Industrial collaborative scheduling method and system based on park industrial chain short board analysis

Abstract

The invention provides an industrial collaborative scheduling method and system based on park industrial chain short board analysis, and belongs to the technical field of park industrial chain digital management and intelligent collaboration, and the method comprises the following steps: S1, collecting original data of a park industrial chain; s2, integrating to form an average standardized value; s3, forming a park industry chain short board list; s4, forming a park industry chain short board priority ranking table; s5, setting a park industry collaborative scheduling target table; s6, in combination with original data of a park industry chain, designing specific solution measures of each short board, and forming a park industry collaborative scheduling implementation scheme; s7, adopting a KF algorithm to form a monthly monitoring evaluation report; and S8, calculating the target achievement rate of each short board, and adjusting the target table or the implementation scheme according to the analysis result. Through deep fusion of six algorithms and strong association of each link, transformation of the park industry chain from passive response to active optimization is realized, and the overall collaboration and competitiveness of the industry chain are improved.

Description

Technical Field

[0001] The present invention relates to the field of digital management and intelligent collaboration technology of industrial parks' industrial chains, and in particular to an industrial collaboration scheduling method and system based on analysis of shortcomings of industrial parks' industrial chains. Background Art

[0002] Currently, industrial chain management in industrial parks generally faces four core pain points, and existing technical solutions are difficult to provide effective support: Data fragmentation and insufficient value mining: Park enterprise data is stored in separate systems such as production, supply chain, and finance. There is a lack of unified collection and standardized processing mechanisms. These data "islands" make it impossible to accurately reflect the operating status of the industrial chain and make it difficult to support decision-making. Weakness identification is subjective and incomplete: Traditional weakness identification relies on manual research and empirical judgment, without integrating data dispersion into the importance of quantitative indicators. This can easily lead to missing links with small but critical data differences (such as the technology transfer cycle), and is unable to identify cross-link coordination bottlenecks (such as the connection between upstream supply and midstream production). Inefficient resource allocation and poorly targeted solutions: Production task allocation and supplier selection rely on "manual equalization" or "proximity" without considering multiple factors such as production capacity, cost, and target alignment, leading to resource waste. Solution formulation lacks alignment with weak point priorities and resource carrying capacity, leading to "overly ambitious goals that cannot be achieved" or "measures that are disconnected from pain points." Lagging execution progress control and lack of closed-loop: Program execution monitoring is mostly "post-statistics" (such as monthly summaries), which cannot predict progress deviations in real time; there is a lack of effective feedback and adjustment mechanisms, and when goals are not achieved, only passive responses can be made, making it difficult to form a continuous improvement chain of "identification-program-execution-optimization".

[0003] These pain points often lead to problems in the industrial chain of the park, such as "upstream supply delays affecting midstream production", "disconnection between technology research and development and market application", and "mismatch between talent allocation and industrial demand", which restrict the improvement of the overall competitiveness of the industrial chain. Summary of the Invention

[0004] This invention provides an industrial collaborative scheduling method and system based on the analysis of shortcomings in the industrial chain of a park. Focusing on the core goal of "solving shortcomings and collaborative optimization of the industrial chain of a park", it constructs an eight-step fully closed-loop technical system of "data collection - preprocessing - shortcoming identification - impact assessment - goal setting - program formulation - implementation monitoring - effect evaluation". The core logic is as follows: Data base construction: Through standardized collection processes, we integrate multi-source data such as production, supply chain, resource consumption, talent and technology of park enterprises to form an original data set; Data preprocessing: Use the IQR method to eliminate outliers and the min-max method to achieve data standardization, and output a standardized data set that can be directly used for algorithm analysis; Short plate accurate identification: based on EWM algorithm to determine the index weight, combined with industry benchmark data to calculate the index and link gap, identify single link short board and cross link bottleneck; Influence evaluation: through AHP-EWM fusion algorithm to balance subjective experience and objective data, calculate the short board comprehensive influence score, and form the priority ranking; Target quantification: according to the short board priority, set the core index target value (such as delivery period, R&D personnel ratio), and clarify the achievement period and resource demand; Intelligent scheme formulation: for high priority short board, design specific measures, embed GA algorithm to optimize production task, supplier selection and other resource allocation problems, and form executable scheme; Real-time execution monitoring: based on KF algorithm to predict the progress of scheme execution, establish a hierarchical early warning mechanism, and dynamically track the progress and risk; Effect evaluation and optimization: calculate the target achievement rate, find out the reasons for reaching the standard / non-standard, dynamically adjust the target and scheme, and form a closed-loop management of continuous iteration.

[0005] The scheme realizes the transformation of the park industrial chain from "passive response" to "active optimization" through the deep fusion of six algorithms (IQR, min-max, EWM, AHP-EWM, GA, KF) and the strong correlation of each link, and improves the overall synergy and competitiveness of the industrial chain.

[0006] In order to achieve the above purpose, the application adopts the following technical scheme: The industrial collaborative scheduling method based on park industrial chain short board analysis comprises the following steps: S1: collect park industrial chain original data, the original data includes enterprise basic information, product production data, supply chain data, resource consumption data and technology and talent data, and integrate to form park industrial chain original data set; S2: preprocess the original data set, first use the interquartile range method (IQR) to detect outliers in the quantitative indicators in the original data, and then use the min-max standardization method to standardize the normal data, while the qualitative indicators in the original data are coded, and the park industrial chain standardized data set and the average standardized value of each industrial chain link are integrated; S3: based on the standardized data set and the average standardized value, combined with the industry benchmark data obtained from national official channels, industry associations or third-party research institutions, use the entropy weight method (EWM) to calculate the objective weight of each index in each industrial chain link, and then calculate the weighted gap rate of each index and the comprehensive gap rate of each link through the objective weight, average standardized value and industry benchmark data, identify single link short board and cross link bottleneck, and form park industrial chain short board list; S4: Based on the objective weight in the short board list, the subjective weight of each evaluation index determined by the analytic hierarchy process (AHP), the fusion coefficient of AHP and EWM is calculated by the coefficient of variation method, the subjective weight and the objective weight are fused to obtain the combined weight through the fusion coefficient, and then the comprehensive influence score of each short board is calculated by combining the scores of the short boards in the production value influence degree, stability influence degree and repair difficulty dimension, and a short board priority ranking table of the park industry chain is formed; S5: Based on the comprehensive influence score in the priority ranking table, set the core index target value, target achievement period and auxiliary target corresponding to each short board, and form a park industry collaborative scheduling target table; S6: Based on the core index target value and auxiliary target in the target table, combined with the production capacity, order demand and other data in the original data of the park industry chain, the genetic algorithm (GA) is used to optimize the production task allocation, supplier selection or technology transformation project sorting, and the specific solution measures of each short board are designed, and a park industry collaborative scheduling implementation scheme is formed; S7: Based on the GA optimization result in the implementation scheme as the initial state, the Kalman filter (KF) algorithm is used to predict the progress of the scheme in real time, and the real-time collected scheme execution data is used as the observation value to dynamically correct the progress prediction result. If the progress lag reaches the preset warning threshold, an early warning is triggered, and a monthly monitoring and evaluation report is formed; S8: Based on the progress prediction result in the monthly monitoring and evaluation report and the core index target value in the target table, the target achievement rate of each short board is calculated, the attribution analysis of the achievement is carried out from the measure effectiveness, resource input, external environment and data quality dimension, and the target table or the implementation scheme is adjusted according to the analysis result, and fed back to S5 or S6, forming a closed loop of "data collection-preprocessing-short board identification-influence evaluation-target setting-scheme development-execution monitoring-effect evaluation-adjustment and optimization".

[0007] In the specification, in step S2, the abnormal value detection result of the interquartile range method (IQR) is used as the input data of the min-max standardization method, specifically: for each quantitative index of each industry chain link, calculate the quartiles Q1, Q3 and interquartile range IQR, remove the abnormal values outside the range [Q1-1.5xIQR, Q3+1.5xIQR], obtain the normal data, and then standardize the normal data based on the minimum and maximum values of each quantitative index of each link, to ensure that the standardized data is used for the entropy weight method calculation in step S3 without abnormal interference.

