similarity can be determined based on the degree of similarity of status code segments; text similarity can be determined based on keyword overlap; alarm similarity can be determined based on the consistency of alarm code, alarm level and alarm keywords; and measurement point similarity can be determined based on the distance between measurement point features such as the ratio of AC output power to rated power, grid connection indicator, power limit indicator and communication quality indicator.

[0119] When the similarity is higher than the preset similarity threshold and the associated measurement point data meets the judgment conditions of the candidate standard state, or after manual confirmation, the candidate update rule is written into the rule base and the corresponding unknown state record is written into the training sample base.

[0120] In a preferred embodiment, the dynamic update layer generates candidate update rules based on the similarity between unknown state records and historical mapping samples, and confirms and judges the candidate standard states corresponding to the candidate update rules.

[0121] When the similarity between an unknown state record and a historical mapping sample is higher than a preset similarity threshold, and the associated measurement point data meets the judgment conditions corresponding to the candidate standard state, the candidate update rule is determined as a writable rule.

[0122] When the similarity between an unknown state record and a historical mapping sample does not reach a preset similarity threshold, or when the associated measurement point data does not meet the judgment conditions corresponding to the candidate standard state, the unknown state record and candidate update rules are retained and written into the rule base after confirmation.

[0123] After confirmation, the candidate update rules are written into the rule base, and the corresponding unknown state records are written into the training sample base.

[0124] The rule records in the rule base include rule number, applicable manufacturer, applicable model version, matching conditions, target standard status, effective time, and rule version number. The sample records in the training sample base include state feature vectors, confirmed standard status, confirmation time, and data source. When the number of newly added samples in the training sample base reaches a preset number or the proportion of newly added samples reaches a preset proportion, incremental training or retraining of the multi-decision tree classification layer is triggered to update the binding relationship between leaf nodes or classification labels and candidate rule sets.

[0125] In a preferred embodiment, the preset reliability threshold can be dynamically set according to the state type. For example, the preset reliability threshold for normal operation can be set to 0.70 to 0.85, and the preset reliability threshold for fault shutdown, communication interruption, and alarm operation can be set to 0.80 to 0.95 to improve the mapping reliability of critical states. The preset similarity threshold can be set to 0.75 to 0.95, preferably 0.85.

[0126] Furthermore, standardized status data is output. Standardized status data includes manufacturer identifier, model version, original status code, final standard status, mapping rule number, overall confidence level, status occurrence time, and mapping version number.

[0127] The mapping version number is used to distinguish mapping results under different rule base versions. When the rule base is updated, new mapping results adopt the new rule version number, while historical mapping results retain the original rule version number. By using the mapping rule number and mapping version number, the mapping basis, rule source, and rule effective time of each standardized status data can be traced, avoiding confusion in the definition of historical status data after the rule base is updated.

[0128] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The embodiments and descriptions in the specification are merely principles of the invention. Various changes and modifications can be made to the invention without departing from its spirit and scope, and all such changes and modifications fall within the scope of the claimed invention. The scope of protection claimed by the appended claims and their equivalents is defined.

Claims

1. An automatic mapping model and management method for the original state and standard state of inverter manufacturers, characterized in that, The automatic mapping model includes a multi-decision tree classification layer, a rule engine matching layer, a confidence level arbitration layer, and a dynamic update layer; the management method includes: S1. Establish or import the inverter standard state system, which includes standard state codes, standard state names, state priorities, and state transition constraints. S2. Obtain original status data of inverter manufacturers from multiple manufacturers. The data includes manufacturer identification, model version, original status code, status text, alarm information, associated measurement point data, and timestamp. S3. Standardize the format and handle anomalies of the manufacturer's original status data, and construct a status feature vector that includes code value, text, alarm, measurement point and time sequence features; S4. Input the state feature vector into the automatic mapping model and perform multi-decision tree classification, rule engine matching, confidence arbitration and dynamic update in the order of coarse screening, fine matching, arbitration and evolution. S5. The multi-decision tree determines the leaf node or classification label based on the manufacturer type, model version and status type, and retrieves the candidate standard status set and candidate rule set bound to it. S6. Perform multi-evidence matching on the candidate rule set to generate candidate mapping results. Determine the final standard state according to the comprehensive confidence level, state priority and state transition constraints. Reduce the weight or mark the results that violate the state transition constraints as abnormal. S7. When the overall confidence level is lower than the threshold or no candidate rule set is hit, candidate update rules are generated based on unknown state records and similar historical mapping samples. After confirmation, the rule base and sample base are updated, and standardized state data containing the final standard state, rule number, confidence level, occurrence time and version number are output.

