An online efficiency running state detection method and system based on industrial flow data

Abstract

The present application relates to an online efficiency operation state detection method and system based on industrial flow data, which comprises: periodically aggregating original data flow to generate periodic sampling records, and constructing newly added decision unit blocks characterized by input-output vectors; based on the efficiency and multiplier solution of historical decision units, calculating the default value of each historical decision unit corresponding to the newly added unit, determining the set of historical affected units, and identifying directly confirmable efficient / very efficient newly added units, and updating the efficient / very efficient set; constructing the evaluation set and its necessary reference mapping, and solving the efficiency and multiplier solution on the limited scale sub-sample reference set constrained by the mapping; finally, based on the change of the very efficient set and the change of the efficiency value, the operation state detection result is generated. Through incremental local update and limited scale solving, the online efficiency evaluation and interpretable early warning output of multi-input and multi-output production units under flow data are realized.

Description

Technical Field

[0001] This invention relates to the field of industrial data processing and online monitoring technology, specifically to an online efficiency operation status detection method and system based on industrial flow data. Background Technology

[0002] In industrial production sites, real-time monitoring platforms based on sensors, automated control systems, and information systems are typically deployed to continuously collect and display equipment operating status, production progress, energy consumption levels, and quality indicators. In manufacturing enterprises with multiple factories, work teams, or production lines, this type of on-site data is often generated continuously at a frequency of minutes or even seconds. Existing industrial monitoring systems generally use key performance indicator (KPI) dashboards, with indicators such as overall equipment utilization, downtime, number of micro-stops, energy consumption levels, and pass rates as core metrics for real-time display and management. These dashboards are usually based on pre-set thresholds and empirical rules to monitor the production process, primarily answering questions such as "Has a downtime occurred?", "Has the indicator exceeded the threshold?", and "Has the predetermined goal been achieved?". Essentially, this type of system focuses on the execution of established production specifications and operating rules, that is, judging whether the production process deviates from preset standards.

[0003] However, in actual production, even if all key performance indicators (KPIs) are within normal ranges and overall equipment utilization meets requirements, enterprises may still face problems such as resource shortages, insufficient capacity, or limited overall performance improvement. The key reason is that KPI dashboards emphasize single-indicator achievement and threshold warnings, making it difficult to characterize "resource allocation efficiency" under the comprehensive constraints of multiple inputs and outputs, and also making it difficult to conduct comparable horizontal evaluations of the operation of different production units under the same criteria. Therefore, managers often need to further answer the following more decision-making-value questions: Among production units with similar resource conditions, which has higher and replicable overall operating efficiency, and which consistently maintains a lower level of overall operating efficiency? How does a newly emerging superior efficiency performance affect the system's efficiency pattern? When the system's efficiency pattern changes, which units should be prioritized? Data Envelopment Analysis (DEA) can evaluate the relative efficiency of similar decision-making units under multiple input and output conditions and characterize the production frontier composed of effective samples, thus providing theoretical tools for "horizontally comparable efficiency evaluation" and "best practice identification." However, traditional DEA methods are mainly geared towards offline, batch-based historical data analysis. In industrial flow data scenarios, if the snapshot of the operating status of each production unit within each sampling period is considered as a decision-making unit, the set of decision-making units will continuously accumulate as data continues to flow in. If a full DEA recalculation is performed on all historical decision-making units at the end of each new period, the size and number of linear programming problems to be solved will continue to grow, leading to unbearable computational load and failing to meet the requirements of online or near real-time analysis. Furthermore, the addition of new data often only affects the efficiency evaluation results and frontier structure of some historical decision-making units, resulting in significant computational redundancy during the full recalculation.

[0004] Therefore, in industrial data flow scenarios, there is an urgent need for an efficiency evaluation and operational monitoring method oriented towards online updates. This method not only needs to solve the problem of computational efficiency and realize an online computing mechanism of "incremental access, partial updates, and limited-scale solutions" to ensure the consistency and correctness of data envelopment analysis results, but more importantly, it must transform the structural evolution of the efficiency frontier and the changes in efficiency value sequences revealed during the update process into a series of structured, interpretable, and verifiable early warning information and decision-making clues. Through this mechanism, the system can dynamically track changes in the production frontier under the continuous drive of data flow, sensitively capture structural shocks caused by new operational states, accurately locate new state points that can influence the system, and continuously assess whether each production unit has long-term inefficiency or short-term abnormal efficiency fluctuations. Ultimately, the system needs to automatically answer the core questions that managers care about, such as comprehensive efficiency comparison, best practice identification, shock source judgment, and unit state assessment, thereby providing in-depth decision support for production optimization and resource allocation that goes beyond simple threshold alarms, achieving a leap from "monitoring whether standards are met" to "understanding why efficiency is high." Summary of the Invention

[0005] This invention provides an online efficiency operation status detection method and system based on industrial streaming data. Its purpose is to solve the technical problems that existing industrial monitoring systems are unable to conduct online and efficient horizontal efficiency evaluation of production units with multiple inputs and outputs in a streaming data environment, and cannot automatically convert the efficiency evaluation results into interpretable decision-making warnings.

[0006] To achieve the above objectives, the first aspect of the present invention provides an online efficiency operation status detection method based on industrial flow data, comprising the following steps: The raw data streams of multiple homogeneous production units are periodically summarized to generate periodic sampling records for each production unit in each sampling period. A new decision unit block is constructed based on the periodic sampling records, and each decision unit in the new decision unit block is represented by an input vector and an output vector. Obtain the historical decision unit set up to the previous sampling period, along with its efficiency value and multiplier solution. Based on the default values ​​between the historical decision units and each decision unit in the newly added decision unit block, determine the historical affected set and identify the decision units in the newly added decision unit block that can be directly confirmed as effective or highly effective. This yields the newly added effective decision units and the newly added highly effective decision units. Update the current effective set and highly effective set. Based on the historical affected set and the decision units in the newly added decision unit block that were not directly identified as effective or highly effective, construct a set to be evaluated, and construct a mandatory reference set mapping for the decision units in the set to be evaluated. Using the required reference set mapping as a constraint, the efficiency of the decision units in the set to be evaluated is solved on the limited-size subsample reference set to obtain the efficiency value and multiplier solution of each decision unit. Based on the efficiency value and multiplier solution, the newly added highly efficient decision unit, and the highly efficient set and efficiency value of the previous sampling period, the operation status detection result is generated.

[0007] Furthermore, methods for periodically summarizing the raw data streams of multiple homogeneous production units to generate periodic sampling records for each production unit in each sampling period include: Based on the preset sampling period length and period index, the start and end times of the current sampling period are determined to form a period time window; For each production unit, various data segments whose timestamps fall within the cycle time window are extracted from the raw data stream, including operation status data segments, event log data segments, energy consumption metering data segments, and output and quality count data segments. The extracted data fragments are processed for field normalization, mapping the field names, data types and units of measurement to a preset standard format. Basic data quality marking is performed on each of the extracted data segments, including: if there are records with the same field reported repeatedly under the same timestamp, deduplication is performed and a duplicate anomaly is marked; if there are missing key fields or data types that cannot be parsed, a missing field anomaly is marked; if there are abnormal values ​​due to out-of-bounds values ​​or inconsistent units, an out-of-bounds anomaly is marked. Based on the operational status data segments, the cumulative downtime and cumulative micro-stop time of the production unit within the cycle time window are statistically analyzed, and the consistency of start and end times is checked in conjunction with the event record data segments. If time inconsistency is found or the sum of the cumulative downtime and cumulative micro-stop time exceeds the sampling cycle length, the time is marked as abnormal. The energy consumption increment of the production unit within the cycle time window is calculated based on the energy consumption metering data segment. If the calculation result is negative or there is an obvious out-of-bounds reading, an out-of-bounds anomaly is marked. Based on the production and quality count data segments, the total production increment, qualified production increment, and first-pass qualified quantity increment of the production unit within the cycle time window are statistically analyzed. If the logical relationship that the first-pass qualified quantity does not exceed the qualified production and the qualified production does not exceed the total production is not satisfied, then a logical anomaly is marked. The statistically obtained cumulative downtime, cumulative micro-downtime, energy consumption increment, total output increment, qualified output increment, and first-pass qualified output increment, together with the production unit identifier, cycle index, cycle time window, sampling cycle length, and marked data quality markers, are assembled into a cycle sampling record.