[0008] In the present specification, in step S3, the interaction process of the entropy weight method (EWM) and the industry benchmark data is as follows: first, the objective weights of each index are calculated by the entropy weight method to highlight the importance of indexes with high data dispersion, and then the objective weights are substituted into the index weighted gap rate calculation formula, so that indexes with higher weights have greater influence on short board identification, ensuring that the identified single-link short board and cross-link bottleneck are more in line with the core needs of the industrial chain, and each short board information in the short board list is directly used as the basis data for calculating the comprehensive influence score in step S4.

[0009] In the present specification, in step S4, the core improvement of the AHP-EWM fusion algorithm is as follows: the variation coefficients of AHP subjective weights and EWM objective weights are calculated by the variation coefficient method, and the inverse ratio of the variation coefficient is used as the fusion coefficient, so that the weights with smaller dispersion and higher reliability have a higher proportion in the fusion, and the combined weights are directly used for calculating the comprehensive influence score of each short board, and the comprehensive influence score is the only basis for setting the target coefficient in step S5.

[0010] In the present specification, in step S6, the interaction process of the genetic algorithm (GA) and the target table is as follows: the core index target values in the target table (such as equipment utilization target, delivery cycle target) are used as the fitness function constraint conditions of GA, and in the iteration process of GA, the solution that meets the target value requirement and has the optimal resource consumption is selected by selection, crossover and mutation operators, and the GA optimization result is not only used as the initial state of Kalman filtering in step S7, but also used as the execution basis for production task allocation or supplier selection in the implementation scheme.

[0011] In the present specification, in step S7, the interaction process of the progress prediction and early warning mechanism of Kalman filtering (KF) is as follows: the state equation of KF is based on the planned progress in the implementation scheme to set the state transition matrix and the control matrix, and the observation equation is based on the real-time collected execution data (such as equipment update progress, order delivery rate) to set the observation matrix, and after each iteration, the optimal progress estimation value is obtained by balancing the predicted value and the observed value through Kalman gain; when the deviation between the optimal progress estimation value and the planned progress in the target table exceeds 10%, a yellow warning is triggered, and when the deviation exceeds 20%, a red warning is triggered, and the warning information is fed back to step S8 for attribution analysis.

[0012] In the present specification, in the closed loop formed by step S8, the output data of each link is used as the input data of the corresponding link in the next round, specifically: the adjusted target table of S8 is fed back to S5 as the update basis for the next round of target setting; the adjusted implementation scheme of S8 is fed back to S6 as the basis for the next round of scheme optimization; the attribution analysis result of S8 is also fed back to S3 for updating the gap rate calculation parameters of short board identification, to ensure that the closed loop continuously adapts to the changes of the park industrial chain.

[0013] In the specification, in step S8, the end condition of the closed loop is that, in the two consecutive target achievement periods, the target achievement rates of all the extremely high priority and high priority short boards are all ≥95%, and the target achievement rates of the medium and low priority short boards are all ≥80%, or the expert committee assesses and confirms that the core short board of the park industry chain has been eliminated, at which time the next round of full-process circulation can be entered, otherwise the closed loop optimization is continuously promoted by adjusting the target or scheme.

[0014] In the specification, in steps S3 to S7, the parameters of each algorithm adopt a dynamic optimization mechanism: the entropy weight method weight of step S3 is updated every half year, and is recalculated based on the latest standardized data set; the AHP subjective weight of step S4 is updated every year by re-scoring of experts; the process noise covariance and observation noise covariance of step S7 Kalman filter are corrected based on the monitoring data deviation of the last year, to ensure that the algorithm parameters continuously fit the actual application scene.

[0015] In the specification, the attribution analysis of step S8 adopts a multi-dimensional linkage mechanism: if the attribution result is that the measure effectiveness is insufficient (such as slow response of technical service providers), the implementation scheme of step S6 is adjusted, and service provider incentive measures are supplemented; if the attribution result is that the resource investment is insufficient (such as the lack of talent subsidies), the target resource demand of step S5 is adjusted, and the park financial supplementary funds are coordinated; if the attribution result is external environmental influence (such as extreme weather), the KF early warning threshold of step S7 is adjusted, and an external risk response coefficient is added, to ensure that the attribution analysis result directly guides the closed loop adjustment.

[0016] The industrial collaborative scheduling system based on park industry chain short board analysis applies any one of the industrial collaborative scheduling methods based on park industry chain short board analysis described above, and the industrial collaborative scheduling system based on park industry chain short board analysis comprises: A collection module for collecting park industry chain original data to form a park industry chain original data set; A preprocessing module for preprocessing the park industry chain original data set to form a park industry chain standardized data set and average standardized values of each industry chain link; An identification module for calculating the objective weight of each index in each industry chain link by entropy weight method based on the park industry chain standardized data set and the average standardized values of each industry chain link, combining with industry benchmark data, and then calculating the weighted gap rate of each index and the comprehensive gap rate of each link by the objective weight, the average standardized value and the industry benchmark data, to identify single-link short boards and cross-link bottlenecks, and form a park industry chain short board list; The sorting module is used for determining subjective weights of each evaluation index based on objective weights in the park industry chain short board list in combination with an analytic hierarchy process, calculating a fusion coefficient of the analytic hierarchy process and the entropy weight method by using a coefficient of variation method, fusing the subjective weights and the objective weights to obtain combined weights through the fusion coefficient, combining scores in the output value influence degree, the stability influence degree and the repair difficulty dimension of the single-link short board, calculating a comprehensive influence score of the single-link short board, and forming a park industry chain short board priority ranking table; The setting module is used for setting a core index target value, a target achievement period and an auxiliary target corresponding to the single-link short board based on the comprehensive influence score in the park industry chain short board priority ranking table, and forming a park industry collaborative scheduling target table; The optimization module is used for optimizing production task allocation, supplier selection or technology transformation project sequencing based on the core index target value and the auxiliary target in the park industry collaborative scheduling target table in combination with a park industry chain original data set, designing specific solutions for the single-link short board, and forming a park industry collaborative scheduling implementation scheme.

[0017] To sum up, the present application has at least the following advantages: The present application realizes four core technical effects through full-process algorithm fusion and closed-loop management, and can show significant value without relying on specific numerical values: Short board identification is more accurate and objective: the EWM algorithm quantifies the importance of the index, and combines industry benchmark data and cross-link connection analysis to not only locate the single-link dominant short board (such as long supply cycle), but also to excavate the cross-link implicit bottleneck (such as insufficient technology flow connection), avoiding the subjectivity and one-sidedness of manual judgment; Resource allocation is more efficient and adaptive: the GA algorithm can find the optimal solution under multiple constraints (such as capacity, cost, target), realize the precise matching of production tasks, suppliers and technology projects, and reduce resource waste; the AHP-EWM fusion algorithm balances subjective experience and objective data to ensure that resources are tilted towards high-priority short boards, and improves the input-output ratio; Progress control is more dynamic and timely: the KF algorithm predicts progress deviation in advance, realizes risk "early detection, early treatment" through a hierarchical early warning mechanism (yellow / red warning), avoids target failure due to lagging control, and combines real-time monitoring data with algorithm prediction to provide "real-time portrait" rather than "post-mortem"; The resilience and synergy of the industry chain are significantly improved: through the full closed loop of "data-algorithm-decision-feedback", the short board solution and target setting are continuously optimized to enhance the ability of the industry chain to respond to external changes (such as extreme weather and market fluctuations); the cross-link synergy mechanism (such as upstream-middlestream supply coordination and middlestream-downstream technology transformation) strengthens the linkage of each link in the industry chain, reduces the risk of "chain breakage", and improves overall operational efficiency.

