Power plant income index deep mining method based on historical operation data

Abstract

The invention relates to the technical field of data mining, in particular to a power plant income index deep mining method based on historical operation data, which comprises the following steps: based on input power plant operation data, selecting core income dimension time and key income measurement generating capacity to carry out initial structure design, and setting a hierarchical division standard of the income dimension time, and aggregating the key income measurement generating capacity according to a set time hierarchy, and establishing a hierarchical aggregation power plant income data structure. According to the method, the time dimension and the power generation capacity measurement in the input power plant operation data are subjected to structured processing, a hierarchical aggregation structure with hierarchical division capability is constructed, and higher query efficiency and clear expression of time sequence aggregation logic are provided in a data organization mode. By embedding the data integrity state and accuracy into the aggregation unit, a composite data structure with business measurement and quality marking capabilities at the same time is constructed.

Description

Technical Field

[0001] The present invention relates to the technical field of data mining, and in particular to a method for deep mining of power plant revenue indicators based on historical operating data. Background Art

[0002] Data mining refers to the process of automatically or semi-automatically extracting valuable information patterns and knowledge that are implicit but unknown to people in advance from large, complex, and potentially related data sets.

[0003] Existing technologies lack structural design to embed quality information. Consequently, evaluating data validity often requires the construction of external quality models. This leads to fragmented analysis logic, lengthened response times, and reduced credibility of response results. During metric queries, full table scans and generalized matching are often employed, lacking the binding of target fields to time constraints. This results in performance degradation under high-frequency query pressure and loss of data consistency. Therefore, improvements are needed. Summary of the Invention

[0004] The purpose of the present invention is to solve the shortcomings of the existing technology and propose a deep mining method for power plant profit indicators based on historical operation data.

[0005] To achieve the above objectives, the present invention adopts the following technical solution: a method for deep mining of power plant profit indicators based on historical operation data, comprising the following steps:

[0006] Based on the input power plant operation data, the core revenue dimension time and the key revenue metric power generation are selected for initial structure design. The hierarchical division standard of the revenue dimension time is set, and the key revenue metric power generation is aggregated according to the set time hierarchy to establish a hierarchical aggregated power plant revenue data structure.

[0007] Based on the hierarchical aggregated power plant revenue data structure, for the aggregated power generation, evaluating the data integrity status and data accuracy of the corresponding original power plant operation data to obtain a core data quality assessment set; based on the core data quality assessment set, embedding the data integrity status and data accuracy into corresponding aggregation units of the hierarchical aggregated power plant revenue data structure to generate a revenue data structure;

[0008] Based on the revenue data structure, responding to a real-time analysis query request for a target power plant revenue indicator, parsing the power plant revenue indicator name and time range parameters required by the query request to obtain a target revenue indicator query parameter package, retrieving the corresponding power plant revenue indicator from the revenue data structure based on the target revenue indicator query parameter package, combining the results into a single query response, and establishing a revenue indicator query response;

[0009] Based on the profit indicator query response, the trend of the power plant profit indicator over time is examined, the time point of profit fluctuation is identified, and a preliminary profit fluctuation identification sequence is obtained. Based on the preliminary profit fluctuation identification sequence and combined with the information attached to each data point in the sequence, the credibility score of the overall fluctuation pattern is calculated, and a profit fluctuation analysis summary is established.

[0010] Preferably, the steps for obtaining the hierarchical aggregated power plant revenue data structure are:

[0011] Based on the input power plant operation data, the operation time field and power generation field in each record are extracted. The operation time field is uniformly converted into a standard timestamp format and divided according to the preset units. The corresponding power generation field within each divided time unit range is then grouped and aggregated to generate a power generation aggregate distribution set with a fixed time scale.

[0012] Calculating a composite value of power generation offset in each time interval based on the aggregated distribution set of power generation with a fixed time scale;

[0013] According to the composite value of power generation offset, determine whether it exceeds the upper limit of the power generation offset reference threshold. If the condition is met that the composite value of power generation offset is greater than the upper limit of the power generation offset reference threshold, then the time interval is marked as an abnormal operating profit interval; otherwise, it is marked as a normal profit interval, forming a hierarchical aggregated power plant profit data structure.

[0014] Preferably, the steps for obtaining the core data quality assessment set are:

[0015] Based on the hierarchical aggregated power plant revenue data structure, the original power plant operation data entries associated with the aggregated power generation in each time interval are parsed, and the timestamp, power generation field, operation duration field, temperature field, and equipment number field contained in each original power plant operation data are extracted to obtain the associated original power plant operation data set;

[0016] Based on the associated original power plant operation data set, check each time whether the timestamp field is missing or has a format error, determine whether the power generation field, the operation time field, and the temperature field are missing or have abnormal value ranges, and mark each original power plant operation data with a completeness and accuracy dual status to form an original data quality status mark set;

[0017] Based on the original data quality status mark set, the percentage of the completeness and accuracy status of all original power plant operation data in each time interval is counted to form a core data quality assessment set.

[0018] Preferably, the steps for obtaining the income data structure are:

[0019] Based on the core data quality assessment set, extract the integrity status label and accuracy status label corresponding to each time interval one by one, assign a unique identification number to each time interval, form a structured state mapping list with the status label and the number, and generate a numbered quality status label list;

[0020] According to the numbered quality status tag list, traverse the aggregation units at all time levels in the hierarchical aggregated power plant revenue data structure, compare the time interval numbers of the aggregation units with the numbers in the status tag list in turn, and after confirming a successful match, insert the integrity status tag into the metadata area field of the aggregation unit, and at the same time insert the accuracy status tag into the metadata area field of the aggregation unit, to obtain the hierarchical aggregated power plant revenue data structure with the quality status field embedded;

[0021] Based on the hierarchical aggregated power plant revenue data structure with embedded quality status fields, the integrity status identification fields and accuracy status identification fields of all aggregated units are traversed, the number of units with abnormal status is counted, and each type of status is clustered and marked under each time level dimension, and corresponding cluster labels are generated. The quality status, cluster labels and aggregated content of all aggregated units are uniformly saved and set to read-only state to generate a revenue data structure.

[0022] Preferably, the steps for obtaining the target revenue index query parameter package are:

[0023] Based on the revenue data structure, after receiving a power plant revenue indicator analysis query request initiated by a user, the complete request body content submitted in the query request is extracted, and the keywords in the body content are marked as four types of elements: target power plant identification tag, revenue indicator name keyword, start time field and end time field, to generate a structured parsed field set;

[0024] Based on the structured parsed field set, verify whether the target power plant identification tag exists in the registered power plant directory table in the revenue data structure. If a match is successful, read the revenue indicator field summary table corresponding to the power plant and compare whether there is a complete match for the revenue indicator name keyword in the field summary table. At the same time, check whether the time format of the start time field and the end time field is legal and the range is reasonable, and generate a target revenue indicator field and time parameter set that passes the verification;

[0025] Based on the verified target profit indicator field and time parameter set, a combination object with four types of content: target power plant identification label, standardized profit indicator name, standardized start time field and standardized end time field is constructed to generate a target profit indicator query parameter package.

[0026] Preferably, the steps of obtaining the revenue indicator query response are:

[0027] Based on the target profit indicator query parameter package, the target power plant identification tag, the standardized profit indicator name, the standardized start time field, and the standardized end time field are extracted; the power plant identification tag is used to lock the corresponding power plant branch node in the profit data structure; the aggregate field set corresponding to the target profit indicator name is located under the node; and all aggregate field contents that meet the time conditions are filtered according to the start time field and the end time field to generate the target profit indicator data set;

[0028] Based on the target benefit indicator data set, traverse each aggregate field content and uniformly parse the included power generation aggregate value, completeness status identifier, and accuracy status identifier, combine the field contents within the same time interval into a single result record, sort all records according to the time interval sequence and combine them into a structured response body, and generate a sorted target benefit indicator structured record set;

[0029] Based on the sorted target profit indicator structured record set, a query response structure in a unified format is constructed, and the indicator name, power plant tag, time range and structured record content required by the response body are filled in to generate a profit indicator query response.

[0030] Preferably, the steps for obtaining the preliminary return fluctuation identification sequence are:

[0031] Based on the revenue index query response, extract the revenue index value field, equipment load rate field, and timestamp field from all records, convert the equipment load rate field into a decimal format according to the percentage, uniformly convert the timestamp field into a real value in hours, and sort them in ascending time order. After removing the first and last time points, generate a standardized revenue index time series structure;

[0032] Calculate the normalized multi-factor volatility coefficient corresponding to each time point based on the time series structure of the standardized return indicator;

[0033] Based on the normalized multi-factor volatility coefficient, the entire time series structure is traversed, and the time points where the normalized multi-factor volatility coefficient is more than twice the average value of the two adjacent points are marked as candidate nodes for abnormal returns. All marked time points are combined in timestamp order to form a candidate point sequence, and a preliminary return fluctuation identification sequence is generated.