[0008] Furthermore, the method for constructing new decision unit blocks based on the periodic sampling records includes: Obtain available period sampling records for each production unit in each sampling period. The available period sampling records are period sampling records with data quality marked as empty set or containing only duplicate anomaly marks. For each available cycle sampling record, extract the production unit identifier, cycle index, sampling cycle length, cumulative downtime, cumulative micro-downtime, energy consumption increment, total output increment, qualified output increment, and first-pass qualified output increment. The qualified output increment, the first-pass qualified output increment, and the total output increment are assembled into a production vector in a preset order; The cumulative downtime, cumulative micro-downtime, and energy consumption increment are used as the first three input indicators, and the speed loss indicator is calculated as the fourth input indicator. The speed loss indicator is calculated as follows: the theoretical output is obtained by multiplying the sampling period length by the target production rate effective within the sampling period of the production unit. The difference between the theoretical output and the actual total output increment is taken. If the difference is positive, the difference is taken; otherwise, it is taken as zero. The cumulative downtime, cumulative micro-downtime, energy consumption increment, and speed loss indicator are assembled into an input vector in a preset order. The assembled input vector and output vector are used as decision units corresponding to the available periodic sampling records, and the index of the decision unit is added to the set of new decision unit blocks to construct new decision unit blocks.

[0009] Furthermore, methods for determining the historically affected set based on the default values ​​between historical decision-making units and each decision-making unit in the newly added decision-making unit block include: For each historical decision unit and each decision unit in the newly added decision unit block, calculate the default value:

[0010] in, This is the default value; An index for historical decision-making units; For the newly added decision unit block, the decision unit index is provided. For historical decision-making units In the previous sampling period The output multiplier vector; For historical decision-making units In the previous sampling period The input multiplier vector; For historical decision-making units In the previous sampling period The free variables; For the newly added decision-making unit The input vector; For the newly added decision-making unit The output vector; Represents the transpose of a vector; For each historical decision unit, calculate its maximum default value on the newly added decision unit block:

[0011] in, The maximum default value; For the newly added decision-making unit block set; The historical decision-making units with the maximum default value greater than zero are included in the historical affected set.

[0012] Furthermore, the methods for identifying newly added effective decision units and newly added highly effective decision units from the newly added decision unit block that can be directly confirmed as effective or highly effective include: For each decision unit in the newly added decision unit block, calculate the maximum default value of the newly added decision unit on the set of historical decision units, and obtain the historical decision units that have reached the maximum default value as reference objects; If both of the following conditions are met simultaneously, the newly added decision unit and other newly added decision units associated with the same reference object and having the same maximum default value are jointly identified as newly added valid decision units: First, the maximum default value of the newly added decision unit on the historical decision unit set is greater than zero; Second, the maximum default value is equal to the maximum default value of the historical decision unit used as the reference object on the newly added decision unit block. For each decision unit that is identified as a new valid decision unit, its multiplier solution is constructed to be the same as that of the historical decision unit that serves as the reference, and the free variables are updated to be the sum of the free variables of the historical decision unit and its maximum default value on the new decision unit block. New highly effective decision-making units are selected from the set of newly identified effective decision-making units. The selection rules include: if a newly identified effective decision-making unit is the only element in its associated set of decision-making units, or has the smallest value in its associated set of decision-making units for a certain input indicator, or has the largest value in its associated set of decision-making units for a certain output indicator, or has the best input-output ratio in its associated set of decision-making units, then the newly identified effective decision-making unit is identified as a newly identified highly effective decision-making unit.

[0013] Furthermore, the method for constructing a necessary reference set mapping for the decision-making units in the set to be evaluated includes: For each decision unit in the historical affected set, obtain the set of new decision units corresponding to the point when it reaches the maximum default value in the new decision unit block: And map the newly added set of decision-making units as the mandatory reference set of decision-making units in the historically affected set; For each decision unit in the set of decision units that are not directly identified as valid or highly valid in the newly added decision unit block, its mandatory reference set mapping is set to an empty set.

[0014] Furthermore, the restricted-size subsample reference set is constructed as follows: For any target decision unit in the set to be evaluated, obtain the required reference set of the target decision unit and use it as the initial member of the subsample reference set; When the number of members in the subsample reference set does not reach the preset upper bound, target decision units are selected from the historical valid set in order of increasing difference from the target decision unit in terms of input-output characteristics, until the number of members in the subsample reference set reaches the upper bound minus one. The target decision-making unit itself is included in the subsample reference set to obtain a complete restricted-size subsample reference set.

[0015] Furthermore, methods for obtaining the efficiency values ​​and multiplier solutions for each decision-making unit include: For any target decision unit in the set to be evaluated, solve the multiplier model on the constructed restricted-size subsample reference set to obtain the candidate efficiency value and candidate multiplier solution of the target decision unit; The candidate multiplier solution is used to perform constraint checks on all currently available decision units, and the default indication of each decision unit for the candidate multiplier solution is calculated. If the default indication does not exceed zero, then the candidate efficiency value and candidate multiplier solution are confirmed as the final efficiency value and multiplier solution of the target decision unit, and the solution is then entered into the next target decision unit. If the default indication is greater than zero, the set of decision units that cause the default is identified. Decision units that meet the extremely efficient condition are selected from the default decision unit set and added to the extremely efficient set. The default decision unit set is then extended to the subsample reference set to form an updated subsample reference set. The multiplier model is re-solved on the updated subsample reference set until the default indication does not exceed zero, thus obtaining the final efficiency value and multiplier solution of the target decision unit.

[0016] Furthermore, methods for generating runtime status detection results include: Obtain the most efficient set of the current sampling period and the previous sampling period, and generate an early warning of changes in the efficiency frontier state by comparing the changes in the most efficient set of the two periods; Obtain the historical affected set and the efficiency change of each decision unit in the set between adjacent periods. If the proportion of the number of decision units with decreased efficiency in the historical affected set to the total number of the historical affected set reaches a preset collective decrease proportion threshold, a historical collective efficiency decrease warning is generated. The number of times each decision unit in the historical affected set and the newly added decision unit block is referenced in the mandatory reference set mapping is obtained. If the number of references of a newly added decision unit reaches the preset single-point impact coverage intensity threshold and the proportion of the total number of the historical affected set reaches the collective decline ratio threshold, then the newly added decision unit is identified as a strong impact source point, and a strong impact source location warning is generated. For each production unit, obtain its efficiency sequence within the preset analysis window length, and count the proportion of decision units with efficiency values ​​lower than the preset inefficiency threshold to the total number of decision units within the window. If this proportion reaches the preset inefficiency ratio threshold, generate a long-term inefficiency solidification warning for that production unit, and set the warning level according to the severity of the inefficiency ratio. For each production unit, the efficiency jump amplitude between adjacent sampling periods within the preset analysis window length is obtained. If there are multiple consecutive jump amplitudes exceeding the preset adjacent efficiency jump threshold, an abnormal efficiency fluctuation warning for that production unit is generated. The generated warnings are summarized in layers according to system level, block level, and unit level. Unit-level warnings of the same production unit are merged according to preset priority rules, and the final operation status detection results are output.