[0018] In addition, the modular design of the scheme (data acquisition, preprocessing, identification, evaluation, target, scheme, monitoring, evaluation) can be flexibly adapted to different industrial parks (such as new energy, electronic information), and has strong universality and scalability. BRIEF DESCRIPTION OF DRAWINGS

[0019] Figure 1 A schematic diagram of an industrial collaborative scheduling method based on short board analysis of an industrial chain in a park involved in the present application.

[0020] Figure 2 A schematic diagram of a short board influence evaluation process involved in the present application.

[0021] Figure 3 A schematic diagram of a target and scheme development process involved in the present application.

[0022] Figure 4 A schematic diagram of an industrial collaborative scheduling system based on short board analysis of an industrial chain in a park involved in the present application. DETAILED DESCRIPTION

[0023] The embodiments of the present application will be described in detail below with reference to the accompanying drawings.

[0024] As shown in Figure 1 , the present embodiment provides an industrial collaborative scheduling method based on short board analysis of an industrial chain in a park, comprising: S1: Park industrial chain basic data acquisition 1.1 Data acquisition range and type Enterprise number: i (i=1, 2,..., N, N is the total number of enterprises in the park, such as i=001 represents the first enterprise in the park); Product number: j (j=1, 2,..., M, M is the total number of main products of the enterprise, such as j=001 represents the first product of the enterprise); Time period: t (t=1, 2,..., T, T is the number of data acquisition periods, monthly data t=1 represents January, t=12 represents December); Supplier number: s (s=1, 2,..., S, S is the total number of upstream suppliers of the enterprise); Customer number: c (c=1, 2,..., C, C is the total number of downstream customers of the enterprise); Raw material number: k (k=1, 2,..., K, K is the total number of raw materials used by the enterprise); Technology number: m (m=1, 2,..., , Total number of core technologies of the enterprise); Work type number: n (n=1, 2,..., , is the total number of skilled talent jobs); Industrial chain links: g (g=up=upstream raw material supply, g=mid=midstream production and manufacturing, g=down=downstream product sales, g=sup=supporting services).

[0025] The following five types of data are collected: 1. Basic information of the enterprise: enterprise name, industry chain link, registered address, contact information, establishment time, enterprise scale (number of employees) : Small <100 people, medium 100≤ ≤500 people, large >500 people), annual output value (10,000 yuan), sourced from the enterprise registration system of the park management committee and the basic information form filled out annually by the enterprise; 2. Product production data: main product names, monthly output (units / pieces), production cycle (days / batch), number of production equipment (unit), equipment operating status (e=device number, Status=1 running, 0 shut down), product qualification rate = (number of qualified products / total output) × 100%, sourced from the monthly production report submitted by the enterprise production management system before the 5th of each month; 3. Supply chain data: upstream supplier name, product type, and delivery cycle (days, from order placement to delivery), monthly supply (tons / unit), procurement cost (Yuan / unit); downstream customer name, sales product type, sales cycle (days, from order receipt to delivery), monthly sales volume (unit / piece), sales price (Yuan / unit), sourced from data exported from the enterprise supply chain management system + real-time data captured by the park supply chain monitoring platform; 4. Resource consumption data: Monthly energy consumption (electricity kWh, natural gas Cubic meters, water resources Tons), monthly raw material consumption (tons / unit), land area (square meters), sourced from the park's energy management center's metering data (electricity / natural gas / water resources), the company's raw material procurement records, and the park's natural resources bureau's land use registration certificate; 5. Technology and talent data: core technology name, technology maturity (1-5 scale: 1 = laboratory stage, 2 = small test stage, 3 = pilot stage, 4 = early industrialization, 5 = mature industrialization), number of R&D personnel (undergraduate and below people, master people, doctor people), number of skilled personnel (by type: e.g. mechanical operator n = 1, electrician n = 2), from the annual technical report of the enterprise R&D department, and the talent record table of the park human resources bureau.

[0026] 1.2 Data collection and preprocessing After data aggregation through the park unified data collection platform, the original data set is formed

[0027] This data set needs to be imported into S2 for preprocessing, because the original data has three major problems: first, there are abnormal values of output caused by equipment failure, and cost abnormal values caused by input errors; second, the quantitative indicators have large dimension differences (such as output unit "unit", energy unit "kilo watt hour"); third, the data format is not unified (such as the qualitative indicator "technology maturity" is described in words), which needs to be eliminated through the algorithm of S2 to provide usable data for subsequent short board identification.

[0028] S2: preprocessing of industry chain basic data 2.1 Preprocessing target and algorithm selection logic The core goal of preprocessing is to generate a standardized, noise-free, and comparable data set, which lays the foundation for S3 entropy weight method to calculate the index weight. The "interquartile range (IQR) method" is chosen to handle outliers, because the park industry chain data (such as output, delivery cycle) often has extreme values (such as sudden drop in output caused by equipment failure), and IQR method is not sensitive to extreme values, which is more suitable for industrial data than mean-standard deviation method; The "min-max standardization" is chosen to handle quantitative indicators, because this method can compress data to the [0,1] interval, preserving the relative differences between indicators while eliminating the dimension effect (such as output "1000 units" and cycle "10 days" can be directly compared for dispersion); Qualitative indicators are coded to convert non-numeric data (such as technology maturity) into numbers to meet the input requirements of subsequent algorithms.

[0029] 2.2 Outlier detection algorithm (IQR method) 2.2.1 Model construction IQR method identifies outliers by dividing the four quartiles of the data, the specific steps and formula definition are as follows: 1. Quartile calculation: for any quantitative indicator X (such as output , pass rate ), calculate the lower quartile of this indicator in all enterprise data in stage g , the upper quartile : : the lower quartile of indicator X in stage g, i.e. 25% of enterprise data ≤ , the calculation formula is , where represents the original value of indicator X of stage g, enterprise i, is the quantile function in Excel / Python (returning the 25% quantile); : the upper quartile of indicator X in stage g, i.e. 75% of enterprise data ≤ , the calculation formula is ; 2. Quartile range and outlier range: Quartile range (reflecting the distribution range of the middle 50% of data, the larger the range, the higher the data dispersion); Outlier determination range: indicator value or , the formula is: Outlier determination: ; (2-1) Note: 1.5xIQR is the default outlier determination coefficient in statistics, suitable for most industrial data sets.

[0030] 2.2.2 Model training (parameter solidification process) The core of training is to determine the , , of each indicator in each stage as the baseline parameter for subsequent outlier detection, the steps are as follows: 1. Data grouping: group the original data set of S1 according to the industry chain stage g (up / mid / down / sup), for example, group all upstream enterprise (g=up) production data into one group, and group the equipment utilization rate data (g=mid) of the midstream enterprises ( = running time / total time x 100%) into another group; 2. Parameter calculation: for each quantitative indicator X in each group, use the PERCENTILE function to calculate , , and derive , for example: Water consumption of upstream stage (g=up) : calculated = 200 tons, = 800 tons, = 800 - 200 = 600 tons; Midstream segment (g = mid) product pass rate : calculated = 85%, = 95%, = 10%; 3. Parameter solidification: store all segments, all indicators of , , to the "park industry chain outlier detection parameter library" as the judgment basis for subsequent application stage (such as directly calling the library parameters when detecting new data).

[0031] 2.2.3 Model application The application stage needs to combine the parameter library to detect outliers for new data collection, and process according to the "verification-correction-labeling" process, as follows: Example 1: Midstream enterprise A (i = 003, g = mid) 20XX March product A (j = 001) production anomaly detection 1. Extract data: enterprise A 3 product A production = 1200 units; 2. Call parameter library: the = 300 units, = 800 units, = 500 units of midstream segment production index Q; 3. Calculate the outlier range (formula 2-1): [300-1.5×500, 800+1.5×500]=[-450, 1550] units; 4. Judgment: 1200 ∈ [-450, 1550], is normal data, keep the original value.