[0034] Preferably, the steps for obtaining the income fluctuation analysis summary are:

[0035] Calculating a composite credibility score for the overall volatility pattern based on the preliminary return volatility identification sequence;

[0036] Based on the composite credibility score, and compared with the established credibility score standard curve, it is determined whether the current score is higher than the upper limit of the credibility threshold within the 30-day rolling interval. If it is higher, the fluctuation sequence is marked as "abnormally significant high volatility", otherwise it is marked as "regular volatility", and a return volatility analysis summary is generated.

[0037] Compared with the prior art, the advantages and positive effects of the present invention are:

[0038] The present invention constructs a hierarchical aggregation structure with hierarchical division capabilities through structured processing of the time dimension and power generation measurement in the input power plant operation data, providing higher query efficiency and clear expression of time series aggregation logic in terms of data organization. By embedding the data integrity status and accuracy into the aggregation unit, a composite data structure with both business measurement and quality marking capabilities is constructed, which achieves more detailed quality tracking in the data credibility assessment process and enhances the reliability of subsequent analysis. In the query response process, a combined screening mechanism based on the target revenue indicator name and time parameters ensures a high degree of consistency in response granularity and accuracy, thereby supporting high-frequency, low-latency indicator call scenarios. By dynamically examining the temporal change trend of the revenue indicator and combining multi-dimensional information such as load, temperature, and integrity status to jointly construct a fluctuation credibility score, the sensitivity and accuracy of revenue anomaly identification are improved, and the risk of misjudgment due to single-dimensional judgment is avoided when identifying critical fluctuation nodes. BRIEF DESCRIPTION OF THE DRAWINGS

[0039] Figure 1 Schematic diagram of the steps of the present invention. DETAILED DESCRIPTION

[0040] In order to make the purpose, technical solutions and advantages of the present invention more clearly understood, the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.

[0041] See also Figure 1 The present invention provides a technical solution, a method for deep mining of power plant profit indicators based on historical operation data, comprising the following steps:

[0042] Based on the input power plant operation data, the core revenue dimension time and the key revenue metric power generation are selected for initial structure design. The hierarchical division standard of the revenue dimension time is set, and the key revenue metric power generation is aggregated according to the set time hierarchy to establish a hierarchical aggregated power plant revenue data structure.

[0043] Based on the hierarchical aggregated power plant revenue data structure, the data integrity status and data accuracy of the corresponding original power plant operation data are evaluated for the aggregated power generation to obtain a core data quality assessment set. Based on the core data quality assessment set, the data integrity status and data accuracy are embedded in the corresponding aggregation units of the hierarchical aggregated power plant revenue data structure to generate a revenue data structure.

[0044] Based on the revenue data structure, respond to the real-time analysis query request for the target power plant revenue indicator, parse the power plant revenue indicator name and time range parameters required by the query request, obtain the target revenue indicator query parameter package, retrieve the corresponding power plant revenue indicator from the revenue data structure based on the target revenue indicator query parameter package, combine them into a single query response, and establish a revenue indicator query response;

[0045] Based on the response to the revenue indicator query, the trend of the power plant's revenue indicators over time is examined, the time points of revenue fluctuations are identified, and a preliminary revenue fluctuation identification sequence is obtained. Based on the preliminary revenue fluctuation identification sequence, the credibility score of the overall fluctuation pattern is calculated in combination with the information attached to each data point in the sequence, and a revenue fluctuation analysis summary is established.

[0046] The steps to obtain the hierarchical aggregated power plant revenue data structure are:

[0047] Based on the input power plant operation data, the operation time field and power generation field in each record are extracted. The operation time field is uniformly converted into a standard timestamp format and divided according to the preset units. The corresponding power generation field within each divided time unit range is then grouped and aggregated to generate a power generation aggregate distribution set with a fixed time scale.

[0048] According to the aggregated distribution set of power generation with a fixed time scale, the composite value of power generation offset in each time interval is calculated. The calculation formula is:

[0049]

[0050] Among them, Q k is the composite value of power generation offset in the kth time interval, P k,i is the power generation value of the i-th record in the k-th time interval (unit: MWh), is the average value of power generation in this time interval (unit: MWh), OT k,i The device running time of this record (unit: hours), Temp k,i is the temperature reading of the device where the record is located (unit: Celsius), n k is the number of records in the kth time interval;

[0051] According to the composite value of power generation offset, it is judged whether it exceeds the upper limit of the power generation offset reference threshold. If the condition is met that the composite value of power generation offset is greater than the upper limit of the power generation offset reference threshold, the time interval is marked as an abnormal operating profit interval, otherwise it is marked as a normal profit interval, forming a hierarchical aggregated power plant profit data structure.

[0052] Specifically, based on the input power plant operation data, first through the data interface protocol, such as OPC

[0053] (OLE for Process Control) or Modbus, batch read the original operation records from the distributed control system (DCS) or data acquisition and supervisory control system (SCADA) of the power plant. Each record contains dozens to hundreds of parameters such as timestamp, generator set number, power generation, equipment temperature, operating time, etc. The operation time field in each record is specifically extracted, such as "2023-10-01T10:15:30.050Z" or "1696155330", as well as power generation fields, such as the recorded instantaneous power value or the accumulated electric energy in a very short period. Then, all the extracted operation time fields are uniformly parsed and converted into the internationally standardized UTC timestamp format, such as the string format of "YYYY-MM-DDHH:MM:SS" or the Unix timestamp (the number of seconds since January 1, 1970 UTC 00:00), to ensure the consistency of time data for subsequent sequence analysis. Then, according to the preset time aggregation unit, this unit is analyzed based on The precision requirement setting is as follows. For example, if hour-level analysis is required, the preset unit is 1 hour. If minute-level analysis is required, it is 1 minute or 5 minutes. Each standardized running timestamp is divided into the corresponding time unit interval. For example, the timestamp "2023-10-0110:15:30" belongs to the hour interval "2023-10-0110:00:00" to "2023-10-0110:59:59" when divided into hours. After the time division is completed, the time interval of the same preset time is Group aggregation is performed on all power generation field values ​​within a unit interval. The total power generation within that time unit is usually calculated using sum aggregation. For example, if there are 60 minute-level power generation records in the interval "2023-10-01 10:00:00" to "2023-10-01 10:59:59", each record is 0.5MWh, then the aggregated power generation is 60×0.5=30MWh. This process iterates over all input data and ultimately generates a power generation aggregate distribution set with a fixed time scale.

[0054] formula: The usefulness of the formula is that it provides a quantitative indicator Q that comprehensively considers the impact of power generation fluctuations, fluctuation duration, and equipment operating temperature.k , which is used to evaluate the stability of power plant operation and potential profit risk within a specific time interval. k,i The average power generation in its interval Deviation, and combined with the corresponding equipment operation time OT k,i , can amplify the impact of power generation deviation with a longer duration, that is, the power generation deviation of the same size, if it lasts longer, will have a greater impact on Q k The weight in is larger, which can better reflect its impact on stability. In addition, this weighted deviation is divided by the device temperature Temp k,i , is intended to make a certain degree of correction or consideration for changes in power generation efficiency or stability under the influence of temperature. For example, in some equipment, extreme temperatures may make power generation fluctuations more likely to occur or more difficult to control. The introduction of this parameter helps to distinguish between abnormal fluctuations under normal temperature and expected fluctuations under extreme temperature conditions. The root mean square calculation method is used to make the larger deviation term have a greater impact on Q k The contribution of Q is more significant, thus more sensitively capturing significant power generation deviation events. This comprehensive quantitative evaluation makes subsequent Q-based k The operating status judgment based on the power generation value (such as whether it is in the abnormal operating profit range) is more objective and comprehensive, rather than relying solely on the single absolute value of power generation or simple difference.

[0055] n k The steps for obtaining the parameter are as follows: This parameter represents the number of raw power plant operation data records in the kth time interval. The input power plant operation data has been divided according to the preset time unit (for example, 1 hour). For a specific kth time interval (for example, from 10:00 to 10:59 am on May 12, 2025), n is obtained by directly counting the total number of raw data record entries contained in the interval. k For example, if the power plant SCADA system records operating data once a minute, then in the kth interval of an hour, there should theoretically be 60 records. If there is no missing data during the period, then n k =60. When actually obtaining, all records whose timestamp field values ​​fall within the kth time interval are filtered out from the database or data file, and then these filtered records are counted. For example, executing the SQL query

[0056] SELECT COUNT(*) FROM operation_data WHERE timestamp>='2025-05-1210:00:00' AND timestamp<'2025-05-1211:00:00' The result is n k In this example, it is assumed that within a specific 1-hour interval k, a total of nk =3 valid power generation records.

[0057] P k,i The parameter is obtained as follows: This parameter represents the power generation value of the i-th raw record in the k-th time interval, in MWh. This data is directly extracted from the input power plant operation data. Each raw operation data record (for example, a SCADA system log entry) contains a power generation field, the value of which is P k,i For example, in the first record of the kth time interval, if the "power generation" or "accumulated electric energy" related fields are consulted and the record value is 50000kWh, then it is converted to P k,1 =50MWh. For this example, set P k,1 =50MWh, P k,2 =60MWh, P k,3 =40MWh.