[0017] To achieve the above objectives, a second aspect of the present invention provides an online efficiency operation status detection system based on industrial flow data, comprising: The data acquisition and access module is used to periodically summarize the raw data streams of multiple homogeneous production units and generate periodic sampling records for each production unit in each sampling period. The decision unit construction module is connected to the data acquisition and access module and is used to construct new decision unit blocks based on the periodic sampling records. Each decision unit is represented by an input vector and an output vector. The efficiency update object identification module, connected to the decision unit construction module, is used to obtain the historical decision unit set up to the previous sampling period, along with its efficiency value and multiplier solution. Based on the default values ​​between the historical decision units and each decision unit in the newly added decision unit block, it determines the historical affected set and identifies decision units in the newly added decision unit block that can be directly confirmed as effective or highly effective, thus obtaining newly effective decision units and newly highly effective decision units, and updating the current effective set and highly effective set. It is also used to construct an evaluation set based on the historical affected set and the decision units in the newly added decision unit block that are not directly confirmed as effective or highly effective, and to construct a mandatory reference set mapping for the decision units in the evaluation set. An efficiency acceleration solution module, connected to the efficiency update object identification module, is used to solve the efficiency of decision units in the set to be evaluated on a limited-size subsample reference set with the mandatory reference set mapping as a constraint, and to obtain the efficiency value and multiplier solution of each decision unit. The operation status change early warning module is connected to the efficiency acceleration solution module and is used to generate operation status detection results based on the efficiency value and multiplier solution, the newly added highly efficient decision unit, and the highly efficient set and efficiency value of the previous sampling period.

[0018] The beneficial effects of this invention are: Compared with existing technologies, this invention provides an online efficiency operation status detection method and system based on industrial flow data. By constructing snapshots of the operation status of each production unit in each sampling period into a unified decision-making unit, it provides a quantitative basis for horizontal efficiency evaluation under conditions of multiple inputs and multiple outputs. By calculating the default value between historical decision-making units and newly added decision-making units, it identifies the set of historical decision-making units affected by the newly added units based on the default value, and simultaneously identifies decision-making units among the newly added units that can be directly confirmed as effective or highly effective. This avoids the need to recalculate all historical decision-making units in each period, realizes an incremental local update mechanism, and solves the problem of unbearable computational load for traditional data envelopment analysis methods in industrial flow data scenarios. The technical problem is addressed by constructing a mandatory reference set mapping for the decision-making units in the set to be evaluated and performing efficiency solutions on a limited-size subsample reference set. This controls the reference size of each linear programming solution within a preset range, further ensuring computational efficiency and stability in high-frequency data stream environments. Finally, by generating operational status detection results based on efficiency values ​​and multiplier solutions, newly added highly efficient decision-making units, and historical highly efficient sets and efficiency values, the evolution of the efficiency frontier structure and changes in the efficiency value sequence are automatically transformed into interpretable early warning information. This solves the technical problem that existing industrial monitoring systems can only provide in-depth efficiency evaluation and decision support based on single-index threshold alarms, achieving a technological leap from "whether monitoring meets standards" to "understanding why efficiency is achieved." Attached Figure Description

[0019] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below.

[0020] Figure 1 This is a flowchart of an online efficiency operation status detection method based on industrial flow data disclosed in an embodiment of the present invention. Detailed Implementation

[0021] To enable those skilled in the art to better understand the present invention, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of the present invention.

[0022] According to embodiments of the present invention, it should be noted that the steps shown in the flowcharts of the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions, and although a logical order is shown in the following methods, in some cases the steps shown or described may be executed in a different order than that shown here.

[0023] like Figure 1 As shown, this invention provides an online efficiency operation status detection method based on industrial flow data, comprising the following steps: Step S100: Periodically summarize the raw data streams of multiple homogeneous production units to generate periodic sampling records for each production unit in each sampling period. In this step, it is necessary to uniformly model multiple homogeneous production units deployed in the industrial production site. The production unit can be a factory, work group, production line, or equipment unit. Homogeneity means that the technological processes and functional attributes of each production unit are consistent, and input and output indicators are generated under a unified indicator dictionary and consistent statistical caliber, so that they are horizontally comparable.

[0024] With uniform sampling period length The operational data of each production unit is periodically summarized. The length of this sampling period can be preset according to actual business needs, but it is not required that different production units sample synchronously at the same physical moment. Data from each production unit is allowed to arrive asynchronously on the timeline.

[0025] First, based on the current cycle index With the preset sampling period length Determine the start and end times of the current sampling period to form a period time window. ,in, For the first The start time of each sampling period For the first The end time of each sampling period satisfies For each production unit Extracting various data segments from the original data stream set whose timestamps fall within the time window of that period, specifically including: Running status data fragments It is used to reflect the switching of the operating status of the production unit on the time axis, and can at least distinguish between the states of effective operation, shutdown, and minor shutdown. Event log data fragments This includes the start and end times, type identifiers, etc. of shutdown and micro-shutdown events; Energy consumption metering data fragments This includes energy consumption metering or integral readings within a cycle; And production and mass count data fragments It includes total output, qualified output, and first-pass yield, or other counts or detailed data that can be statistically analyzed within the cycle.

[0026] After obtaining the above data fragments, the fields of various data are standardized, and the field names, data types and units of measurement are uniformly mapped to the preset standard format to ensure the consistency and computability of the data.

[0027] Based on this, basic data quality marking is performed: if the same field is found to be repeatedly reported or recorded at the same timestamp in any period segment, deduplication is performed according to the preset unique key, and only one record is retained for the current period's statistics. At the same time, the DUPLICATE mark is added to the data quality mark set and the duplicate field name, source summary, and duplicate count summary are recorded; if a key field is missing or its type cannot be parsed, the MISSING mark is added to the data quality mark set and the list of missing fields and the source summary of the missing fields are recorded; if abnormal values ​​are found due to non-numerical, significant out-of-bounds, or inconsistent units, the RANGE mark is added to the data quality mark set and the field name and value range summary are recorded.

[0028] Next, the cumulative downtime will be calculated based on the runtime status data fragments. Cumulative time with micro-stop The system performs consistency checks on the start and end times and status segments of shutdown and micro-shutdown events based on event log data fragments: if time inconsistencies are found, the TIME flag is added to the data quality flag set and an inconsistency summary is recorded; simultaneously, time validity checks are performed, and if the cumulative shutdown time or micro-shutdown time is negative, the RANGE flag is added to the data quality flag set; if the sum of the cumulative shutdown time and the cumulative micro-shutdown time exceeds the sampling period length... If so, the TIME flag will be added to the data quality flag set and the time difference exceeding the limit will be recorded.

[0029] Calculate the periodic energy consumption increment based on energy consumption metering data fragments. If the calculation result is negative or there is a significant out-of-range reading, the RANGE flag is added to the data quality flag set, and the energy consumption value and range summary are recorded. The total output increment is calculated based on the output and quality count data segments. Qualified production increase And the increase in the number of qualified items passed at one time And perform a logical consistency check; if it does not meet the requirements... If the logical relationship is correct, the LOGIC tag will be added to the data quality tag set and the field pairs that violate the constraints and the difference summary will be recorded.

[0030] Finally, identify the production unit. Periodic Index Periodic Time Window Sampling period length Total downtime Accumulated time for micro-stops Energy consumption increment Total output increase Qualified production increase Increase in the quantity of qualified products passed in one go and the generated set of data quality tags (Include Marked sets and Evidence summary mappings are jointly assembled into a structured periodic sampling record. This record contains both core business metrics for subsequent efficiency evaluation and complete data quality information.

[0031] Step S200: Construct a new decision unit block based on the periodic sampling records. Each decision unit in the new decision unit block is represented by an input vector and an output vector. In this step, the available periodic sampling record set for each production unit in each sampling period is obtained. The definition of the available periodic sampling record is: for the periodic sampling record generated in step S100 If its data quality label It is an empty set, or Includes only duplicate exception markers Then the record is considered available and included. For records that do not meet the above conditions, they will only be archived and will not participate in the decision-making unit construction and subsequent efficiency update chain of this step.