[0032] Example 2: Upstream enterprise B (i = 005, g = up) 20XX March water resource consumption anomaly detection 1. Extract data: enterprise B 3 water resource consumption = 1800 tons; 2. Call parameter library: the = 200 tons, = 800 tons, = 600 tons of upstream segment water resource index; 3. Calculate the outlier range: [200-1.5×600, 800+1.5×600]=[-700, 1700] tons; 4. Judgment: 1800> 1700, is an outlier, start the processing flow: 4.1. Verification: contact enterprise B to verify the data Step 1: The park big data center staff contacts enterprise B (feedback within 1 working day), verifies the abnormal reason; Step 2: Enterprise B feedbacks that "due to the addition of a production line in March, the water consumption has increased", which belongs to actual abnormality, not input error; Step 3: Mark "actual abnormality (addition of production line)" in the data, and keep the original value (if it is input error, such as recording 1 "0" more, correct it to 180 tons and recheck).

[0033] 2.3 Data standardization algorithm (min-max method + qualitative coding) 2.3.1 Quantitative index standardization (min-max method) Model construction The min-max standardization converts the original value of the quantitative index X into a standardized value in the interval [0, 1] , the formula is: ; (2-2) : standardized value of link g, enterprise i, index X (0≤ ≤1); : minimum value of index X in link g (statistically from S1 original data, such as =100 tons in the upstream link ); : maximum value of index X in link g (such as =1700 tons in the upstream link , including actual abnormal value); denominator : value range of index X in link g, if the range is 0 (all enterprises have the same value of this index), then =0.5 (default middle value).

[0034] Model training and application Training steps: calculate , of each quantitative index according to link g, for example: Downstream link (g=down) sales price : =100 yuan / unit, =500 yuan / unit; Supporting service link (g=sup) R&D personnel ratio =( ) / ×100%: =5%, =30%.

[0035] Application example 1: water resource consumption standardization of upstream enterprise B Enterprise B water resource consumption in March = 1800 tons (actual outlier), = 100 tons, = 1700 tons, substitute formula (2-2): ; Because the standardized value needs to be in the interval [0, 1], it is truncated to 1.0 (outliers exceeding the maximum value are uniformly set to 1.0).

[0036] Application Example 2: Downstream Enterprise C (i = 008, g = down) sales price standardization Enterprise C sales price of product B (j = 002) in March = 300 yuan / unit, = 100 yuan / unit, = 500 yuan / unit, substitute formula (2-2): .

[0037] 2.3.2 Qualitative index coding For non-numeric indicators, coding is done according to the "one-to-one correspondence" principle to ensure that subsequent algorithms can be calculated. The coding rules are as follows: 1. Technology maturity : 1st level → 1, 2nd level → 2, 3rd level → 3, 4th level → 4, 5th level → 5, for example, Enterprise A core technology A (m = 001) is in the pilot stage (3rd level), the coding is = 3; 2. Enterprise size: small ( <100 people) → 1, medium (100 ≤ ≤ 500 people) → 2, large ( > 500 people) → 3, for example, Enterprise B has 200 employees (medium), the coding is = 2; 3. Equipment operating status : running (1) → 1, shutdown (0) → 0, for example, Enterprise C equipment 1 (e = 001) is in shutdown state in March, the coding is = 0.

[0038] 2.4 Data integration 2.4.1 Integration results output Integrate the quantitative standardized data and qualitative coded data after outlier processing according to the "industrial chain link-enterprise-index" three-dimensional structure to form the park industrial chain standardized data set ( is the qualitative coded data), and calculate the average index value of each link: (2-3) where is the number of enterprises in the gth link, e.g. 10 enterprises in the upstream link (g = up) with = 10), and the sum of the normalized water consumption values is 5.2, then = 5.2 / 10 = 0.52.

[0039] 2.4.2 Connection logic with S3 is the core input of the S3 entropy weight method, for two reasons: first, the entropy weight method needs to calculate the weight based on the dispersion of the index, and the standardized data eliminates the dimension effect (such as “0.5” of output and “0.6” of period can be directly compared for dispersion), ensuring that the weight calculation is unbiased; second, can be directly compared with industry benchmark data, providing a link benchmark value for the “gap rate” calculation of subsequent short board identification - if not standardized, the “90% pass rate” of the industry benchmark and the “80% pass rate” of the park can be directly compared, but the “10-day delivery period” of the benchmark and the “20-day delivery period” of the park can be compared due to the same dimension, while the “500-unit output” of the benchmark and the “10-day period” of the park cannot be directly included in the same calculation. After standardization, all indicators are in the [0, 1] interval, and the gap rate can be calculated uniformly. S3: Short board identification of industrial chain

[0040] 3.1 Identification logic and algorithm selection Short board identification needs to solve the problem of “which indicators are key short boards”, and the core is to quantify the importance of indicators (weight) and the gap between the park and the industry (gap rate). The entropy weight method (EWM) is chosen to calculate the objective weight because this method is based entirely on the dispersion data of D_{std}, avoiding the influence of expert subjective preferences (such as considering “output” important or “period” important); at the same time, industry benchmark data (sourced from various benchmarks, standard data in the industry, such as: National Bureau of Statistics of China National Key Industry Development Report, Industry Association Industry Chain Link Benchmark Indicator White Paper, etc., which needs to be consistent with the type of park industrial chain, such as electronic information industry chain corresponding to electronic information industry benchmark), determine the short board through “weighted gap rate” - indicators with high weight and large gap rate are the core short board.

[0041] 3.2 Entropy weight method (EWM) 3.2.1 Model construction The core principle of the entropy weight method is that “the higher the dispersion of the index, the smaller the information entropy, and the greater the weight” (high dispersion indicates that the index has a large difference between enterprises and has a more significant impact on the industrial chain), and the specific formula and definition are as follows: ​1. Indicator normalization proportion: calculate the standardized value of link g, enterprise i, and indicator X Proportion of the sum of the standardized values of all enterprises of the link for the indicator , the formula is: ; (3-1) If = 0, then = 1 (the denominator is 0, indicating that all enterprises have the same indicator value, and the dispersion is 0, so assign the same proportion to each enterprise); 2. Information entropy calculation: information entropy measures the information uncertainty of indicator X (0 ≤ ≤ 1, the smaller E, the higher the dispersion), the formula is: ; (3-2) If = 0, then = 0; is a normalization coefficient to ensure in the interval [0, 1]; 3. Entropy weight (objective weight) calculation: convert information entropy to weight (0 ≤ = 1, is the number of indicators of link g), the formula is: ; (3-3) 1- is the information utility value, the larger the utility value, the larger the indicator weight.

[0042] 3.2.2 Model training (parameter calculation and solidification) The training steps are taken as an example of "midstream production and manufacturing link (g = mid)", as follows: 1. Determine the indicators and data: select 4 core indicators for the midstream link ( = 4): equipment utilization (X1), product pass rate (X2), production cycle (X3), and technology maturity (X4), extract the standardized data of 5 enterprises in the midstream of S2 ( = 5) (as shown in Table 3-1): Table 3-1. Standardized data example

[0043] 2. Calculate (formula 3-1): X1 (equipment utilization):​ =0.8+0.6+0.7+0.5+0.9=3.5, then =0.8 / 3.5≈0.23, =0.6 / 3.5≈0.17, =0.7 / 3.5=0.2, =0.5 / 3.5≈0.14, =0.9 / 3.5≈0.26; Similarly, calculate the p value of X2, X3, X4 (omit here, only list the results of X1); 3. Calculate (Formula 3-2): X1: =ln5≈1.61, =0.23×ln0.23+0.17×ln0.17+0.2×ln0.2+0.14×ln0.14+0.26×ln0.26≈-0.23×1.56-0.17×1.77-0.2×1.61-0.14×1.96-0.26×1.45≈-1.52, then =-(-1.52) / 1.61≈0.94; Similarly, calculate =0.96, =0.92, =0.93; 4. Calculate (Formula 3-3): =(1-0.94)+(1-0.96)+(1-0.92)+(1-0.93)=0.06+0.04+0.08+0.07=0.25; =0.06 / 0.25=0.24, =0.04 / 0.25=0.16, =0.08 / 0.25=0.32, =0.07 / 0.25=0.28; 5. Parameter solidification: store the entropy weight of the four indicators of the middle link to the "park industry chain indicator weight library", and calculate and store the entropy weight of all indicators of the upstream, downstream and supporting service links.