[0058] The steps for obtaining the parameter are as follows: This parameter represents all the original recorded power generation values ​​P in the kth time interval k,i The arithmetic mean of the units is MWh. It is calculated by taking the n k P of records k,i The values ​​are summed and then divided by the number of records n k For example, using the previously obtained P k,i Value and n k Value, if P k,1 =50MWh, P k,2 =60MWh, P k,3 =40MWh, and n k =3, then

[0059] OT k,i The steps to obtain the parameter are as follows: This parameter represents the actual operating time of the equipment corresponding to the i-th record in the k-th time interval, in hours. This data is also extracted from the input power plant operation data for each record. In some power plant data records, the actual online operating time of the generator set in the corresponding data sampling period will be clearly recorded, or it can be calculated from the start-up and shutdown event log. For example, a SCADA record may correspond to a 5-minute sampling period. If the unit continues to operate within these 5 minutes, the OT k,i =5 / 60=0.0833 hours. If the record itself represents an operation block, the fields it contains will directly indicate the duration of the block. For this example, the equipment operation time corresponding to these three records is set to be OT k,1 =0.3 hours, OT k,2 =0.4 hours, OT k,3= 0.3 hours, and the total duration is 1 hour, which is consistent with the total duration of time interval k, indicating that the device is divided into these three sub-operation periods for recording within this hour.

[0060] Temp k,i The steps for obtaining the parameter are as follows: This parameter represents the temperature reading of the key parts of the power generation equipment corresponding to the record when the i-th record occurs in the k-th time interval, and the unit is degrees Celsius (℃). This data is extracted from the equipment status parameters collected synchronously with the power generation, operating time and other information in the power plant operation data. Usually it is the key temperature measurement point of the generator set, such as bearing temperature, winding temperature or ambient temperature, etc. The specific measurement point depends on the monitoring configuration of the power plant. For example, by querying the "generator winding temperature" field associated with the i-th record in the k-th time interval, its value is obtained. For this example, the corresponding temperature data is queried from the power plant database and set as Temp. k,1 =70℃, Temp k,2 =72℃, Temp k,3 =75℃.

[0061] Calculation process: Select the kth time interval for calculation, and obtain according to the above parameters: n k =3,P k,1 =50MWh, P k,2 =60MWh, P k,3 =40MWh, OT k,1 =0.3h, OT k,2 =0.4h, OT k,3 =0.3h, Temp k,1 =70℃, Temp k,2 =72℃, Temp k,3 =75℃;

[0062] Calculate the contribution of each record i For i=1:

[0063]

[0064] For i=2:

[0065]

[0066] For i=3:

[0067]

[0068] Sum and average:

[0069]

[0070] Taking the square root, we get Qk :

[0071]

[0072] The result shows that in the kth time interval, after comprehensively considering the degree of deviation of power generation from the mean, the equipment operating time under the deviation, and the equipment temperature at that time, the composite value of power generation offset is 0.039524MWh·h / ℃. This value itself represents a weighted, standardized fluctuation amplitude. The higher the value, the greater the fluctuation or the more "abnormal" the power generation process in the time interval (under given temperature and duration weights). If the value is lower than the preset "power generation offset reference threshold upper limit", the power generation fluctuation in this interval is considered to be within an acceptable range; if it is higher than the threshold, it may indicate potential operating problems or efficiency losses, which require further attention. This Q k The value will serve as the key basis for subsequent judgment on whether the time interval is an "abnormal operating income interval".

[0073] The composite value Q of power generation offset in each time interval calculated in the previous step k , then we will judge these Q k The value exceeds the preset power generation deviation reference threshold upper limit, which is set based on the Q k The statistical analysis of the values ​​and the combination of expert operation experience determine, for example, collecting Q values ​​for all time intervals in the past year k The 95th percentile of this data set is calculated and used as the upper limit of the initial reference threshold. For example, if the 95th percentile is 0.045MWh·h / ℃, the upper limit of the reference threshold for power generation deviation is initially set to 0.045. Then, a team of engineers with more than 5 years of experience in power plant operation and management is invited to analyze the Q of several known abnormal operation events (such as equipment failure, unstable fuel supply, sudden change in grid scheduling, etc.) marked in the historical data. k The value is reviewed and fine-tuned in combination with the false alarm rate and missed alarm rate under the current threshold until the threshold can achieve acceptable anomaly recognition accuracy on historical data. Finally, a reference threshold upper limit of power generation deviation such as 0.050MWh·h / ℃ is determined. After the threshold is determined, the system will traverse each time interval and its corresponding Q k Value, perform comparison judgment, if the Q of the current time interval k Value, such as Q obtained in the previous example k ≈0.039524MWh·h / ℃, which is less than the upper limit of the reference threshold of power generation deviation 0.050MWh·h / ℃, then this time interval is marked as "normal benefit interval" in the data structure. If the Q of another time interval is kIf the value is 0.058MWh·h / ℃, then because it is greater than 0.050MWh·h / ℃, this time interval will be marked as an "abnormal operating profit interval". By completing this marking process for all time intervals, each record in the original "aggregate distribution set of power generation with a fixed time scale" is added with a status label (normal or abnormal), thus forming a richer hierarchical aggregated power plant profit data structure.

[0074] The steps for obtaining the core data quality assessment set are:

[0075] Based on the hierarchical aggregated power plant revenue data structure, the original power plant operation data entries associated with the aggregated power generation in each time interval are parsed. The timestamp, power generation field, operation duration field, temperature field, and equipment number field contained in each original power plant operation data are extracted to obtain the associated original power plant operation data set.

[0076] Based on the associated original power plant operation data set, the timestamp field is checked one by one to see if it is missing or has format errors. The power generation field, operating time field, and temperature field are checked to see if they are missing or have abnormal value ranges. Each piece of original power plant operation data is marked with a completeness and accuracy dual status to form a set of original data quality status marks.

[0077] Based on the original data quality status mark set, the percentage of completeness and accuracy status of all original power plant operation data in each time interval is counted to form a core data quality assessment set.

[0078] Specifically, based on the hierarchical aggregated power plant revenue data structure, each time interval entry in the data structure is first traversed. Each entry contains the identifier of the interval (such as the start timestamp) and the aggregated power generation and operating status mark calculated in the previous step. The start and end timestamps of the time interval are used as query conditions to reversely query the database or log system storing the original power plant operation data, and retrieve all original record entries whose timestamps fall within the specific time interval. For example, for the time interval entries from "2025-05-12 10:00:00" to "2025-05-12 10:59:59", a similar query is performed as follows: SELECT timestamp, power_generation, operation_time, temperature, equipment_id FROM raw_data WHERE timestamp>='2025-05-12 10:00:00' AND timestamp<'2025-05-12 11:00:00 0' query operation, extract the required data fields from each raw power plant operation data record returned by the query, including the timestamp field (such as the value of the 'timestamp' column '2025-05-12 10:15:32'), the power generation field (such as the value of the 'power_generation' column '55.2', unit MWh), the equipment operation time field (such as the value of the 'operation_time' column '0.0167', unit hour), the equipment temperature field (such as the value of the 'temperature' column '78.5', unit Celsius), and the equipment ID field identifying the specific power generation equipment (such as the value of the 'equipment_id' column 'GEN002'). All raw records retrieved from the same time interval and the extracted five fields are combined to form a raw data subset associated with the time interval. This process is repeated for all time intervals, and finally the raw data subsets of all time intervals are integrated to obtain a complete associated raw power plant operation data set.

[0079] According to the associated original power plant operation data set, the system will perform an automated check process for each original power plant operation data record in the data set. First, the timestamp field is checked to verify whether there is a blank value (such as NULL or an empty string) and whether its format conforms to the predefined standard timestamp format (for example, "YYYY-MM-DDHH:MM:SS"). If there is a blank or the format does not match (for example, an invalid date such as "2025 / 13 / 0110:00:00"), the timestamp field of the record is considered incomplete or inaccurate. Then, the power generation is determined. The three numeric fields, namely, the power field, the operating time field, and the temperature field, are checked for missing values ​​(such as NULL). If any field is missing, the record has a problem with integrity. Furthermore, the accuracy of the non-missing numeric fields is checked, and their values ​​are compared with the preset abnormal value ranges. These ranges are set based on physical limitations, equipment design specifications, and the statistical distribution of long-term operating data (such as eliminating data points that are outside the range of plus or minus three times the standard deviation of the mean). For example, the reasonable range of the power generation field (MWh) is set to 0 to 100MWh (for example, a single record Corresponding to a shorter time, a high value may be an error). The reasonable range of the operation duration field (hour) is set to 0 to 1 (for example, the time granularity of a single record does not exceed 1 hour). The temperature field (degrees Celsius) is set to different ranges according to different measurement points. For example, the generator bearing temperature range is set to 20℃ to 120℃. If the value of a field falls outside its corresponding range (for example, the temperature reading is -10℃ or 200℃), the field is considered to be inaccurate data. Based on the above inspection results, a dual status tag is assigned to each piece of raw power plant operation data. The tag consists of two parts: integrity status (for example, "Complete" or "Missing Field") and accuracy status (for example, "Accurate", "Wrong Format" or "Value Out of Limit"). For example, if all fields of a record are complete and the values ​​are within a reasonable range, it is marked as {Completeness: "Complete", Accuracy: "Accurate"}; if the temperature field is missing, it is marked as {Completeness: "Missing Field", Accuracy: "Accurate"} (other fields are accurate); if the power generation value is 150MWh and exceeds the range, it is marked as {Completeness: "Complete", Accuracy: "Value Out of Limit"}. After completing the marking of all records, the original data quality status mark set is formed.