[0032] for Each available periodic sampling record in Read the production unit identifier from it. Periodic Index Sampling period length Total downtime Accumulated time for micro-stops Energy consumption increment Total output increase Qualified production increase Increase in the quantity of qualified products passed in one go Fields such as [field name], and record the index of the decision unit as [index name]. .

[0033] The output vector is assembled based on a pre-defined output indicator system. Specifically, the output vector consists of three output indicators, assembled in the order of qualified output increment, first-pass qualified output increment, and total output increment:

[0034] in, Let be the output vector of decision-making unit i,t. For production units During the sampling period The increase in qualified production within the region. Indicates period The number of products that meet quality requirements without rework or repair. Indicates period The total number of products produced by the internal production unit (regardless of whether they are qualified); This represents the transpose of a vector.

[0035] An input vector is assembled based on a pre-defined input indicator system. The input vector consists of four input indicators, the first three of which are derived from the cumulative downtime, cumulative micro-downtime, and energy consumption increment, respectively:

[0036] in, Downtime loss time (first input indicator); For micro-stop time (the second input indicator); This is the increase in energy consumption (the third input indicator).

[0037] The fourth input indicator is speed loss. Used to characterize the cycle Potential production loss due to internal factors causing the operating speed to fall below the target cycle time. The system acquires production units. During the sampling period Target production rate effective within the period Based on the sampling period length Multiply by the target production rate to obtain the theoretical output. ,in, For production units During the sampling period The target production rate is then determined. The theoretical output is then calculated as the increment between the theoretical and actual total output. The difference is taken as the integer part; if the difference is positive, the integer part is taken as zero.

[0038] in, This is an indicator of speed loss.

[0039] The above four input indicators are assembled into an input vector in the order of downtime loss, micro-downtime, energy consumption increment, and speed loss:

[0040] in, Decision-making unit The input vector.

[0041] Ensure that all input and output indicators satisfy the non-negativity boundary conditions, i.e. Furthermore, in order to satisfy the solvability and numerical stability of the input-oriented DEA model, the decision-making unit entering the efficiency update calculation should also simultaneously satisfy the conditions that the total input is greater than zero and the total output is greater than zero.

[0042] For the assembled decision-making unit Index it Add new decision unit block index set and output the corresponding data payload. This newly added decision-making unit block set The input and output data will serve as standardized inputs for the subsequent efficiency update object identification and efficiency solution modules, driving the incremental efficiency evaluation process.

[0043] Step S300: Obtain the historical decision unit set up to the previous sampling period, its efficiency value and multiplier solution. Based on the default value between the historical decision units and each decision unit in the newly added decision unit block, determine the historical affected set and identify the decision units in the newly added decision unit block that can be directly identified as effective or extremely effective. Obtain the newly added effective decision units and the newly added extremely effective decision units, and update the current effective set and extremely effective set. In this step, the previous sampling period is obtained. Historical cumulative decision-making unit set Each decision unit uses a binary index. express, For production unit identification, This represents the sampling time corresponding to the decision-making unit. Simultaneously, historical efficiency and multiplier solutions are obtained: for each... Given its efficiency value and the solution of multipliers ,in To produce multiplier vectors, To input the multiplier vector, These are free variables.

[0044] In addition, the historical valid set is also obtained. A highly effective combination with history and the set of newly added decision unit blocks in the current sampling period. The intra-block decision unit index is denoted as And give each new decision unit Input vector With output vector .

[0045] First, the calculation is performed based on the default values ​​between each decision unit in the historical decision unit and the newly added decision unit block: for any historical decision unit... With newly added decision-making units Define the default value:

[0046] in, This is the default value; An index for historical decision-making units; For the newly added decision unit block, the decision unit index is provided. For historical decision-making units In the previous sampling period The output multiplier vector; For historical decision-making units In the previous sampling period The input multiplier vector; For historical decision-making units In the previous sampling period The free variables; For the newly added decision-making unit The input vector; For the newly added decision-making unit The output vector; Represents the transpose of a vector; This default value measures historical decision-making units. The multiplier solution in evaluating the newly added decision-making unit Does it violate the constraints? If This indicates that the historical multiplier cannot "accommodate" the newly added decision-making unit. Based on this, for each historical decision-making unit... Calculate its maximum default value on the newly added decision unit block:

[0047] in, The maximum default value; For the new decision-making unit block set Then, based on calculations, all historical decision-making units with a maximum default value greater than zero are included in the historical affected set:

[0048] in, The set of historically affected units refers to the set of historical decision-making units whose efficiency may change due to the emergence of new decision-making units.

[0049] For each Further record the set of newly added decision units that reach their maximum default value:

[0050] in, For historical decision-making units The set of new decision units corresponding to the maximum default value reached on the new block.

[0051] It identifies newly added decision units that can be directly confirmed as valid or highly valid, thus obtaining newly added valid and highly valid decision units, and updating the current valid and highly valid sets. Specifically, a temporary set is initialized. This is used to store newly added decision units that have not yet been processed. Each decision unit in Calculate its maximum default value relative to the set of historical decision-making units. And record the historical decision-making units that reached that maximum default value. If both conditions are met... and Then determine the decision-making unit To add valid entries, and to include all those that meet the requirements. of Form a set This set All decision-making units are added to the new valid set. And set an efficiency value for it. Multiplication of the child ( , , Constructed as follows: For each ,make .

[0052] Then, highly efficient representatives are selected from the newly added valid set. If Then, the unique decision-making unit is added to the newly added highly efficient set. ;like Select the most effective representative set according to preset rules. The selection rule is a union filter: there is an input dimension. Make There may be output dimensions. Make or calculate each input-output ratio And select the one with the largest and unique ratio. Add the selected most efficient representatives. and will All elements from Remove from the list. If the above conditions are not met, then remove the current list. from Removed from the set but not added to any valid set, to be processed in subsequent steps.

[0053] Finally, based on the above identification results, the current valid set and the highly efficient set are updated, and the updated valid set is output. The updated highly efficient set and the set to be evaluated For each decision unit to be evaluated Constructing a mapping that requires referencing a set: if ,but ;like ,but Simultaneously, output the results for each newly added valid decision-making unit. efficiency value and its multiplier solution and historically affected sets This is for use in subsequent steps.

[0054] Step S400: Based on the historical affected set and the decision units in the newly added decision unit block that have not been directly identified as effective or highly effective, construct a set to be evaluated, and construct a mandatory reference set mapping for the decision units in the set to be evaluated; In this step, based on the historically affected set already identified in step S300... The set of remaining decision units that have not yet been determined to be valid in the new block is further constructed into the set to be evaluated that needs to be entered into the efficiency solution in this cycle, and a mandatory reference set mapping is assigned to each decision unit in it to guide the constrained scale solution in the subsequent step S500.

[0055] First, define the set of decision units in the newly added decision unit block that have not been directly identified as valid or highly valid. Let the set of newly identified valid decision units in step S300 be denoted as... Then the set of remaining decision units in the newly added block that have not yet been determined as valid or highly valid is:

[0056] in, This refers to the set of decision units in the newly added block that have not been directly identified as valid or highly valid.

[0057] The decision-making units in this set were neither directly confirmed as valid or highly efficient through the default value condition in step S300, nor were they included in the newly added valid set (the newly added highly efficient set is already included in the newly added valid set). Therefore, they need to enter the subsequent efficiency solution process to determine their efficiency value.

[0058] Next, the set of items to be evaluated for this period is constructed. The set to be evaluated consists of two parts: one is the set of historically affected individuals. The first category consists of historical decision-making units whose efficiency may change due to the emergence of new decision-making units; the second category consists of the set of decision-making units in the newly added block that have not been directly identified as effective. Therefore, the set to be evaluated Defined as:

[0059] All decision units in this set need to have their efficiency values ​​and multiplier solutions recalculated or calculated for the first time in step S500.