[0044] 3.2.3 Model application (combined with gap rate to identify short board) Weighted gap rate calculation Combined with the entropy weight and industry benchmark data, calculate the weighted gap rate of each indicator (A positive value indicates that the park lags behind the benchmark; the larger the value, the more significant the shortcoming). The formula is: ; (3-4) :Industry benchmark link g, standardized value of indicator X (the original benchmark value needs to be standardized according to the min-max method of S2 to ensure consistency with comparable); : The average standardized value of the park link g and index X calculated by S2; if (the park is better than the benchmark), then =0 (does not constitute a shortcoming).

[0045] At the same time, calculate the comprehensive gap rate of link g (the sum of the weighted gap rates of all indicators), the formula is: ; (3-5) Shortcoming determination threshold Setting thresholds (Based on the development stage of the park: 30% for start-up phase, 20% for growth phase, 15% for mature phase), if or , which is determined to be a shortcoming. The following is a complete example of a typical industrial chain (for a growth park, =20%): New energy vehicle industry chain: 1. Upstream link (g=up, battery material supply): Index X=Energy (energy density of positive electrode material): =0.9 (original benchmark 400Wh / kg, after standardization), =0.6 (park average 300Wh / kg, after standardization), =0.35, (<20%, not a short board); Indicator X = SupCycle (supply cycle): =0.3 (original benchmark 7 days, after standardization), =0.7 (average 14 days in the park, after standardization), =0.3, (Absolute value 40%>20%, shortcoming: supply chain connection shortcoming); Comprehensive gap rate =11.67%+40%=51.67% (>20%, there are shortcomings in the link); 2. Midstream (g=mid, parts processing): Indicator X = EquipEff (equipment utilization): =0.9 (original benchmark 90%), = 0.7 (70% average in the park), = 0.24, = 0.24 x 22.22% = 5.33% (<20%, not a short board); Index X = TechMat (processing technology maturity): = 0.8 (original benchmark 4 levels), = 0.5 (average 2.5 levels in the park), = 0.28, = 0.28 x 37.5% = 10.5% (<20%, not a short board); Comprehensive gap rate = 5.33% + 10.5% = 15.83% (<20%, no short board in the link); 3. Downstream link (g = down, vehicle assembly): Index X = AssemEff (assembly efficiency): = 0.8 (original benchmark 50 units / person / day), = 0.5 (average 35 units / person / day in the park), = 0.3, = 0.3 x 37.5% = 11.25% (<20%, not a short board); Index X = R&DProp (R&D personnel ratio): = 0.9 (original benchmark 30%), = 0.4 (average 15% in the park), 0.35, = 0.35 x 55.56% = 19.44% (<20%, not a short board); Comprehensive gap rate = 11.25% + 19.44% = 30.69% (>20%, there is a short board in the link, but no single index exceeds the threshold value); 4. Supporting service link (g = sup, detection service): Index X = TestAcc (detection accuracy): = 0.98 (original benchmark 99.9%), = 0.85 (average 98.5% in the park), = 0.4, = 0.4 x 13.27% = 5.31% (<20%, not a short board); Index X = TestSpeed (detection speed): = 0.9 (original benchmark 100 pieces / hour), = 0.5 (average 60 pieces / hour in the park), = 0.3, ≈ 0.3 x 44.44% ≈ 13.33% (<20%, not short board); Comprehensive Gap Rate = 5.31% + 13.33% = 18.64% (<20%, no short board in the link).

[0046] 3.3 Short Board List Generation 3.3.1 Short Board List (as shown in Table 3-2) Table 3-2. Example of Short Board List

[0047] 3.3.2 Connection Logic with S4 The generated park industry chain short board list needs to be transmitted to S4 "Short Board Impact Assessment" link, and two types of data are core transmitted: 1. Entropy Weight : As the "objective weight input" of S4 AHP-EWM fusion algorithm, combined with AHP subjective weight, a more comprehensive combined weight is formed - if only AHP subjective weight (expert scoring) is used, the discrete characteristics of the data itself (such as "software iteration period" with high dispersion but not emphasized by experts) may be ignored, and the fusion of entropy weight can balance the subjective and objective; 2. Weighted Gap Rate Comprehensive Gap Rate : As the basis for S4 assessment of "short board impact range" - The larger, the more significant the impact of the index on the industry chain; The larger, the more prominent the short board effect of the entire link, which needs to be prioritized in S4 to allocate evaluation resources (such as extremely high priority short boards that require more expert effort to calculate AHP weights).

[0048] At the same time, the evaluation results of S4 (comprehensive impact score ) will verify the rationality of S3 short board identification in reverse: if a certain index is high but is low (such as "delivery cycle" gap is large but impact on output value is small), S3's entropy weight calculation or industry benchmark data needs to be rechecked for reasonableness, forming a preliminary closed loop of "identification-evaluation-verification".

[0049] S4: Short Board Impact Assessment (short board impact assessment process as shown in Figure 2 ) 4.1 Evaluation Core Logic and Index System The short board impact assessment needs to solve the problem of "which short board to solve first". The core is to combine subjective expert experience and objective data weight to quantify the "output value impact, stability destruction, and repair difficulty" of the short board to the industrial chain. The AHP-EWM fusion algorithm is selected because single AHP (analytic hierarchy process) is easily affected by expert subjective preference (such as a certain expert paying more attention to technology and ignoring cost), and single EWM (entropy weight method) only relies on data dispersion and lacks industry experience guidance. The fusion of the two can achieve a balance between "data objectivity and experience subjectivity".

[0050] The evaluation index system strictly corresponds to the core demands of the industrial chain, and there are 3 first-level indicators (Table 4-1). The weight is determined by "expert team Delphi method" (5 industry experts + 3 park management cadres + 2 enterprise executives, 3 rounds of scoring convergence): Table 4-1. Definition of first-level indicators

[0051] 4.2 Analytic Hierarchy Process (AHP) - Subjective Weight Calculation 4.2.1 Model Construction AHP converts the qualitative judgment of experts on the importance of indicators into quantitative weights through "judgment matrix". The key steps are as follows: 1. Scale definition: Use 1-9 scale method (Table 4-2) to describe the relative importance between first-level indicators: Table 4-2. Definition of scale method

[0052] 2. Judgment matrix construction: The expert team scores the 3 first-level indicators to form a judgment matrix , where is the importance scale of relative to , satisfying , =1.

[0053] 3. Consistency check: Avoid logical contradictions in expert scoring (such as > and > but > ), steps are: Calculate the maximum eigenvalue of the judgment matrix ; Calculate the consistency index (n=3 is the number of indicators); Calculate the random consistency ratio (RI is the random consistency index, RI=0.58 for 3-order matrix, source: AHP standard coefficient table). If CR < 0.1, the matrix passes the consistency test, otherwise it needs to be re-scored.

[0054] 4. Subjective weight calculation: solve the normalized eigenvector of the judgment matrix by "eigenvector method", that is, the AHP subjective weight ( = 1).

[0055] 4.2.2 Model training and application Step 1: Expert scoring and judgment matrix construction For the "short board of new energy vehicle upstream supply chain connection" (long supply cycle), the final converged judgment matrix of the 10-person expert team is: ; = 3: The (production value influence degree) is slightly more important than (The stability influence degree); = 5: The (repair difficulty) is obviously more important than ; = 3: The (production value influence degree) is slightly more important than .