[0080] Based on the original data quality status mark set, which contains the dual status marks of completeness and accuracy of each original power plant operation data in the dataset, the next step is to summarize these status information by time interval. The system first regroups all records with status marks into the fixed time scale interval (for example, hourly interval) to which they belong based on the timestamp field of the original record. For each time interval, such as "2025-05-12 10:00:00" to "2025-05-12 10:59:59", the number of various status mark combinations of all original power plant operation data records in the interval is counted. Specifically, the number of records marked as {completeness: "complete", accuracy: "accurate"}, the number of records marked as {completeness: "complete", accuracy: "format error"}, the number of records marked as {completeness: "complete", accuracy: "value out of limit"}, and the number of records marked as {completeness: "missing field", accuracy: "accurate"} are calculated.}, as well as the number of records with other possible status combinations (such as missing and exceeded at the same time), and then calculate the percentage of the number of records in each status combination to the total number of original records in the time interval. For example, if there are 60 original records in the hourly interval, 55 of them are marked as {completeness: "complete", accuracy: "accurate"}, 2 are marked as {completeness: "missing field", accuracy: "accurate"}, and 3 are marked as {completeness: "complete", accuracy: "value exceeded"}, then the status proportions of this interval are: complete and accurate account for 55 / 60≈91.7%, missing fields account for 2 / 60≈3.3%, and value exceeded account for 3 / 60=5.0%. These status statistics for each time interval (usually the proportion or absolute number of various states) are organized to form a data set indexed by time interval, which contains quantitative evaluation information of the integrity and accuracy of the original data in the corresponding interval, namely the core data quality assessment set.

[0081] The steps to obtain the income data structure are:

[0082] Based on the core data quality assessment set, the integrity status label and accuracy status label corresponding to each time interval are extracted one by one, and each time interval is assigned a unique identification number. The status label and number are combined into a structured status mapping list to generate a numbered quality status label list;

[0083] According to the numbered quality status tag list, traverse the aggregation units at all time levels in the hierarchical aggregated power plant revenue data structure, compare the time interval numbers of the aggregation units with the numbers in the status tag list in turn, and after confirming a successful match, insert the integrity status tag into the metadata area field of the aggregation unit, and at the same time insert the accuracy status tag into the metadata area field of the aggregation unit, to obtain the hierarchical aggregated power plant revenue data structure with the quality status field embedded;

[0084] Based on the hierarchical aggregated power plant revenue data structure with embedded quality status fields, the integrity status identification fields of all aggregated units are traversed, the number of units with abnormal status is counted, and each type of status is clustered and marked under each time level dimension to generate corresponding cluster labels. The quality status, cluster labels and aggregate content of all aggregated units are uniformly saved and set to read-only state to generate the revenue data structure.

[0085] Specifically, based on the core data quality assessment set, which contains detailed statistical percentages of the integrity and accuracy of the original data in each time interval, the system first processes each time interval entry and maps it to a predefined discrete state label based on the percentage values ​​of the integrity status (for example, the percentage of "complete and accurate", the percentage of "missing fields", etc.) and accuracy status (for example, the percentage of "accurate", the percentage of "format errors", the percentage of "value out of limit", etc.) of the records therein. This mapping rule is set according to the data quality requirements. For example, the integrity status label rule is set: if the sum of the percentage of records of "complete and accurate" plus "value out of limit" and "format errors" is greater than or equal to 98%, the integrity label is "high"; if the percentage is between 90% and 98%, the label is "medium"; if it is less than 90%, the label is "low". Similarly, the accuracy status label rule is set: if the sum of the percentage of records of "complete and accurate" plus "missing fields" is greater than or equal to 98%, the accuracy label is If the percentage is between 90% and 98%, the label is "medium"; if it is less than 90%, the label is "low". Apply these rules to the statistical percentage data of each time interval in the core data quality assessment set, and extract the corresponding integrity status label (such as "high") and accuracy status label (such as "medium"). At the same time, assign a program-internal unique identification number to each time interval in the database or internal data structure, such as using an auto-incrementing integer sequence starting from 1 or directly using a standard format time interval start time string (such as "2025051210") as the unique number. Subsequently, the unique identification number of each time interval is combined with its calculated integrity status label and accuracy status label into a structured record. All these records are collected to form a status mapping list, which clearly corresponds each uniquely identified time interval with a concise assessment label of its data quality to generate a numbered quality status label list.

[0086] Based on the numbered quality status tag list generated in the previous step and the hierarchical aggregated power plant revenue data structure formed in the earlier step, which contains information such as the aggregated power generation of each time interval (each time interval aggregation unit in this structure already has or is assigned a unique identification number consistent with the status tag list in this step), the system starts the data embedding process, which traverses each aggregation unit in the hierarchical aggregated power plant revenue data structure. These aggregation units represent different time intervals (for example, the data aggregation results of an hour or a day). For the aggregation unit currently traversed, its unique identification number is obtained, and this number is used to search in the numbered quality status tag list. By accurately matching the numbers, the integrity status tag (for example, "high", "medium" or "low") and accuracy status tag (for example, "high", "medium" or "low") corresponding to the time interval are located. ”), once the match is confirmed successfully, the completeness status tag value found is inserted as a new attribute into a specified field in the metadata area preset in the current aggregation unit, such as the field named 'data_completeness_level'. At the same time, the accuracy status tag value found is inserted into another specified field in the metadata area, such as the field named 'data_accuracy_level'. This search and insertion operation is repeated for all aggregation units under all time levels (such as hourly level, daily level, etc., if there is a multi-level structure) in the hierarchical aggregated power plant revenue data structure until all aggregation units have been successfully matched according to their time interval numbers and embedded with the corresponding completeness and accuracy status tags, and finally a hierarchical aggregated power plant revenue data structure with embedded quality status fields is obtained.

[0087] Based on the hierarchical aggregated power plant revenue data structure with embedded quality status fields, each aggregate unit (time interval) in the structure is now labeled with the 'data_completeness_level' and 'data_accuracy_level' fields. The system performs the final analysis and solidification steps. First, it traverses all aggregate units, checks their integrity status identification and accuracy status identification fields, and determines whether the unit status is abnormal based on the preset standards. The abnormal standard here is defined as: Aggregate units with a completeness status label of "low" or an accuracy status label of "low" are regarded as quality abnormal units. The total number of units judged to be quality abnormal in all aggregate units is counted. Next, these abnormal states are analyzed in the time level dimension. For example, if the basic aggregation unit is at the hourly level, the system will further count and mark them at higher time levels such as days, weeks or months, and calculate the number of abnormal states that occur daily, weekly or monthly. The number of hourly units with abnormal quality (low integrity or low accuracy) is counted, or whether there are multiple consecutive hourly units with abnormal quality, and cluster labels are generated based on these statistical results. For example, if the quality of more than 3 or 2 consecutive hourly units in a day is "low", then the daily aggregation unit (if any) can be assigned a cluster label of "frequent low-quality data within the day". If the quality of hourly units exceeding a certain proportion (for example, 20%) in a week is "low", then the weekly aggregation unit can be assigned a cluster label of "frequent low-quality data within the week". After completing the status statistics and cluster marking, the original aggregation content of each aggregation unit (such as total power generation), the newly embedded quality status label (completeness and accuracy level), and the generated cluster label (if any) are integrated together to form the final record structure, and the status of the entire data set is set to read-only to prevent subsequent accidental modifications, and finally a revenue data structure containing rich quality dimension information is generated.

[0088] The steps to obtain the target profit indicator query parameter package are as follows:

[0089] Based on the revenue data structure, after receiving a power plant revenue indicator analysis query request initiated by a user, the complete request body content submitted in the query request is extracted, and the keywords in the body content are marked as four types of elements: target power plant identification tag, revenue indicator name keyword, start time field and end time field, to generate a structured parsed field set;

[0090] Based on the structured parsing field set, check whether the target power plant identification tag exists in the registered power plant directory table in the revenue data structure. If a match is successful, read the revenue indicator field summary table corresponding to the power plant and compare whether the revenue indicator name keyword exists in the field summary table. At the same time, check whether the time format of the start time field and the end time field is legal and the range is reasonable, and generate a target revenue indicator field and time parameter set that passes the verification;

[0091] Based on the verified target profit indicator fields and time parameter sets, a composite object with four types of contents, namely, target power plant identification label, standardized profit indicator name, standardized start time field, and standardized end time field, is constructed to generate a target profit indicator query parameter package.