[0060] Then, for each decision unit in the set to be evaluated Construct a map of a set of required references The meaning of the mandatory reference set mapping is: in the subsequent efficiency solution process, the solution decision unit is... The subsample reference set used for efficiency must be required to include The decision-making unit within the system. The construction rules fall into two categories: (1) For decision units in the newly added block that are not directly identified as valid or highly valid, i.e. Each one has already been recorded in the previous step S300. The set of newly added decision units that reaches the maximum default value on the newly added decision unit block. This set reflects the relationship with historical decision-making units. The most tightly constrained new decision-making unit is the cause of... The key reference point where efficiency may decline. Therefore, [the following will be taken as an example]. As Required reference set : ,

[0061] Notice It may contain multiple new decision-making units, or it may be an empty set (but according to...) Definition of maximum default value ,and It is achieved by one or more newly added decision-making units, and is therefore usually not empty.

[0062] (2) For decision-making units in the newly added block that are not directly identified as valid, i.e. Since these decision-making units have not yet established strong constraints with historical decision-making units, and no directly inheritable multiplier solutions have been identified, their mandatory reference set is mapped to an empty set:

[0063] in, For unconfirmed decision units in the newly added block The required reference set is set to an empty set.

[0064] This means that when solving the efficiency of these decision units in the future, it is not necessary to force the inclusion of any specific reference units, but to determine it entirely by the constrained-size subsample reference set construction mechanism in step S500.

[0065] Through the above construction, not only is the aggregation of the set to be evaluated completed, but also differentiated prior constraint information is assigned to each decision unit to be evaluated: the mandatory reference set for historically affected decision units points to the newly added decision units that caused their default, ensuring that these key reference objects are not overlooked when re-evaluating efficiency; while the decision units in the new blocks that are not confirmed (i.e., not confirmed as effective or highly effective) have no mandatory constraints, and their reference sets will rely entirely on the distance metric and highly effective set supplementation strategy in subsequent steps. Finally, the set to be evaluated... Required reference set mapping And the efficiency values ​​of the newly added effective decision-making units obtained in step S300. and its multiplier solution The updated valid set With the most efficient set Historically affected collections Output together for use in steps S500 and S600. Among them, and This will serve as the initial efficient set and the highly efficient set for subsequent efficiency-solving modules, and It will be passed to the operational status change early warning module to generate collective descent and impact positioning early warnings.

[0066] Step S500: Using the required reference set mapping as a constraint, perform efficiency calculations on the decision units in the set to be evaluated on the limited-size subsample reference set to obtain the efficiency value and multiplier solution of each decision unit. In this step, the set to be evaluated is based on the output of step S400. (For the sake of simplicity, superscripts are omitted here) Actual correspondence ), Initially efficient set (corresponding to the output of step S400) Initial valid set (corresponding to the output of step S400) Candidate reference set Required reference set mapping and the preset upper bound parameter of the scale For each decision unit to be evaluated, solve the DEA multiplier model under the assumptions of input orientation and variable returns to scale, and output the efficiency value. Decision-making unit to be evaluated Multiplier solution and update the most efficient set. With valid sets .

[0067] First, perform initialization: set If the initial highly efficient set If not empty, proceed directly to iterative solution; if Then in the set to be evaluated Construct the initial highly efficient set according to the following rules. (Simply satisfy one of the following conditions): Rule (a) Minimum unique input: If an input dimension exists. , making a certain satisfy And this minimum value is unique, then ; Rule (b) Output unique maximum: if an output dimension exists. , making a certain satisfy And this maximum value is unique, then ; Rule (c) Ratio Supplement: For each calculate ,like If it is the largest and unique, then .

[0068] make And record the current maximum effective set size. Initialize the subset to be evaluated that has not yet yielded final efficiency results. .

[0069] Next, the iterative solution process begins. The following steps are executed repeatedly.

[0070] from The first target decision unit is retrieved in index order. First, construct the target decision unit. Limited-size subsample reference set : set (i.e., the set that must be referenced); if the current Then only the distance is retained. Recent There are 1 decision-making unit, and the distance is defined as:

[0071]

[0072] in, Decision-making unit and The distance between them is used for reference set filtering; then the remaining capacity is calculated. ,like Then all Merging Otherwise from According to distance Select from smallest to largest order Each is incorporated; finally, the target decision-making unit itself is added to the reference set: ; This represents the size of the current most efficient set.

[0073] In the constructed subsample reference set Above, solve the following multiplier model (input-oriented, variable-scale returns):

[0074]

[0075]

[0076]

[0077] Obtain the optimal solution And the corresponding candidate efficiency values ​​obtained on the subsample reference set. (in fact) And due to constraints This model is equivalent to the standard input-oriented DEA model. The KKT indicator is then calculated. :

[0078] according to The value is used for branching.

[0079] like If so, then the current candidate solution is the global optimal solution. (Record) and and will from Remove from the middle. If Then perform a highly efficient update: define a tight constraint set. .like Then Join the most efficient set ;like Then in Construct a highly efficient set according to the rules described in step S300 (minimum input, maximum output, or optimal input-output ratio). and add it At the same time from Remove from (if) There are elements that belong to ), and for each set up , multiplier solution Then return to the starting point of the iteration and continue processing. The remaining decision-making units.

[0080] like If the current subsample reference set is insufficient to constrain the global set, then construct the default set. And let the set of decision units corresponding to the maximum default value be... .exist Selecting the most effective candidate set :if only An element that satisfies any one of the aforementioned rules (minimum input, maximum output, or optimal input-output ratio) can be placed into the database. Then Join the most efficient set For any ,like Then take it from Remove from and set , multiplier solution (Note that free variables need to be added) Next, update the subsample reference set: Let ,like Then from According to distance Eliminate redundant units from farthest to nearest, until... Finally, the target decision-making unit was reinstated. Then return to the multiplier model solution steps and resolve the decision unit. .

[0081] when The iterative solution ends when the set becomes empty. If the initial set is highly efficient... Empty (i.e., the initial highly efficient set was constructed using the above rules) Then it is necessary to... Each decision unit in Further solution: In the most efficient set (This already includes) Solving the multiplier model on (and other newly added highly effective units) yields... and This is to ensure that the multiplier solutions of these initial highly effective units have the correct reference frame.

[0082] Finally, output all efficiency value Solution with multipliers Output the updated highly efficient set. With valid sets These results will serve as input to step S600 to generate the running status detection results, and will also serve as the historical efficiency and multiplier solution for the next sampling period, driving the subsequent incremental update process.

[0083] Step S600: Generate the operating status detection result based on the efficiency value and multiplier solution, the newly added highly efficient decision unit, and the highly efficient set of the previous sampling period.

[0084] In this step, based on the current sampling period Compared with the previous sampling period cumulative decision-making unit set and Add a set of running statuses Historically affected collections (Potentially empty), Required reference set mapping The most efficient set of the current period and the previous period and And the efficiency values ​​of all decision-making units over two cycles. and And based on a preset set of configurable parameters Output a structured set of early warnings .in, The threshold for the collective decrease ratio (configurable parameter); Single-point impact coverage intensity threshold (configurable parameter); Length of the production unit analysis window (configurable parameter); Inefficient threshold (configurable parameter); Inefficient percentage threshold (configurable parameter); Threshold for the proportion of persistently inefficient operations (configurable parameter); The trend significance threshold (configurable parameter); The threshold for adjacent efficiency jumps (configurable parameter); Minimum number of jumps required to confirm abnormal fluctuations (configurable parameter); Length of the Top-K evidence list (configurable parameter).

[0085] First, preprocessing is performed: for any condition that simultaneously exists... and Decision-making unit Calculate the change in efficiency Define the historical actual decline set. ,like Then calculate the minimum value of the decrease. , median and maximum value Otherwise, set it to NA. Simultaneously initialize the warning set. and set the basic evidence fields. .