2. The automatic mapping model and management method for the original state and standard state of inverter manufacturers according to claim 1, characterized in that, In step S1, the inverter standard state system includes: Standard state classification field, standard state code field, standard state name field, state priority field, allowed preceding state field, allowed subsequent state field, state mutual exclusion constraint field, and state determination description field; The standard states include at least normal operation, standby, alarm operation, fault shutdown, power-limited operation, communication interruption, and maintenance shutdown.

3. The automatic mapping model and management method for the original state and standard state of inverter manufacturers according to claim 1, characterized in that, In step S2, the associated measurement point data includes: At least one of the following: DC voltage, DC current, AC output power, grid connection indicator, power limit indicator, inverter temperature, and communication quality indicator; The associated measurement point data and the original status code are associated according to the same inverter number and the same sampling time window.

4. The automatic mapping model and management method for the original state and standard state of inverter manufacturers according to claim 1, characterized in that, In step S3, constructing the state feature vector includes: The original status code is parsed to generate code value features; Keyword extraction is performed on status text and alarm information to generate text features and alarm features; The associated measurement point data is normalized to generate measurement point features; And generate timing features based on the previous standard state, state duration, and state transition interval; When there is no previous standard state, the timing features are generated with a preset initial state.

5. The automatic mapping model and management method for the original state and standard state of inverter manufacturers according to claim 1, characterized in that, In step S4, the multi-decision tree classification layer includes: Cascaded decision trees for manufacturer identification, model / version identification, and status type identification; Among them, the manufacturer identification decision tree is used to determine the manufacturer type, the model version identification decision tree is used to determine the inverter model version under the manufacturer type constraint, and the status type identification decision tree is used to determine the status type corresponding to the manufacturer's original status data under the manufacturer type and model version constraints.

6. The automatic mapping model and management method for the original state and standard state of inverter manufacturers according to claim 1, characterized in that, In step S5, the leaf node or classification label is pre-bound to the candidate standard state set and candidate rule set number in the rule index table; the multi-decision tree classification layer queries the rule index table according to the leaf node or classification label to retrieve the corresponding candidate standard state set and candidate rule set.

7. The automatic mapping model and management method for the original state and standard state of inverter manufacturers according to claim 1, characterized in that, In step S6, the candidate rule set includes at least one of the following: code value direct mapping rule, status keyword matching rule, alarm information matching rule, status code and associated measurement point joint judgment rule, and status code, alarm information and associated measurement point joint judgment rule.

8. The automatic mapping model and management method for the original state and standard state of inverter manufacturers according to claim 1, characterized in that, In step S6, the multi-evidence matching includes: The code value matching score, text matching score, alarm matching score, measurement point consistency score, and state transition consistency score of the candidate mapping results are calculated respectively, and a weighted calculation is performed based on each score and its corresponding weight to generate the comprehensive confidence of the candidate mapping results. The corresponding weights are determined based on the manufacturer type, status type, or historical mapping accuracy.

9. The automatic mapping model and management method for the original state and standard state of inverter manufacturers according to claim 8, characterized in that, In step S6, the state transition constraint is represented by a state transition matrix or a sequential state relationship table; when the state transition corresponding to the candidate mapping result does not satisfy the state transition constraint, the overall confidence of the candidate mapping result is reduced, or the candidate mapping result is marked as an abnormal mapping result.

10. The automatic mapping model and management method for the original state and standard state of inverter manufacturers according to claim 1, characterized in that, In step S7, the unknown state record includes: Manufacturer identification, model version, original status code, status text, alarm information, associated measurement point data, candidate standard status, and generation time; The dynamic update layer generates candidate update rules based on the similarity between unknown state records and historical mapping samples; When the similarity is higher than the preset similarity threshold and the associated measurement point data meets the judgment conditions of the candidate standard state, or after manual confirmation, the candidate update rule is written into the rule base and the corresponding unknown state record is written into the training sample base. The rule records in the rule base include rule number, applicable manufacturer, applicable model version, matching conditions, target standard status, effective time, and rule version number.

Images