[0092] Specifically, based on the revenue data structure, after the system receives a power plant revenue indicator analysis query request initiated by the user through interface input or API call, for example, the request submitted by the user may be a text string "Show the daily trend data of the total power generation of XXA plant in the past month" or a JSON object containing specific parameters {"plantName":"XXA plant","indicator":"total power generation","period":"last_month","granularity":"daily"}, the system first extracts the content of the request body in full, and then uses predefined parsing rules and keyword matching logic to process the body. These rules include a regular expression pattern library to identify date and time expressions (such as "past month", "May 1 to May 10, 2025"), The target name (such as "total power generation" and "on-grid electricity price") and the power plant name or number are matched with the built-in power plant alias dictionary (for example, "XXA plant" is mapped to the internal ID "HNA001") and the indicator synonym library (for example, "total power generation" and "power generation performance" are both mapped to the standard indicator name "TOTAL_GENERATION"). The identified information is marked as the target power plant identification label (such as "HNA001"), the benefit indicator name keyword (such as "TOTAL_GENERATION"), the start time field (such as the specific start date "2025-04-12" calculated by combining "the past month" with the current date), and the end time field (such as the calculated "2025-05-11"). The set of extracted and preliminarily classified fields is the generated structured parsed field set.

[0093] Based on the structured parsed field set generated in the previous step, which includes the target power plant identification tag, profit indicator name keywords, start time field, and end time field extracted from the user request, the system performs a series of verification operations. First, the target power plant identification tag is used to query a registered power plant directory table maintained within the profit data structure. This table records all power plants registered in the system and with data stored, as well as their unique identifiers and basic information. For example, if the target power plant identification tag in the structured parsed field set is "HNA001", the system will check whether "HNA001" exists in the directory table. If the match fails, it will be considered an invalid power plant request. If the match succeeds, the system will then read the profit indicator field summary table associated with the power plant ("HNA001"). This summary table is also part of the profit data structure and lists the standard names and definitions of all profit indicators available for query for this specific power plant, such as "TOTAL_GENERATION", "NET_PROFIT", "AVG_CO2", etc. AL_COST", etc., and then the profit indicator name keywords in the structured parsing field set are fully matched with the standard names in this general table. If no match is found, it is judged as an invalid indicator request. At the same time, the system performs format and logic verification on the start time field and end time field in the structured parsing field set to check whether they conform to the system's preset standard date and time format, such as "YYYY-MM-DDHH:MM:SS" or ISO8601 standard, and verifies the rationality of the time range. For example, the start time must be earlier than or equal to the end time, and the query time range should not exceed the earliest date and latest date of the power plant data record in the system. These two date boundaries are usually obtained from the metadata of the profit data structure. For example, if the earliest data date recorded in the system is "2020-01-01" and the start time of the user query is "2019-12-01", the range is judged to be unreasonable. After all verifications pass, a target profit indicator field and time parameter set that passes the verification is formed.

[0094] Based on the verified target benefit indicator field and time parameter set obtained in the previous step, which already contains the verified target power plant identification tag, benefit indicator name keyword, and legal start and end times, the system performs the final parameter packaging processing. First, the time parameters are standardized. Even if the input start time field and end time field have multiple legal formats (such as "2025-05-01" or "2025 / 05 / 0110:00"), they are uniformly converted to the standard timestamp format used for internal system processing. For example, "2025-05-01" is converted to "2025-05-01T00:00:00Z" as the standardized start time field, and the user-specified end date (such as "2025-05-07") is converted to the end time of that date "2025-05-07T23: The standardized end time field is "59:59.999Z" to ensure the precise closure of the time range. The profit indicator name keyword may also be converted into a strictly corresponding unique standard indicator code within the system at this stage (for example, the user query "total power generation" may be converted into the internal code "GEN_TOTAL_MWH"). Finally, the four types of content, namely the confirmed valid target power plant identification tag (such as "HNA001"), the standardized profit indicator name (such as "GEN_TOTAL_MWH"), the standardized start time field (such as "2025-05-01T00:00:00Z"), and the standardized end time field (such as "2025-05-07T23:59:59.999Z"), are organized into a structured data object or parameter set. This combined object is the generated target profit indicator query parameter package.

[0095] The steps to obtain the revenue indicator query response are:

[0096] Based on the target profit indicator query parameter package, extract the target power plant identification tag, standardized profit indicator name, standardized start time field, and standardized end time field. Use the power plant identification tag to lock the corresponding power plant branch node in the profit data structure. Under the node, locate the aggregate field set corresponding to the target profit indicator name. Then, use the start time field and end time field to filter all aggregate field contents that meet the time conditions to generate the target profit indicator data set.

[0097] Based on the target revenue indicator data set, each aggregate field is traversed and the power generation aggregate value, completeness status identifier, and accuracy status identifier contained therein are uniformly parsed. The contents of each field in the same time interval are combined into a single result record. All records are sorted according to the time interval sequence and combined into a structured response body to generate a sorted target revenue indicator structured record set;

[0098] Based on the sorted target benefit indicator structured record set, a query response structure with a unified format is constructed, and the indicator name, power plant label, time range and structured record content required by the response body are filled in to generate a benefit indicator query response.

[0099] Specifically, based on the target profit indicator query parameter package, the parameter package encapsulates the target power plant identification tag, standardized profit indicator name, standardized start time field and standardized end time field after the user query request is parsed and verified. The system first extracts these four core parameters from the parameter package, and then uses the target power plant identification tag (for example, "HNA001") to navigate in the constructed profit data structure. The profit data structure may be layered or indexed by power plant ID, then by indicator name, and then by time, so as to quickly locate the data branch node of the specific power plant. Under this power plant branch node, the system further uses the standardized profit indicator name (for example, "GEN_TOTAL_MWH") to find the corresponding aggregate field set. This set contains a series of aggregate data points of the indicator changing over time. Each data point records The indicator values ​​and associated quality information of a specific time interval. Then, the system performs time range filtering on the located aggregate field set based on the standardized start time field (for example, "2025-05-01T00:00:00Z") and the standardized end time field (for example, "2025-05-07T23:59:59.999Z") in the parameter package, and filters out all aggregate field content entries whose timestamps (or time interval identifiers) fall within the specified query range. For example, if the timestamp of an entry in the aggregate field set is "2025-05-03T10:00:00Z", the entry meets the filtering conditions, while the entry with the timestamp of "2025-04-30T23:00:00Z" is excluded. After filtering, all eligible aggregate field contents together constitute the target revenue indicator data set.

[0100] According to the target profit indicator data set generated in the previous step, which contains all the original aggregated data points of the specified power plant and the specified indicator within the specified time range, the system will traverse each aggregate field content in this data set. For each content, it represents the data of a specific time interval. The system uniformly parses out the core data items, including the aggregate value corresponding to the queried indicator (for example, if the total power generation is queried, it is the total power generation MWh value of the time interval), as well as the integrity status identifier (for example, "high", "medium", "low") and accuracy status identifier (for example, "high", "medium", "low") embedded in the previous step. These information parsed from the same aggregate field content, namely the time interval identifier, indicator aggregate value, integrity status identifier and accuracy status identifier, are combined and packaged into one. A structured result record, for example, forms a single record such as {"interval_timestamp":"2025-05-03T10:00:00Z","indicator_value":150.75,"completeness_status":"High","accuracy_status":"High"}. This parsing and combination process is repeated for all aggregated field contents in this data set to generate a set of result records. Subsequently, this set of result records is strictly sorted in ascending order according to the order of their time interval identifiers to ensure the time continuity of the output data. After the arrangement is completed, all records are combined together to form an ordered, structured data list, which is the sorted target yield indicator structured record set.

[0101] Based on the sorted target benefit indicator structured record set generated in the previous step, which already contains the data of each time interval arranged in chronological order, including the indicator value and quality status, the system begins to construct the query response structure that is ultimately returned to the user. This structure adopts a unified, pre-defined format, usually JSON or XML format, to ensure the standardization and ease of use of the interface. The response structure mainly consists of two parts: metadata information and result data. First, fill in the metadata information part, which includes the query parameters obtained from the target benefit indicator query parameter package and echoed to the user, such as the standardized benefit indicator name of the queried indicator (such as "GEN_TOTA L_MWH"), the target power plant identification tag (such as "HNA001"), and the standardized start time field and standardized end time field used for the query (such as "2025-05-01T00:00:00Z" to "2025-05-07T23:59:59.999Z"). Then, the sorted target profit indicator structured record set is embedded as a whole into the result data part of the response structure, for example, as an array named 'data_points'. Each element in the array is a single structured result record generated in the previous step. After all information is filled in, the system generates a complete profit indicator query response.

[0102] The steps to obtain the preliminary return volatility identification sequence are:

[0103] Based on the revenue index query response, extract the revenue index value field, equipment load rate field, and timestamp field from all records. Convert the equipment load rate field to a decimal format as a percentage, uniformly convert the timestamp field to a real value in hours, and sort them in ascending time order. After removing the first and last time points, generate a standardized revenue index time series structure.