[0086] Early warning of changes in the efficiency frontier (Type I warning): Compare the changes in the highly efficient set over two periods: Calculate the newly added highly efficient set. and exit the most efficient set If both are empty, the leading edge remains unchanged, and an info-level warning is generated; otherwise... If it is, then it is determined to be a "frontier extension"; otherwise, if and If the condition is met, it is determined as "frontier surface reconstruction, local boundary optimization"; otherwise, it is determined as "significant frontier surface contraction and reconstruction, substantial improvement of the technical boundary". Regardless of the type of change (except for invariance), a warning-level alert is generated, with the alert type being FrontierStatus, the scope being system, and the relevant set being... Key metrics include The evidence field contains .

[0087] Historical efficiency collective decline warning (Category II warning): If If so, skip this type of warning. Otherwise, calculate the decrease percentage. .like This will trigger a CollectiveDrop warning, with a scope of system and the relevant collection being... Key indicators are Severity: If If so, it is critical; otherwise, it is warning. The evidence field contains... List (from) Chinese Press Before selecting descending order (Details of each decision-making unit and its efficiency changes). If If the output is not specified, then no output will be provided.

[0088] Strong impact source location early warning (Category III early warning): If If not, skip it. For each new decision unit... Count the number of times it is cited. Calculate the maximum number of citations. and determine the set of impact sources. Secondary discrimination: When and If at the same time, and This will trigger a DominantShock warning. The warning is defined when triggered. and each corresponding The warning scope is block, and the relevant set is... Key indicators are Severity: If If it is critical, then it is critical; otherwise, it is warning. The evidence field contains... .

[0089] Long-term inefficient solidification early warning (Category 4 early warning): For each production unit Define the most recent A set of windows for each period If the number of decision units within the window is insufficient If so, skip that unit. Define an inefficient indicator function. Calculate the proportion of inefficiency .like If this occurs, a PersistentInefficiency alert is triggered. Further severity grading is then performed: the slope of a simple linear regression of the efficiency sequence within the window is calculated. (If the number of cycles is sufficient), if and If the severity is critical, then the severity is critical; otherwise, if and If there is no significant negative trend, it is a warning; if only the proportion of inefficiency meets the standard but the trend is positive or stable, it is still a warning. The warning scope is unit, and the relevant set is... Key metrics include The evidence field contains List of inefficient decision-making units and their efficiency values.

[0090] Efficiency fluctuation warning (Category 5 warning): For each production unit Similarly, take the window. ,Require Calculate the efficiency jump amplitude between adjacent sampling periods. (in Define the jump indicator function. If there exists a continuous times or more That is, there exists continuity If the number of consecutive transitions exceeds the threshold, a VolatilityAnomaly warning is triggered. Let the set of consecutive transition locations be denoted as . Calculate the maximum jump amplitude `maxjump` and the median jump amplitude `median(jump)`. The warning scope is `unit`, and the relevant set is... ,in To trigger the collection. Key metrics include Severity: If the condition is met (consecutive transitions) and any single transition amplitude exceeds [a certain threshold], then [the severity is determined by the severity of the transition]. or consecutive times If it is true, then it is critical; otherwise, it is a warning. The evidence field contains... .

[0091] Summary and Deduplication Merging (Sixth Type of Warning): The generated warnings are divided into hierarchical categories: System and block-level alerts are retained and not merged. Unit-level alerts are grouped by production unit: for each production unit... ,like Then do not merge; if If severity is equal, the primary warning is selected based on priority: critical > warning > info. If severity is equal, the primary warning is selected based on type: VolatilityAnomaly > PersistentInefficiency. The explanatory descriptions of other merged warnings are added as additional tags to the primary warning's interpretation field, and a merged list is appended to the primary warning's evidence_fields, recording the summaries (warning_type, severity, key_metrics) of the merged warnings and their key evidence indexes. Final output. ,in For each unit, there is a primary warning (if merged) or a primary warning (if not merged). This set of warnings is the operational status detection result generated in step S600, used for operational monitoring and decision support, helping managers quickly locate changes in efficiency patterns, sources of impact, long-term inefficient units, and abnormal fluctuations, achieving in-depth support from "whether monitoring meets standards" to "understanding why efficiency is high".

[0092] The method of this invention is executed according to the following process under continuous sampling period to form an end-to-end online efficiency update and operation monitoring closed loop.

[0093] First, the system maintains cross-period state variables, including: the previous sampling period. cumulative decision-making unit set Efficiency value Multiplication solution Valid set With the most efficient set When the system starts up for the first time, set... Set cold start indicator .

[0094] Each sampling period At the beginning, step S100 is executed first to generate periodic sampling records. Then, the available sample records are filtered based on data quality markers. (Records whose data quality is marked as an empty set or contain only the DUPLICATE mark), execute step S200 to construct the new decision unit block. Update the cumulative set .

[0095] Then, the three-stage gating judgment begins: (1) If cold start indicator and (Cold start accumulation phase), then efficiency calculation and early warning determination will not be performed in this cycle, and the process will directly jump to the cross-cycle write-back step, in which, This is the threshold for the cumulative number of DMUs during the cold start phase.

[0096] (2) If and Then, we enter the initial efficiency benchmark establishment stage: setting the set to be evaluated. If the initial valid set and the highly efficient set are empty, call step S500 (efficiency acceleration solution module) to process all... Solving for the efficiency value and the multiplier solution yields the following results. and , place Then, it will jump to the operation status change warning step.

[0097] (3) If Then it enters the online incremental efficiency update chain: If a new block is added Then the efficiency and multiplier solutions of all historical decision-making units will follow the results of the previous cycle. This will redirect you directly to the warning procedure.

[0098] Otherwise, proceed with steps S300 and S400 to obtain the set to be evaluated. Required reference set mapping, updated valid set With the most efficient set Historically affected collections And the efficiency and multiplier solutions of newly added effective decision-making units.

[0099] like If the result is satisfactory, then the updated set described above should be used directly; otherwise, step S500 should be called. The efficiency of the decision-making unit in the algorithm is calculated to obtain its efficiency value and multiplier solution, and then updated. and .

[0100] For historical decision units that have not entered the set to be evaluated, their efficiency and multiplier solutions use the results from the previous period.

[0101] Finally, step S600 generates the running status detection result (this step uses the highly efficient set and efficiency value from the previous sampling period as input), and performs a cross-period write-back: , , , , Write back to each , , , , Then let Then, proceed to the next cycle.

[0102] According to another aspect of the embodiments of this application, an online efficiency operation status detection system based on industrial flow data is also provided, comprising: The data acquisition and access module is used to periodically summarize the raw data streams of multiple homogeneous production units and generate periodic sampling records for each production unit in each sampling period. The decision unit construction module is connected to the data acquisition and access module and is used to construct new decision unit blocks based on the periodic sampling records. Each decision unit is represented by an input vector and an output vector. The efficiency update object identification module, connected to the decision unit construction module, is used to obtain the historical decision unit set up to the previous sampling period, along with its efficiency value and multiplier solution. Based on the default values ​​between each decision unit in the newly added decision unit block and the historical decision units, it determines the historical affected set and identifies decision units in the newly added decision unit block that can be directly confirmed as effective or highly effective, thus obtaining newly effective decision units and newly highly effective decision units. It is also used to construct an evaluation set based on the historical affected set, the decision units in the newly added decision unit block that were not directly confirmed, and the newly effective and newly highly effective decision units, and to construct a mandatory reference set mapping for the decision units in the evaluation set. An efficiency acceleration solution module, connected to the efficiency update object identification module, is used to solve the efficiency of decision units in the set to be evaluated on a limited-size subsample reference set with the mandatory reference set mapping as a constraint, and to obtain the efficiency value and multiplier solution of each decision unit. The operation status change early warning module is connected to the efficiency acceleration solution module and is used to generate operation status detection results based on the efficiency value and multiplier solution, the newly added highly efficient decision unit, and the highly efficient set of the previous sampling period.