[0056] Step 2: Consistency test 1. Calculate the eigenvalue of the judgment matrix: through matrix operation ≈ 3.038; 2. Calculate = 0.019; 3. Calculate ≈ 0.033 < 0.1, pass the consistency test.

[0057] Step 3: Subjective weight calculation Solve the characteristic equation , get the normalized eigenvector: ; That is = 0.637 (production value influence degree weight), = 0.258 (stability influence degree weight), = 0.105 (repair difficulty weight), stored in the "short board evaluation weight library".

[0058] 4.3 AHP-EWM fusion algorithm - combined weight calculation 4.3.1 Fusion logic and formula definition The core of fusion is to measure the "credibility" of AHP subjective weight and S3EWM objective weight by "coefficient of variation": the smaller the dispersion (the more concentrated the weight, the higher the expert consensus or the lower the data dispersion), the higher the credibility, and the higher the fusion coefficient. The specific formula is as follows: 1. Coefficient of variation calculation: the coefficient of variation CV reflects the degree of data dispersion, the smaller the CV, the lower the dispersion, and the formula is: ; (4-1) is the standard deviation (such as = [0.637, 0.258, 0.105] of the new energy automobile short board) ≈ 0.23); is the mean (such as = 0.333 of the new energy automobile short board) = (0.637 + 0.258 + 0.105) / 3 ≈ 0.333).

[0059] 2. Fusion coefficient determination: the fusion coefficient (AHP weight ratio) and 1- (EWM weight ratio) are allocated in inverse proportion to CV, and the formula is: ; (4-2) The smaller the CV, the higher the weight ratio, such as AHP CV small large, more dependent on subjective experience.

[0060] 3. Combined weight calculation: the combined weight of link g, index X, and first-level index is: ; (4-3) Ensure = 1, which can be used for subsequent comprehensive impact score weighting.

[0061] 4.3.2 Fusion application Take "electronic information supporting service collaborative innovation short board" (long software iteration cycle) as an example, and strongly correlate S3 data: 1. Input data: S3EWM objective weight: = 0.4 (entropy weight of software iteration cycle index, high dispersion, weight ratio 40%); S4AHP subjective weight: = [0.637, 0.258, 0.105] (same as AHP weight of new energy automobile short board, same evaluation system); Calculate CV: ≈ 0.69; : 3 indicators (software iteration period, logistics accuracy rate, and detection efficiency) in the supporting service link, = [0.4, 0.3, 0.3], ≈ 0.058, ≈ 0.333, so ≈ 0.174.

[0062] 2. Fusion coefficient calculation (formula 4-2): , 1- ≈ 0.799; Since the CV of EWM is smaller, 1- is larger, and more dependent on the objective data weight.

[0063] 3. Combined weight calculation (formula 4-3): For (value impact degree): = 0.201 x 0.637 + 0.799 x 0.4 ≈ 0.128 + 0.320 = 0.448; For (stability impact degree): = 0.201 x 0.258 + 0.799 x 0.4 ≈ 0.052 + 0.320 = 0.372; For (repair difficulty): = 0.201 x 0.105 + 0.799 x 0.4 ≈ 0.021 + 0.320 = 0.341; The sum of the combined weights is 0.448 + 0.372 + 0.341 ≈ 1.161, which needs to be normalized to 1, and finally [0.386, 0.320, 0.294] is obtained.

[0064] 4.4. Comprehensive impact score calculation 4.4.1 Scoring rules and formulas 1. Criterion layer scoring (1-5 points, 5 points have the greatest impact / difficulty): For each short board, decompose the first-level indicators into quantifiable second-level indicators, and score based on the data (Table 4-3): Table 4-3. Scoring example

[0065] 2. Comprehensive impact score formula: weighted sum through combined weight, formula: ; (4-4) Where is the score of short board (g, X) in the first-level indicator , and ∈[1,5], the greater the value, the higher the priority.

[0066] 4.4.2 Evaluation (4 core short boards of 2nd industry chain) Enumeration 1: New energy vehicle upstream supply chain connection short board (long supply cycle) 1. Scoring basis: Production value impact: Long supply cycle leads to insufficient upstream material supply, 18% reduction in midstream capacity utilization, 18% reduction in production value → =4 points; Stability impact: 3 chain breaks in the past year due to delayed supply, total running time 12 months, chain break probability 3 / 12=25% → =4 points; Repair difficulty: Need to optimize supplier layout + establish safety stock, cycle 8 months, funds 30 million → =3 points.

[0067] 2. Combination weight: =[0.39,0.31,0.30] (after normalization); 3. Comprehensive impact score (formula 4-4): =0.39×4+0.31×4+0.30×3=1.56+1.24+0.9=3.7 (high priority: 3.0≤S<4.0).

[0068] Enumeration 2: Electronic information supporting service collaborative innovation short board (long software iteration cycle) 1. Scoring basis: Production value impact: Slow software iteration leads to downstream equipment unable to adapt to new chips, 12% reduction in production value → =3 points; Stability impact: 2 downstream shutdowns in the past year due to software adaptation problems, chain break probability 2 / 12≈16.7% → =3 points; Repair difficulty: Need to jointly develop with universities + introduce engineers, cycle 10 months, funds 40 million → =3 points.

[0069] 2. Combination weight: =[0.386,0.320,0.294]; 3. Comprehensive impact score: =0.386×3+0.320×3+0.294×3=3×(0.386+0.320+0.294)=3.0 (high priority threshold value).

[0070] Enumeration 3: New energy vehicle downstream comprehensive capacity short board (no single indicator exceeds threshold) 1. Scoring basis: Impact on output value: Low downstream assembly efficiency, 10% reduction in output value → = 2 points; Impact on stability: No chain breakage, only order delay, fluctuation probability 8% → = 2 points; Repair difficulty: Need to add production line, cycle 15 months, funds 80 million → = 4 points.

[0071] 2. Combined weight: = [0.4, 0.3, 0.3]; 3. Comprehensive impact score: = 0.4 x 2 + 0.3 x 2 + 0.3 x 4 = 0.8 + 0.6 + 1.2 = 2.6 (Medium priority: 2.0 ≤ S < 3.0).

[0072] Enumeration 4: Shortcomings in electronic information midstream comprehensive manufacturing efficiency (no single indicator exceeds threshold) 1. Scoring basis: Impact on output value: Low yield and equipment utilization, 15% reduction in output value → = 3 points; Impact on stability: Equipment failure leads to short-term shutdown, fluctuation probability 12% → = 3 points; Repair difficulty: Need to update equipment and optimize process, cycle 18 months, funds 120 million → = 5 points.

[0073] 2. Combined weight: = [0.4, 0.3, 0.3]; 3. Comprehensive impact score: = 0.4 x 3 + 0.3 x 3 + 0.3 x 5 = 1.2 + 0.9 + 1.5 = 3.6 (High priority).

[0074] 4.5 Priority ranking and S5 connection 4.5.1 Priority classification criteria According to From high to low, divide into 4 priorities (Table 4-4): Table 4-4. Priority classification criteria example

[0075] 4.5.2 Priority ranking results (Table 4-5) Table 4-5. Priority ranking results example

[0076] 4.5.3 Connection logic with S5 The generated park industry chain short board priority ranking table is transmitted to S5, which transmits two key information: 1. Comprehensive impact score : Directly determines the S5 target coefficient k (the higher the priority, the larger k and the more aggressive the target, such as very high priority k=0.8, high priority k=0.6); 2. Repair difficulty score : Determine the S5 target cycle (such as = 3 points → Repair period is 6-12 months, corresponding to the S5 target period set at 10 months, leaving a buffer time).

[0077] At the same time, the target setting of S5 needs to reversely match the output value impact of S4 :like =4 points (output value reduced by 15% to 20%), then the S5 target needs to ensure that the output value loss after repair is reduced to less than 5%, forming a logical closed loop of "impact assessment-target matching".