[0104] According to the time series structure of the standardized return indicator, the normalized multi-factor volatility coefficient corresponding to each time point is calculated. The calculation formula is:

[0105]

[0106] Among them, F c is the normalized multi-factor volatility coefficient at the cth time point, S c is the profit index value at the cth time point, S c-1 is the profit index value at the c-1th time point, U c is the equipment load rate at the cth time point (converted to a decimal between 0 and 1), U c-1 is the equipment load rate at the c-1th time point (converted to a decimal between 0 and 1), TS c The timestamp of the c-th time point (unit: hour), TS c-1is the timestamp of the c-1th time point (unit: hour), γ is a set non-zero constant used to prevent the denominator from being zero;

[0107] Based on the normalized multi-factor volatility coefficient, the entire time series structure is traversed, and the time points where the normalized multi-factor volatility coefficient is more than twice the average value of the two adjacent points are marked as candidate nodes for abnormal returns. All marked time points are combined in timestamp order to form a candidate point sequence, generating a preliminary return fluctuation identification sequence.

[0108] Specifically, based on the benefit indicator query response, the response body contains a structured record list sorted by time. Each record contains information such as a time interval identifier, an indicator aggregation value, and a data quality status. The system first extracts the core fields required for fluctuation analysis from these records, namely the timestamp field (for example, the start time of the time interval to which the record belongs, "2025-05-03T10:00:00Z"), the query target benefit indicator value field corresponding to the timestamp (for example, "indicator_value" is 150.75), and the equipment load rate field at the corresponding time point. The equipment load rate data needs to be obtained synchronously with the benefit indicator data, and may be directly included in the record of the benefit indicator query response, or obtained by associating it with the load rate record in the power plant operation database through the timestamp. For example, if the load rate is 85.0%, after extracting these fields, the equipment load rate field is formatted and converted from a percentage (such as 85.0%) to a value between 0 and 1. Next, the timestamp fields are uniformly converted from the standard date and time format to continuous real values ​​in hours. For example, the first timestamp in the sequence can be set to time zero, and subsequent timestamps can be calculated as the difference in hours relative to time zero. Alternatively, the timestamps can be uniformly converted to hours since a fixed historical moment (such as the Unix epoch). For example, "2025-05-03T10:00:00Z" might be converted to 483250.0 hours. After the conversion is complete, ensure that all records are sorted in ascending order according to the converted hourly timestamps (usually the input responses are already sorted, and this step is a verification step). Then, remove the data records at the first and last time points in the time series. This operation is used to avoid boundary issues when subsequently calculating the difference. After the above extraction, conversion, sorting, and removal operations, a standardized time series structure containing the triples of time (hours), revenue indicator value, and equipment load rate (decimal) is generated. This is the standardized revenue indicator time series structure.

[0109] formula: The formula is useful in that it provides a method for measuring the volatility of the return indicator at a specific time point c relative to the previous time point c-1, and this measurement takes into account the changes in the return indicator itself (S c -Sc-1 ) and equipment operating status (in terms of load rate U c changes in (U c -U c-1 ) coupling effect, the molecular term |(S c -S c-1 )(U c -U c-1 )| means that only when the revenue indicator and equipment load rate change significantly in the same or opposite direction at the same time, the absolute value of their product will increase significantly, reflecting a stronger co-movement. If one of them changes very little, even if the other changes very much, the product will be small, indicating that such unilateral changes may not be regarded as a typical multi-factor volatility event that requires attention. The denominator (TS c -TS c-1 ) 2 +γ normalizes the time interval so that the fluctuation coefficient is not affected by the data sampling frequency. That is, if the index and load factor of the same magnitude change in a shorter time, the calculated F c The value will be larger, indicating more severe fluctuations. The introduction of the constant γ ensures that the denominator is not zero, which increases the stability of the calculation. The entire formula can effectively identify the time points when the profit indicators and the equipment operating status undergo rapid and significant synchronous changes by calculating the normalized multi-factor linkage change amplitude, providing a quantitative basis for the subsequent identification of potential abnormal fluctuations in profits.

[0110] S c With S c-1 The steps to obtain the parameters are: c Represents the profit index value at the cth time point in the standardized profit index time series structure, S c-1 Represents the profit index value of the previous (c-1) time point. These two values ​​are directly read in chronological order from the generated standardized profit index time series structure. Each record in the structure contains the profit index value of the corresponding time. The unit is determined according to the specific indicator of the query (for example, power generation is MWh, net profit is 10,000 yuan, etc.). When obtaining, directly read the indicator value field in the corresponding time point record. For example, for the c=5th time point, the profit index value recorded at this time point is S5=155.0, and the profit index value recorded at the previous time point (c-1=4) is S4=150.0.

[0111] U c with U c-1 The steps to obtain the parameters are: c Represents the equipment load rate at the cth time point in the time series structure of the standardized revenue indicator, U c-1Represents the equipment load rate at the previous (c-1th) time point. Both values ​​are converted into decimal format between 0 and 1, and are also read directly from the standardized profit indicator time series structure in chronological order. Each record in the structure also contains the equipment load rate (decimal format) at the corresponding time. When obtaining, the load rate field in the corresponding time point record is read. For example, for the c=5th time point, the equipment load rate recorded at this time point is read as U5=0.90 (representing 90%), and the equipment load rate recorded at the previous time point (c-1=4) is read as U4=0.85 (representing 85%).

[0112] TS c With TS c-1 The steps to obtain parameters are: TS c Represents the timestamp of the cth time point in the time series structure of the standardized return indicator, TS c-1 Represents the timestamp of the previous (c-1th) time point. Both values ​​have been uniformly converted into real values ​​in hours and can be read directly from the standardized return indicator time series structure in chronological order. Each record in the structure contains the hourly real value timestamp of the corresponding time. For example, for the c=5th time point, its timestamp is read as TS5=483251.0 hours, and the timestamp of the previous time point (c-1=4) is read as TS4=483250.0 hours, which means that the two time points are separated by 1 hour.

[0113] The steps to obtain the γ parameter are as follows: γ is a pre-set non-zero constant, whose main function is to prevent the denominator of the formula from being affected by the time difference (TS c -TS c-1 ) is extremely small or zero (which should not happen in theory, but is set to ensure robustness), resulting in calculation errors or numerical instability.

[0114] Calculation process: Select the c=5th time point in the time series structure of the standardized return indicator for calculation, and obtain the following parameters according to the above parameters: S5=155.0, S4=150.0, U5=0.90, U4=0.85, TS5=483251.0h, TS4=483250.0h, γ=0.000001;

[0115] Calculate the numerator term |(S c -S c-1 )(U c -U c-1 )|:

[0116] |(155.0-150.0)(0.90-0.85)|=|(5.0)(0.05)|=|0.25|=0.25;

[0117] Calculate the denominator (TS c -TS c-1 ) 2 +γ:

[0118] (483251.0-483250.0) 2 +0.000001=(1.0) 2 +0.000001=1+0.000001=1.000001;

[0119] Calculate F c :

[0120]

[0121] The results show that at the c=5th time point, the normalized multi-factor volatility coefficient F5 is about 0.50. This value quantifies the intensity of the joint change of the profit index and the equipment load rate at this time point relative to the previous time point, and after the normalization of the time interval, F c The larger the value, the more significant the linkage fluctuation at that time point.

[0122] Based on the normalized multi-factor volatility coefficient F covering the entire time series calculated in the previous step c Sequence, the system then traverses this F c Sequence (since the first and last data points were removed in the previous step, this F c The index c of the sequence is calculated from 2 and ends at the position where the original sequence length minus 2. For each point c in the sequence (the range of c is from F2 calculated from the second point of the original sequence to F calculated from the second to last point), N-1 , where N is the length of the sequence after removing the head and tail), a dynamic threshold rule is applied to determine whether it is an abnormal fluctuation point. The rule is: compare the fluctuation coefficient F of the current point c The previous point F c-1 and the next point F c+1 The fluctuation coefficient is twice the average value, that is, the judgment condition is That is, F c >F c-1 +F c+1 This rule aims to capture the "peak" points where the fluctuation coefficient is significantly higher than its local average level. For example, if the F c =0.8, and the previous point F c-1 =0.2, the last point F c+1 =0.3, then F c-1 +F c+1 =0.5, because 0.8>0.5, so the condition is met, and the time point c will be marked as a "candidate node for abnormal returns". If F c=0.4, and F c-1 +F c+1 =0.5, then 0.4≯0.5, this point is not marked, the system is F c All c points in the sequence that meet this condition (from c=2 to c=N-2) are marked, and all successfully marked time points are collected in the order of their timestamps in the original time series and combined into a time series of candidate points. This sequence only contains those time points that are preliminarily judged to have significant return fluctuations, which is the generated preliminary return fluctuation identification sequence.