[0103] In the above embodiments of the present invention, the descriptions of each embodiment have different focuses. For parts not described in detail in a certain embodiment, please refer to the relevant descriptions of other embodiments.

[0104] In the several embodiments provided in this application, it should be understood that the disclosed technical content can be implemented in other ways. The device embodiments described above are merely illustrative; for example, the division of units can be a logical functional division, and in actual implementation, there may be other division methods. For instance, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the displayed or discussed mutual coupling, direct coupling, or communication connection may be through some interfaces; the indirect coupling or communication connection between units or modules may be electrical or other forms.

[0105] Furthermore, the functional units in the various embodiments of the present invention can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0106] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, read-only memory (ROM), random access memory (RAM), portable hard drives, magnetic disks, or optical disks.

[0107] The above description is only a preferred embodiment of the present invention. It should be noted that for those skilled in the art, several improvements and modifications can be made without departing from the principle of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention.

Claims

1. A method for online efficiency operation status detection based on industrial flow data, characterized in that, Includes the following steps: The raw data streams of multiple homogeneous production units are periodically summarized to generate periodic sampling records for each production unit in each sampling period. A new decision unit block is constructed based on the periodic sampling records, and each decision unit in the new decision unit block is represented by an input vector and an output vector. Obtain the historical decision unit set up to the previous sampling period, along with its efficiency value and multiplier solution. Based on the default values ​​between the historical decision units and each decision unit in the newly added decision unit block, determine the historical affected set and identify the decision units in the newly added decision unit block that can be directly confirmed as effective or highly effective. This yields the newly added effective decision units and the newly added highly effective decision units. Update the current effective set and highly effective set. Based on the historical affected set and the decision units in the newly added decision unit block that were not directly identified as effective or highly effective, construct a set to be evaluated, and construct a mandatory reference set mapping for the decision units in the set to be evaluated. Using the required reference set mapping as a constraint, the efficiency of the decision units in the set to be evaluated is solved on the limited-size subsample reference set to obtain the efficiency value and multiplier solution of each decision unit. Based on the efficiency value and multiplier solution, the newly added highly efficient decision unit, and the highly efficient set and efficiency value of the previous sampling period, the running status detection result is generated; Methods for periodically summarizing the raw data streams of multiple homogeneous production units to generate periodic sampling records for each production unit in each sampling period include: Based on the preset sampling period length and period index, the start and end times of the current sampling period are determined to form a period time window; For each production unit, various data segments whose timestamps fall within the cycle time window are extracted from the raw data stream, including operation status data segments, event log data segments, energy consumption metering data segments, and output and quality count data segments. The extracted data fragments are processed for field normalization, mapping the field names, data types and units of measurement to a preset standard format. Basic data quality marking is performed on each of the extracted data segments, including: if there are records with the same field reported repeatedly under the same timestamp, deduplication is performed and a duplicate anomaly is marked; if there are missing key fields or data types that cannot be parsed, a missing field anomaly is marked; if there are abnormal values ​​due to out-of-bounds values ​​or inconsistent units, an out-of-bounds anomaly is marked. Based on the operational status data segments, the cumulative downtime and cumulative micro-stop time of the production unit within the cycle time window are statistically analyzed, and the consistency of start and end times is checked in conjunction with the event record data segments. If time inconsistency is found or the sum of the cumulative downtime and cumulative micro-stop time exceeds the sampling cycle length, the time is marked as abnormal. The energy consumption increment of the production unit within the cycle time window is calculated based on the energy consumption metering data segment. If the calculation result is negative or there is an obvious out-of-bounds reading, an out-of-bounds anomaly is marked. Based on the production and quality count data segments, the total production increment, qualified production increment, and first-pass qualified quantity increment of the production unit within the cycle time window are statistically analyzed. If the logical relationship that the first-pass qualified quantity does not exceed the qualified production and the qualified production does not exceed the total production is not satisfied, then a logical anomaly is marked. The statistically obtained cumulative downtime, cumulative micro-downtime, energy consumption increment, total output increment, qualified output increment, and first-pass qualified output increment, together with the production unit identifier, cycle index, cycle time window, sampling cycle length, and marked data quality markers, are assembled into a cycle sampling record. Methods for determining the historically affected set based on the default values ​​between each decision unit in the historical decision unit and the newly added decision unit block include: For each historical decision unit and each decision unit in the newly added decision unit block, calculate the default value: in, This is the default value; An index for historical decision-making units; For the newly added decision unit block, the decision unit index is provided. For historical decision-making units In the previous sampling period The output multiplier vector; For historical decision-making units In the previous sampling period The input multiplier vector; For historical decision-making units In the previous sampling period The free variables; For the newly added decision-making unit The input vector; For the newly added decision-making unit The output vector; Represents the transpose of a vector; For each historical decision unit, calculate its maximum default value on the newly added decision unit block: in, The maximum default value; For the newly added decision-making unit block set; The historical decision-making units with the maximum default value greater than zero are included in the historical affected set.

2. The online efficiency operation status detection method based on industrial flow data as described in claim 1, characterized in that, The method for constructing a new decision unit block based on the periodic sampling records includes: Obtain available period sampling records for each production unit in each sampling period. The available period sampling records are period sampling records with data quality marked as empty set or containing only duplicate anomaly marks. For each available cycle sampling record, extract the production unit identifier, cycle index, sampling cycle length, cumulative downtime, cumulative micro-downtime, energy consumption increment, total output increment, qualified output increment, and first-pass qualified output increment. The qualified output increment, the first-pass qualified output increment, and the total output increment are assembled into a production vector in a preset order; The cumulative downtime, cumulative micro-downtime, and energy consumption increment are used as the first three input indicators, and the speed loss indicator is calculated as the fourth input indicator. The speed loss indicator is calculated as follows: the theoretical output is obtained by multiplying the sampling period length by the target production rate effective within the sampling period of the production unit. The difference between the theoretical output and the actual total output increment is taken. If the difference is positive, the difference is taken; otherwise, it is taken as zero. The cumulative downtime, cumulative micro-downtime, energy consumption increment, and speed loss indicator are assembled into an input vector in a preset order. The assembled input vector and output vector are used as decision units corresponding to the available periodic sampling records, and the index of the decision unit is added to the set of new decision unit blocks to construct new decision unit blocks.

3. The online efficiency operation status detection method based on industrial flow data as described in claim 2, characterized in that, Methods for identifying newly added effective or highly effective decision units within a newly added decision unit block, and for obtaining newly added effective and highly effective decision units, include: For each decision unit in the newly added decision unit block, calculate the maximum default value of the newly added decision unit on the set of historical decision units, and obtain the historical decision units that have reached the maximum default value as reference objects; If both of the following conditions are met simultaneously, the newly added decision unit and other newly added decision units associated with the same reference object and having the same maximum default value are jointly identified as newly added valid decision units: First, the maximum default value of the newly added decision unit on the historical decision unit set is greater than zero; Second, the maximum default value is equal to the maximum default value of the historical decision unit used as the reference object on the newly added decision unit block. For each decision unit that is identified as a new valid decision unit, its multiplier solution is constructed to be the same as that of the historical decision unit that serves as the reference, and the free variables are updated to be the sum of the free variables of the historical decision unit and its maximum default value on the new decision unit block. New highly effective decision-making units are selected from the set of newly identified effective decision-making units. The selection rules include: if a newly identified effective decision-making unit is the only element in its associated set of decision-making units, or has the smallest value in its associated set of decision-making units for a certain input indicator, or has the largest value in its associated set of decision-making units for a certain output indicator, or has the best input-output ratio in its associated set of decision-making units, then the newly identified effective decision-making unit is identified as a newly identified highly effective decision-making unit.

4. The online efficiency operation status detection method based on industrial flow data as described in claim 1, characterized in that, The method for constructing a necessary reference set mapping for the decision units in the set to be evaluated includes: For each decision unit in the historical affected set, obtain the set of new decision units corresponding to the point when it reaches the maximum default value in the new decision unit block: And map the newly added set of decision-making units as the mandatory reference set of decision-making units in the historically affected set; For each decision unit in the set of decision units that are not directly identified as valid or highly valid in the newly added decision unit block, its mandatory reference set mapping is set to an empty set.