[0078] S5: Industry collaborative scheduling target setting (based on S4 priority, target and plan formulation process as follows Figure 3 shown) 5.1 Core Principles and Formulas for Goal Setting Goal setting must follow the three principles of "priority-oriented, data-supported, and quantifiable and implementable." The core formula is "current value + (benchmark value - current value) × target coefficient," i.e.: ; (5-1) : target value of the core indicator of link g and indicator X (original value, non-standardized); : The average original value of link g and indicator X in the original data of S1 (e.g., the average upstream supply cycle is 14 days); : The original value of the industry benchmark link g and indicator X (e.g., the upstream supply cycle benchmark is 7 days); : Target coefficient (Table 5-1, from S4 Determine that the higher the priority, the larger k is and the closer the target is to the benchmark); Table 5-1. Target coefficient examples

[0079] In addition to the core indicator targets, "auxiliary targets" (such as technical cooperation and talent recruitment) need to be added to ensure that the targets cover "hard indicators + soft support".

[0080] 5.2 Goal Setting 5.2.1 Shortage of upstream supply chain connection in new energy vehicle industry (high priority, S=3.7, k=0.6) 1. Core indicator: upstream supply cycle

[0081] S1 Raw data: = 14 days (average supply cycle of 10 upstream suppliers); Industry benchmark: = 7 days (supply cycle of leading new energy vehicle material supplier in China, source: new energy vehicle supply chain related materials, etc.); Target value calculation (formula 5-1): = 14 + (7-14) x 0.6 = 14-4.2 = 9.8 ≈ 10 days Target cycle: 3-6 months (match S4 repair difficulty score 3 points, leave 2 months buffer).

[0082] 2. Auxiliary targets: Supplier optimization: eliminate 2 suppliers with supply cycle over 15 days within 6 months, add 3 local suppliers (shorten transportation time); Safety stock: establish upstream material safety stock (meet 15 days production demand) within 3 months, stock funds subsidized by 30% by park (source: park industry support fund); Collaboration platform: launch "upstream supplier collaboration platform" within 4 months, realize real-time visualization of order / logistics / stock.

[0083] 3. Responsible subjects: park economic development bureau (lead), upstream supplier alliance (10 enterprises), park finance bureau (funding guarantee).

[0084] 5.3 Target summary and S6 connection Generate park industry collaboration and scheduling target table (Table 5-2), pass information to S6: 1. Core indicator target value : as the "optimization target" of S6 genetic algorithm (GA) (such as GA needs to optimize chip yield to 91%); 2. Auxiliary target content: as "non-quantitative measures" of S6 scheme (such as school-enterprise cooperation, equipment update), need to be supplemented based on GA optimization; 3. Responsible subjects and cycle: determine the "execution framework" of S6 scheme (such as the park needs to clearly specify the subsidy ratio, and the enterprise needs to clearly specify the responsibility division).

[0085] S6's solution formulation needs to strictly match S5's target: if S5's target is "91% chip yield within 6 months", S6's GA needs to optimize production parameters (such as temperature, pressure) to achieve this target, while supporting equipment updates and process training measures to ensure "algorithm optimization + manual measures" dual driving. Table 5-2: Table 5-2. Park industry coordination scheduling target representation example

[0086] S6: Industry coordination scheduling solution formulation (embedded GA) 6.1 Solution formulation framework and algorithm selection Solution formulation needs to follow the "one short board one solution, algorithm optimization + manual measures" framework, and select appropriate tools for different short board types: Production scheduling type short board (such as midstream manufacturing efficiency, downstream capacity): adopt genetic algorithm (GA) to optimize "order-equipment-personnel" matching to achieve optimal resource allocation; Supply chain type short board (such as upstream supply cycle): adopt "supplier optimization + coordination platform" manual measures, supplemented by GA optimization of safety stock level; Technical coordination type short board (such as supporting service software iteration): adopt "school-enterprise cooperation + talent introduction" manual measures, supplemented by GA optimization of R&D task allocation.

[0087] The reason for choosing GA to optimize production scheduling is that it is good at solving "multi-variable, multi-constraint" combination optimization problems (such as matching of 10 enterprises and 20 orders), and can converge to the optimal solution through iteration, with efficiency improved by more than 30% compared with traditional manual scheduling.

[0088] 6.2 Genetic algorithm (GA) optimization of midstream chip manufacturing scheduling (high priority short board) 6.2.1 Model construction (for "chip yield + equipment utilization" dual target) 1. Problem definition: 8 chip companies (i=1-8) receive 15 chip orders (j=1-15) per month, which need to be allocated to companies to meet: Constraint 1: The equipment capacity of enterprise i ≤ maximum capacity (such as enterprise 1 photolithography equipment maximum capacity 50,000 pieces / month); Constraint 2: Order j delivery period ≤ customer requirement (such as order 1 needs to be delivered within 30 days); Objective: Maximize chip yield (≥91%) + Maximize equipment utilization (≥84%).

[0089] 2. Chromosome coding: binary matrix coding , =1 indicates that enterprise i undertakes order j, = 0 means not to accept (coding length = 8 * 15 = 120 bits, each bit represents a "company-order" matching relationship).

[0090] 3. Fitness function (fusion of S4, S5 data): (6-1) : Fitness value (0 ≤ f(x) ≤ 1, the larger the value, the better the solution); : Expected yield of company i to accept order j (based on S1 historical data, such as company 1 producing order 1 with a yield of 92%); : Yield demand of order j (such as order 1 needs 10,000 pieces); : Maximum capacity of lithography equipment of company i (such as company 1 capacity 500,000 pieces / month); = 84% (S5 target value); = 0.5, = 0.5 (weight determined by S4 combination weight, yield and utilization are equally important).

[0091] 4. Genetic operators (ensure algorithm convergence): selection: roulette wheel selection (chromosomes with top 30% fitness values are directly reserved to the next generation); crossover: single-point crossover (randomly select the 50th bit as the crossover point, exchange the second half of the two parent chromosomes); mutation: bit mutation (randomly flip the values of two positions in the chromosome, mutation probability = 0.01, avoid local optimum).

[0092] 6.2.2 Model training 1. Input data: S1 historical data: capacity of company i , yield of order j , yield of company i producing order j ; S5 target value: = 91%, = 84%; Algorithm parameters: population size 100 (100 feasible solutions per iteration), iteration number 50 (converged after 50 generations), crossover probability = 0.8.

[0093] 2. Iterative training process: Generation 1: 100 chromosomes are randomly generated, and the fitness value is calculated, with the highest f(x) = 0.65 (yield 88%, utilization 78%); Generation 10: through selection, crossover, and mutation, the highest f(x) = 0.78 (yield 89.5%, utilization 81%); Generation 30: the highest f(x) = 0.92 (yield 90.5%, utilization 83%), close to the target; Generation 50: the highest f(x) = 0.98 (yield 91.2%, utilization 84.5%), meeting the S5 target, stopping iteration, and outputting the optimal chromosome .

[0094] 3. Optimal solution Example: ; Enterprise 1 undertakes orders 1-2, enterprise 2 undertakes order 3, …, enterprise 8 undertakes order 15, achieving a yield of 91.2% and a utilization rate of 84.5%.

[0095] 6.2.3 Model application (scheme landing measures) Based on the optimal solution of GA , an electronic information midstream chip manufacturing collaborative scheduling scheme is developed, including three types of measures: 1. Production scheduling optimization (GA-driven): order allocation: according to , the allocation of "enterprise 1→order 1-2, enterprise 2→order 3" is executed, and the allocation scheme is updated once a week (to respond to order changes); equipment scheduling: based on order demand, the running time of photolithography equipment is optimized (e.g., enterprise 1 equipment runs 6 days a week, 20 hours a day, with a utilization rate of 84.5%); quality control: for enterprises that undertake high-yield orders (e.g., order 5 requires a yield of 93%), increase the daily yield detection frequency (2 times / day).