[0123] The steps to obtain the return volatility analysis summary are:

[0124] Based on the preliminary return fluctuation identification sequence, the composite credibility score of the overall fluctuation pattern is calculated using the following formula:

[0125]

[0126] Among them, L m is the composite credibility score corresponding to the current preliminary return volatility identification sequence, r is the total number of time points identified as volatility in the sequence, and A j is the change in income at the jth time point (unit: 10,000 yuan), Θ j is the change in equipment load rate at the jth time point (unit: percentage, converted to a decimal between 0 and 1), W j is the data integrity status level at the jth time point (0 means no missing, increasing means the severity of missing), Temp j is the device operating temperature at the jth time point (unit: Celsius), δ is a constant in Celsius to avoid zero values ​​in the square root;

[0127] Based on the composite credibility score, and compared with the established credibility score standard curve, determine whether the current score is higher than the upper limit of the credibility threshold within the 30-day rolling interval. If it is higher, the fluctuation sequence will be marked as "abnormally significant high volatility", otherwise it will be marked as "regular volatility", and a return volatility analysis summary will be generated.

[0128] Specifically, the formula: The usefulness of the formula is that it provides a quantitative score L for comprehensively evaluating the overall credibility and strength of the "preliminary return fluctuation identification sequence" m It not only counts the number of fluctuation points, but also analyzes the specific characteristics of each fluctuation point j in depth, including the absolute amplitude of the return change |A j |, the significance of the change in equipment load rate The integrity of relevant data j And the operating temperature Temp j , numerator Higher weights are given to fluctuation points with large revenue changes and significant load factor changes. In particular, the impact of load factor changes is amplified by the square, highlighting the revenue fluctuations when the operating status changes drastically. A penalty mechanism is introduced in the denominator: Data integrity status level W j The worse (the larger the value), (W j +1) is larger, the contribution of this point is reduced, which reflects the distrust of fluctuation events supported by low-quality data; at the same time, the device operating temperature Temp j The higher the denominator The larger the value, the lower the contribution of the point, which reflects that the tolerance of fluctuation under high temperature increases or it is normal. The constant δ ensures that the square root and the denominator are not zero. The above weighted contributions of all r fluctuation points in the sequence are summed and based on the number of fluctuation points r (through Factor) is adjusted and averaged, and the final L m The score comprehensively reflects the average "abnormal signal strength" of the entire fluctuation sequence and takes into account the impact of data quality and operating environment, making it a more reliable basis for evaluating fluctuation patterns than simple counting or single indicator fluctuations.

[0129] The steps for obtaining the r parameter are: this parameter represents the total number of time points identified as fluctuations contained in the "preliminary return fluctuation identification sequence" generated in the previous step. The acquisition method is to directly count the elements in the sequence (that is, the time points marked as candidate nodes for return anomaly). For example, if the preliminary return fluctuation identification sequence contains three time points: T5, T12, and T28, then r = 3.

[0130] A j The steps for obtaining the parameter are as follows: This parameter represents the income change value at the jth time point in the preliminary income fluctuation identification sequence, in units of 10,000 yuan. It quantifies the specific income amount change that occurs at this fluctuation point. This value is based on the "income index value" S of the time point c and the previous time point c-1 in the original time series. c and S c-1 The calculated difference S c -S c-1 , and convert the units as needed. For example, if the unit of the income indicator S is megawatt-hour (MWh), it is necessary to convert it into income (yuan) based on the then on-grid electricity price (yuan / MWh), and then divide it by 10,000 to get 10,000 yuan. The calculation formula example is: A j =(S c -S c-1 )×Electricity Price c / 10000, where the electricity price must be obtained from the relevant economic database or contract at time point c. For example, for the j = 1st fluctuation point in the sequence (corresponding to the c = 5th point in the original sequence), if S5 = 155.0 MWh and S4 = 150.0 MWh, and the electricity price at that time was 300 yuan / MWh, then A1 = (155.0 - 150.0) × 300 / 10000 = 5.0 × 300 / 10000 = 150,000 yuan. In this example, the initial revenue fluctuation identification sequence is set to include r = 2 points, with A1 = 150,000 yuan at the j = 1st point and A2 = -200,000 yuan (indicating a decrease in revenue) at the j = 2nd point.

[0131] Θ j The steps for obtaining the parameter are as follows: This parameter represents the change amplitude of the equipment load rate at the jth time point in the preliminary income fluctuation identification sequence, which is the absolute value of the difference between the equipment load rate at the point c and the previous point c-1 in the original time series (converted to a 0-1 decimal), that is, Θ j =|U c -U c-1 |, which reflects the severity of the change in the operating state of the equipment at the fluctuation point, has no unit. For example, for the j=1 fluctuation point in the sequence (corresponding to the c=5 point in the original sequence), if U5=0.90, U4=0.85, then Θ1=|0.90-0.85|=0.05. For the j=2 fluctuation point (for example, corresponding to the c=12 point in the original sequence), if U 12 =0.70, U 11 =0.80, then Θ2 = |0.70-0.80| = |-0.10| = 0.10.

[0132] W j The steps for obtaining the parameter are as follows: This parameter represents the data integrity status level of the original data corresponding to the jth time point in the preliminary revenue fluctuation identification sequence. It is a non-negative integer used to quantify the severity of data missing. The value comes from the integrity status identification (for example, "high", "medium", and "low") of each aggregate unit in the "hierarchical aggregated power plant revenue data structure with embedded quality status fields" in the early step. It is necessary to establish a clear mapping rule to quantify these status labels into level values. For example, set the mapping rule: the integrity status label "high" corresponds to W j =0 (no missing or very few missing), labelled as “medium” corresponding to W j =1, label is “low” corresponding to W j =2, when obtaining, according to the timestamp of fluctuation point j, find the record of the corresponding time interval in the data structure embedded in the quality status field, read the value of its 'data_completeness_level' field, and then apply the above mapping rules to obtain Wj For example, if the time interval integrity label corresponding to the j=1th fluctuation point is "high", then W1=0; if the time interval integrity label corresponding to the j=2th fluctuation point is "medium", then W2=1.

[0133] Temp j The steps for obtaining the parameter are as follows: the parameter represents the equipment operating temperature corresponding to the j-th time point in the preliminary profit fluctuation identification sequence, in degrees Celsius (℃). This value is directly obtained by extracting the temperature field from the original power plant operation data record corresponding to the timestamp of the fluctuation point. The temperature field has been associated and checked in the previous steps. For example, query the original record that matches the timestamp of the j=1 fluctuation point (corresponding to the c=5 point in the original sequence), and read its temperature field value as Temp1=78.0℃. Query the original record that matches the timestamp of the j=2 fluctuation point (corresponding to the c=12 point in the original sequence), and read its temperature field value as Temp2=82.5℃.

[0134] The steps to obtain the δ parameter are as follows: δ is a preset positive constant in degrees Celsius, which is added to the temperature value Temp j To prevent Temp j When the value is extremely low (e.g. close to zero or negative, which is not common in power plant operation but needs to be considered as abnormal), the denominator The result of the calculation is zero or a complex number.

[0135] Calculation process: Set the preliminary return fluctuation identification sequence to contain r = 2 fluctuation points, and its parameters are as follows: point j = 1: A1 = 1500 yuan, Θ1 = 0.05, W1 = 0, Temp1 = 78.0℃, point j = 2: A2 = -2000 yuan, Θ2 = 0.10, W2 = 1, Temp2 = 82.5℃, constant δ = 0.01℃.

[0136] Calculate the contribution of each fluctuation point j For j=1: Numerator: Denominator:

[0137] Contribution

[0138] For j=2: Numerator: Denominator: Contribution

[0139] Calculate L m :

[0140]

[0141] The results show that the composite credibility score of the overall fluctuation pattern of the current preliminary return fluctuation identification sequence containing two fluctuation points is 0.012026. This score combines factors such as fluctuation amplitude, load change, data quality and operating temperature. The value itself represents the average "signal strength" or "credibility" of the fluctuation pattern. A higher score usually means a more significant and credible fluctuation pattern. m The value will be used as the basis for the next step to determine whether the fluctuation pattern belongs to "abnormally significant high fluctuation" and needs to be compared with the upper limit of the dynamically set credible threshold.

[0142] Based on the composite credibility score L calculated in the previous step m , for example L m = 0.012026, the system then compares it with a dynamically adjusted upper limit of the credibility threshold. This threshold is dynamically calculated by referring to an established credibility scoring standard curve and combining the historical scoring data of the recent period (for example, 30 days). First, the credibility scoring standard curve is based on a large number of historical fluctuation events and their corresponding L m The score is established after analysis, it may be L m The scoring range is divided into different levels (such as low, medium, and high confidence), or a historical abnormal probability reference corresponding to a score is given to help understand the relative position of the current score. Secondly, in order to adapt to the cyclical changes and recent trends in the operating status of the power plant, the "upper limit of the credible threshold within the 30-day rolling interval" is used to set the specific judgment limit. The calculation method of the threshold is: collect the daily or event-driven L calculated for each day in the past 30 days m The score value is formed into a data set containing the latest 30 scores, and then a higher percentile of this data set is calculated, such as the 90th percentile or the 95th percentile, and it is used as the upper limit of the credible threshold for the day. For example, if L m The 90th percentile calculated from the scoring data is 0.0150, so the upper limit of the credible threshold for the day is 0.0150. The system will calculate the current L m =0.012026 is compared with the threshold value 0.0150. If L m If L is higher than the upper threshold (0.012026≯0.0150 in this example), the current preliminary return volatility identification sequence is marked as "abnormally significant high volatility", indicating that the volatility pattern is considered to be unconventional and worthy of high attention. mIf the value is not higher than the upper threshold (as in this case), the fluctuation sequence is marked as "normal fluctuation", indicating that although the fluctuation pattern has been initially identified, its comprehensive strength and credibility have not reached the level that requires special warning. Finally, this marking result ("abnormally significant high fluctuation" or "normal fluctuation") is recorded together with the L m The score value and other information together constitute and generate the final return fluctuation analysis summary.