5. The online efficiency operation status detection method based on industrial flow data as described in claim 1, characterized in that, The constrained-size subsample reference set is constructed as follows: For any target decision unit in the set to be evaluated, obtain the required reference set of the target decision unit and use it as the initial member of the subsample reference set; When the number of members in the subsample reference set does not reach the preset upper bound, target decision units are selected from the historical valid set in order of increasing difference from the target decision unit in terms of input-output characteristics, until the number of members in the subsample reference set reaches the upper bound minus one. The target decision-making unit itself is included in the subsample reference set to obtain a complete restricted-size subsample reference set.

6. The online efficiency operation status detection method based on industrial flow data as described in claim 4, characterized in that, Methods for obtaining the efficiency values ​​and multiplier solutions of each decision-making unit include: For any target decision unit in the set to be evaluated, solve the multiplier model on the constructed restricted-size subsample reference set to obtain the candidate efficiency value and candidate multiplier solution of the target decision unit; The candidate multiplier solution is used to perform constraint checks on all currently available decision units, and the default indication of each decision unit for the candidate multiplier solution is calculated. If the default indication does not exceed zero, then the candidate efficiency value and candidate multiplier solution are confirmed as the final efficiency value and multiplier solution of the target decision unit, and the solution is then entered into the next target decision unit. If the default indication is greater than zero, the set of decision units that cause the default is identified. Decision units that meet the extremely efficient condition are selected from the default decision unit set and added to the extremely efficient set. The default decision unit set is then extended to the subsample reference set to form an updated subsample reference set. The multiplier model is re-solved on the updated subsample reference set until the default indication does not exceed zero, thus obtaining the final efficiency value and multiplier solution of the target decision unit.

7. The online efficiency operation status detection method based on industrial flow data as described in claim 5, characterized in that, Methods for generating runtime status detection results include: Obtain the most efficient set of the current sampling period and the previous sampling period, and generate an early warning of changes in the efficiency frontier state by comparing the changes in the most efficient set of the two periods; Obtain the historical affected set and the efficiency change of each decision unit in the set between adjacent periods. If the proportion of the number of decision units with decreased efficiency in the historical affected set to the total number of the historical affected set reaches a preset collective decrease proportion threshold, a historical collective efficiency decrease warning is generated. The number of times each decision unit in the historical affected set and the newly added decision unit block is referenced in the mandatory reference set mapping is obtained. If the number of references of a newly added decision unit reaches the preset single-point impact coverage intensity threshold and the proportion of the total number of the historical affected set reaches the collective decline ratio threshold, then the newly added decision unit is identified as a strong impact source point, and a strong impact source location warning is generated. For each production unit, obtain its efficiency sequence within the preset analysis window length, and count the proportion of decision units with efficiency values ​​lower than the preset inefficiency threshold to the total number of decision units within the window. If this proportion reaches the preset inefficiency ratio threshold, generate a long-term inefficiency solidification warning for that production unit, and set the warning level according to the severity of the inefficiency ratio. For each production unit, the efficiency jump amplitude between adjacent sampling periods within the preset analysis window length is obtained. If there are multiple consecutive jump amplitudes exceeding the preset adjacent efficiency jump threshold, an abnormal efficiency fluctuation warning for that production unit is generated. The generated warnings are summarized in layers according to system level, block level, and unit level. Unit-level warnings of the same production unit are merged according to preset priority rules, and the final operation status detection results are output.

8. An online efficiency operation status monitoring system based on industrial flow data, characterized in that, include: The data acquisition and access module is used to periodically summarize the raw data streams of multiple homogeneous production units and generate periodic sampling records for each production unit in each sampling period. The decision unit construction module is connected to the data acquisition and access module and is used to construct new decision unit blocks based on the periodic sampling records. Each decision unit is represented by an input vector and an output vector. The efficiency update object identification module, connected to the decision unit construction module, is used to obtain the historical decision unit set up to the previous sampling period, along with its efficiency value and multiplier solution. Based on the default values ​​between the historical decision units and each decision unit in the newly added decision unit block, it determines the historical affected set and identifies decision units in the newly added decision unit block that can be directly identified as effective or highly effective, thus obtaining newly effective decision units and newly highly effective decision units, and updating the current effective set and highly effective set. It is also used to construct an evaluation set based on the historical affected set and the decision units in the newly added decision unit block that are not directly identified as effective or highly effective, and to construct a mandatory reference set mapping for the decision units in the evaluation set. An efficiency acceleration solution module, connected to the efficiency update object identification module, is used to solve the efficiency of decision units in the set to be evaluated on a limited-size subsample reference set with the mandatory reference set mapping as a constraint, and to obtain the efficiency value and multiplier solution of each decision unit. The operation status change early warning module is connected to the efficiency acceleration solution module and is used to generate operation status detection results based on the efficiency value and multiplier solution, the newly added highly efficient decision unit, and the highly efficient set and efficiency value of the previous sampling period. Methods for periodically summarizing the raw data streams of multiple homogeneous production units to generate periodic sampling records for each production unit in each sampling period include: Based on the preset sampling period length and period index, the start and end times of the current sampling period are determined to form a period time window; For each production unit, various data segments whose timestamps fall within the cycle time window are extracted from the raw data stream, including operation status data segments, event log data segments, energy consumption metering data segments, and output and quality count data segments. The extracted data fragments are processed for field normalization, mapping the field names, data types and units of measurement to a preset standard format. Basic data quality marking is performed on each of the extracted data segments, including: if there are records with the same field reported repeatedly under the same timestamp, deduplication is performed and a duplicate anomaly is marked; if there are missing key fields or data types that cannot be parsed, a missing field anomaly is marked; if there are abnormal values ​​due to out-of-bounds values ​​or inconsistent units, an out-of-bounds anomaly is marked. Based on the operational status data segments, the cumulative downtime and cumulative micro-stop time of the production unit within the cycle time window are statistically analyzed, and the consistency of start and end times is checked in conjunction with the event record data segments. If time inconsistency is found or the sum of the cumulative downtime and cumulative micro-stop time exceeds the sampling cycle length, the time is marked as abnormal. The energy consumption increment of the production unit within the cycle time window is calculated based on the energy consumption metering data segment. If the calculation result is negative or there is an obvious out-of-bounds reading, an out-of-bounds anomaly is marked. Based on the production and quality count data segments, the total production increment, qualified production increment, and first-pass qualified quantity increment of the production unit within the cycle time window are statistically analyzed. If the logical relationship that the first-pass qualified quantity does not exceed the qualified production and the qualified production does not exceed the total production is not satisfied, then a logical anomaly is marked. The statistically obtained cumulative downtime, cumulative micro-downtime, energy consumption increment, total output increment, qualified output increment, and first-pass qualified output increment, together with the production unit identifier, cycle index, cycle time window, sampling cycle length, and marked data quality markers, are assembled into a cycle sampling record. Methods for determining the historically affected set based on the default values ​​between historical decision units and each decision unit in the newly added decision unit block include: For each historical decision unit and each decision unit in the newly added decision unit block, calculate the default value: in, This is the default value; An index for historical decision-making units; For the newly added decision unit block, the decision unit index is provided. For historical decision-making units In the previous sampling period The output multiplier vector; For historical decision-making units In the previous sampling period The input multiplier vector; For historical decision-making units In the previous sampling period The free variables; For the newly added decision-making unit The input vector; For the newly added decision-making unit The output vector; Represents the transpose of a vector; For each historical decision unit, calculate its maximum default value on the newly added decision unit block: in, The maximum default value; For the newly added decision-making unit block set; The historical decision-making units with the maximum default value greater than zero are included in the historical affected set.

Images