[0143] The above are merely preferred embodiments of the present invention and do not limit the present invention in any other form. Any technician familiar with the profession may use the technical content disclosed above to change or modify it into an equivalent embodiment with equivalent changes and apply it to other fields. However, any simple modification, equivalent change and modification made to the above embodiment based on the technical essence of the present invention without departing from the content of the technical solution of the present invention shall still fall within the scope of protection of the technical solution of the present invention.

Claims

1. A method for deep mining of power plant profit indicators based on historical operating data, characterized in that: The following steps are involved: Based on the input power plant operation data, the core revenue dimension time and the key revenue metric power generation are selected for initial structure design. The hierarchical division standard of the revenue dimension time is set, and the key revenue metric power generation is aggregated according to the set time hierarchy to establish a hierarchical aggregated power plant revenue data structure. Based on the hierarchical aggregated power plant revenue data structure, for the aggregated power generation, evaluating the data integrity status and data accuracy of the corresponding original power plant operation data to obtain a core data quality assessment set; based on the core data quality assessment set, embedding the data integrity status and data accuracy into corresponding aggregation units of the hierarchical aggregated power plant revenue data structure to generate a revenue data structure; Based on the revenue data structure, responding to a real-time analysis query request for a target power plant revenue indicator, parsing the power plant revenue indicator name and time range parameters required by the query request to obtain a target revenue indicator query parameter package, retrieving the corresponding power plant revenue indicator from the revenue data structure based on the target revenue indicator query parameter package, combining the results into a single query response, and establishing a revenue indicator query response; Based on the profit indicator query response, the trend of the power plant profit indicator over time is examined, the time point of profit fluctuation is identified, and a preliminary profit fluctuation identification sequence is obtained. Based on the preliminary profit fluctuation identification sequence and combined with the information attached to each data point in the sequence, the credibility score of the overall fluctuation pattern is calculated, and a profit fluctuation analysis summary is established.

2. The method for deep mining of power plant profit indicators based on historical operation data according to claim 1 is characterized in that: The steps for obtaining the hierarchical aggregated power plant revenue data structure are: Based on the input power plant operation data, the operation time field and power generation field in each record are extracted. The operation time field is uniformly converted into a standard timestamp format and divided according to the preset units. The corresponding power generation field within each divided time unit range is then grouped and aggregated to generate a power generation aggregate distribution set with a fixed time scale. Calculating a composite value of power generation offset in each time interval based on the aggregated distribution set of power generation with a fixed time scale; According to the composite value of power generation offset, determine whether it exceeds the upper limit of the power generation offset reference threshold. If the condition is met that the composite value of power generation offset is greater than the upper limit of the power generation offset reference threshold, then the time interval is marked as an abnormal operating profit interval; otherwise, it is marked as a normal profit interval, forming a hierarchical aggregated power plant profit data structure.

3. The method for deep mining of power plant profit indicators based on historical operation data according to claim 1 is characterized in that: The steps for obtaining the core data quality assessment set are: Based on the hierarchical aggregated power plant revenue data structure, the original power plant operation data entries associated with the aggregated power generation in each time interval are parsed, and the timestamp, power generation field, operation duration field, temperature field, and equipment number field contained in each original power plant operation data are extracted to obtain the associated original power plant operation data set; Based on the associated original power plant operation data set, check each time whether the timestamp field is missing or has a format error, determine whether the power generation field, the operation time field, and the temperature field are missing or have abnormal value ranges, and mark each original power plant operation data with a completeness and accuracy dual status to form an original data quality status mark set; Based on the original data quality status mark set, the percentage of the completeness and accuracy status of all original power plant operation data in each time interval is counted to form a core data quality assessment set.

4. The method for deep mining of power plant profit indicators based on historical operation data according to claim 1 is characterized in that: The steps for obtaining the income data structure are: Based on the core data quality assessment set, extract the integrity status label and accuracy status label corresponding to each time interval one by one, assign a unique identification number to each time interval, form a structured state mapping list with the status label and the number, and generate a numbered quality status label list; According to the numbered quality status tag list, traverse the aggregation units at all time levels in the hierarchical aggregated power plant revenue data structure, compare the time interval numbers of the aggregation units with the numbers in the status tag list in turn, and after confirming a successful match, insert the integrity status tag into the metadata area field of the aggregation unit, and at the same time insert the accuracy status tag into the metadata area field of the aggregation unit, to obtain the hierarchical aggregated power plant revenue data structure with the quality status field embedded; Based on the hierarchical aggregated power plant revenue data structure with embedded quality status fields, the integrity status identification fields and accuracy status identification fields of all aggregated units are traversed, the number of units with abnormal status is counted, and each type of status is clustered and marked under each time level dimension, and corresponding cluster labels are generated. The quality status, cluster labels and aggregated content of all aggregated units are uniformly saved and set to read-only state to generate a revenue data structure.

5. The method for deep mining of power plant profit indicators based on historical operation data according to claim 1 is characterized in that: The steps for obtaining the target revenue indicator query parameter package are as follows: Based on the revenue data structure, after receiving a power plant revenue indicator analysis query request initiated by a user, the complete request body content submitted in the query request is extracted, and the keywords in the body content are marked as four types of elements: target power plant identification tag, revenue indicator name keyword, start time field and end time field, to generate a structured parsed field set; Based on the structured parsed field set, verify whether the target power plant identification tag exists in the registered power plant directory table in the revenue data structure. If a match is successful, read the revenue indicator field summary table corresponding to the power plant and compare whether there is a complete match for the revenue indicator name keyword in the field summary table. At the same time, check whether the time format of the start time field and the end time field is legal and the range is reasonable, and generate a target revenue indicator field and time parameter set that passes the verification; Based on the verified target profit indicator field and time parameter set, a combination object with four types of content: target power plant identification label, standardized profit indicator name, standardized start time field and standardized end time field is constructed to generate a target profit indicator query parameter package.

6. The method for deep mining of power plant profit indicators based on historical operation data according to claim 1, characterized in that: The steps for obtaining the revenue indicator query response are: Based on the target profit indicator query parameter package, the target power plant identification tag, the standardized profit indicator name, the standardized start time field, and the standardized end time field are extracted; the power plant identification tag is used to lock the corresponding power plant branch node in the profit data structure; the aggregate field set corresponding to the target profit indicator name is located under the node; and all aggregate field contents that meet the time conditions are filtered according to the start time field and the end time field to generate the target profit indicator data set; Based on the target benefit indicator data set, traverse each aggregate field content and uniformly parse the included power generation aggregate value, completeness status identifier, and accuracy status identifier, combine the field contents within the same time interval into a single result record, sort all records according to the time interval sequence and combine them into a structured response body, and generate a sorted target benefit indicator structured record set; Based on the sorted target profit indicator structured record set, a query response structure in a unified format is constructed, and the indicator name, power plant tag, time range and structured record content required by the response body are filled in to generate a profit indicator query response.

7. The method for deep mining of power plant profit indicators based on historical operation data according to claim 1 is characterized in that: The steps for obtaining the preliminary return fluctuation identification sequence are: Based on the revenue index query response, extract the revenue index value field, equipment load rate field, and timestamp field from all records, convert the equipment load rate field into a decimal format according to the percentage, uniformly convert the timestamp field into a real value in hours, and sort them in ascending time order. After removing the first and last time points, generate a standardized revenue index time series structure; Calculate the normalized multi-factor volatility coefficient corresponding to each time point based on the time series structure of the standardized return indicator; Based on the normalized multi-factor volatility coefficient, the entire time series structure is traversed, and the time points where the normalized multi-factor volatility coefficient is more than twice the average value of the two adjacent points are marked as candidate nodes for abnormal returns. All marked time points are combined in timestamp order to form a candidate point sequence, and a preliminary return fluctuation identification sequence is generated.

8. The method for deep mining of power plant profit indicators based on historical operation data according to claim 1, characterized in that: The steps for obtaining the return fluctuation analysis summary are as follows: Calculating a composite credibility score for the overall volatility pattern based on the preliminary return volatility identification sequence; Based on the composite credibility score, and against the established credibility score standard curve, a determination is made as to whether the current score is higher than the upper limit of the credibility threshold within the 30-day rolling interval. If so, the fluctuation sequence is marked as "abnormally significant high volatility", otherwise it is marked as "normal volatility", and a return volatility analysis summary is generated.

Images