[0096] 2. Equipment update (manual measures): purchase list: purchase 12 photolithography equipment within 6 months, allocated to 3 enterprises (enterprises 4-6) with a yield below 90%; fund arrangement: total fund 120 million, park subsidy 50% (60 million), enterprise self-funding 50%, subsidy funds are disbursed in 3 phases (30% for equipment arrival, 30% for installation and debugging, and 40% for acceptance); progress node: months 1-2 for bidding and procurement, months 3-4 for equipment arrival, months 5-6 for installation, debugging, and acceptance.

[0097] 3. Process optimization (manual measures): Technology introduction: Introduce a third-party process team in the third month, conduct 2-phase process training (3 days per phase, covering 40 technical personnel from 8 enterprises); Parameter adjustment: According to the GA optimization results, adjust the lithography temperature of Enterprise 1 from 85°C to 82°C (historical data shows that the yield is improved by 1.5% at 82°C); Effect verification: Compare the yield data before and after adjustment every month, and form a process optimization effect report.

[0098] 6.3 Other short board solutions 6.3.1 Upstream supply chain connection solution for new energy vehicles (high priority) 1. Supplier optimization measures: Elimination mechanism: Evaluate 10 suppliers in the first 2 months, eliminate 2 suppliers with a supply cycle of more than 15 days (Suppliers 9-10); New business introduction: Investigate 5 local suppliers (distance <200 km) in the third to fourth months, add 3 (Suppliers 11-13), sign a 3-year supply agreement; Assessment system: Establish a "supply cycle + quality + price" three-dimensional assessment (cycle accounts for 40%, quality accounts for 30%, and price accounts for 30%), score every month, and give the top 3 a 10% order tilt.

[0099] 2. Safety inventory optimization (GA assistance): GA optimization goal: Determine the safety inventory level (meet the 15-day production demand) and minimize inventory cost; Optimization results: 50 tons of lithium material inventory and 10 tons of cobalt material inventory, inventory cost 800,000 yuan per month, park subsidy 30% (24,000 yuan / month); Inventory management: Complete inventory construction in the fifth to sixth months, adopt the "first-in, first-out" principle, and check once a month.

[0100] 3. Collaborative platform construction: Function modules: Order management (online ordering / accepting orders), logistics tracking (real-time positioning), inventory warning (alarm when inventory is less than 10% of safety inventory); Online time: Complete platform development and testing by the end of the fourth month, put into trial operation in the fifth month, and formally operate in the sixth month; Operation responsibility: Led by the Economic Development Bureau of the park, entrust a third-party technology company (such as Alibaba Cloud) to be responsible for operation and maintenance, with an annual operation and maintenance cost of 2,000,000 yuan (borne by the park).

[0101] 6.3.2 Collaborative innovation solution for electronic information supporting services (high priority) 1. School-enterprise cooperation measures: Agreement signing: Sign a cooperation agreement with the Software College of XX University in the first 2 months, and clearly define the responsibilities of the joint laboratory (R&D + talent training); R&D tasks: Carry out the "chip adaptation software rapid iteration" R&D project in the third to sixth months, aiming to compress the iteration cycle from 6 weeks to 4 weeks; Funding investment: Total project funding is 8,000,000 yuan, with a park subsidy of 40% (3,200,000 yuan) and a 60% share (480,000 yuan) among 5 software enterprises.

[0102] 2. Talent introduction measures: Recruitment plan: 3-5 months to Beijing, Shanghai to carry out 2 special recruitment fairs, target to introduce 20 software engineers (including 5 architects); Subsidy policy: the park gives each person 50,000 yuan of talent subsidy (paid in 2 years, 25,000 yuan per year), the enterprise provides housing subsidy (1500 yuan / month for 3 consecutive years); Training integration: After the new employee enters the job, the XX University tutor carries out 1 month pre-job training (focuses on chip adaptation technology).

[0103] 3. Tool upgrade measures: Tool procurement: In the 4th month, 5 software companies are purchased automated testing tools (such as Selenium+Jmeter set for each company, cost 100,000 yuan); Subsidy ratio: total cost 500,000 yuan, park subsidy 40% (200,000 yuan), enterprise bears 60% (300,000 yuan); Training application: In the 5th month, tool use training is carried out to ensure that at least 5 employees of each enterprise are proficient in operation, and application effect is examined in the 6th month.

[0104] 6.3.3 New energy vehicle downstream comprehensive capacity scheme (medium priority) 1. Production line expansion measures: Planning and design: complete production line expansion planning (selecting site in the east of the park, occupying 50 mu) in the 1st-3rd month, capacity increase 20,000 vehicles / year; Construction period: construction in the 4th-10th month, equipment installation and commissioning in the 11th-12th month, acceptance at the end of the 12th month; Fund subsidy: land transfer fee 10,000,000 yuan, park subsidy 30% (300,000 yuan), construction fund 80,000,000 yuan is self-raised by the enterprise (3 enterprises share according to the proportion of capacity).

[0105] 2. Employee training measures: Training plan: carry out 1 period of assembly skill training for 5 days each in the 4th, 6th and 8th months, covering 200 employees (about 67 people for each enterprise); Training content: new production line equipment operation (such as robot welding), quality detection standard, safety specification; Cost bearing: total training cost 1,000,000 yuan, fully subsidized by the park (from the park skill training special fund).

[0106] 3. Order coordination measures: Platform construction: develop "downstream order sharing platform" in the 6th-8th month to realize order load visualization (such as automatically recommend to enterprise B when enterprise A order is saturated); Allocation rules: allocate according to enterprise capacity utilization rate (enterprises with utilization rate <70% are given priority to receive orders), adjust allocation ratio once a month; Effect target: at the end of the 12th month, the capacity utilization rate difference of the 3 enterprises is reduced from the current 20% to within 10%.

[0107] 6.4 Scheme review and S7 connection 6.4.1 Scheme review process All programs need to be reviewed by the office of the park management committee, the process is: 1. The main body of responsibility submits the initial draft of the program (such as the park industrial and information technology bureau submits the midstream chip program); 2. The preliminary examination of the policy research office (audit target matching, fund rationality); 3. Organize expert review (5 industry experts score, full score 100 points, ≥80 points pass the preliminary examination); 4. The director's office deliberation (participants: director of the management committee, deputy director in charge, head of each responsible party); 5. Release and implement (release within 5 working days after the audit is passed, and the responsible party starts to implement).

[0108] 6.4.2 Linkage logic with S7 The approved park industry coordination and scheduling implementation plan needs to be transmitted to S7, and the information needs to be transmitted: 1. GA optimal solution (such as midstream order allocation plan): as the "initial state" of S7 Kalman filter (KF) (such as initial equipment utilization 84.5%); 2. Key progress node (such as completing equipment bidding in the third month): as a "milestone event" monitored by S7, which needs to be closely tracked for progress deviation; 3. Monitoring indicators (such as yield, delivery cycle): as "observation variables" of S7 KF (such as monthly collected yield data ).

[0109] The monitoring data of S7 will real-time feedback the implementation effect of S6 program: if a node progress lags behind (such as equipment bidding is delayed for one month), S6 program needs to be adjusted (such as compressing subsequent installation and debugging time), forming a dynamic closed loop of "program-monitoring-adjustment".

[0110] S7: Real-time monitoring of program implementation and data (embedded KF) 7.1 Monitoring framework and algorithm selection Monitoring needs to achieve "real-time tracking progress, early warning deviation", and the core adopts Kalman filter (KF) algorithm: Advantages: KF is good at handling "dynamic systems with noise" (such as equipment failure, personnel change, etc. in the implementation of park program), which can output the optimal progress estimation value through "prediction-update" iteration, 7-10 days earlier than traditional manual statistical early warning; Application scenario: for linear tasks with clear progress targets (such as equipment update, order delivery), non-linear tasks (such as technology research and development) use "milestone monitoring + manual evaluation" combination.

[0111] The monitoring system is divided into three layers: 1. Data collection layer: real-time data is collected through sensors, system interfaces, and manual reporting (such as equipment running status, order delivery progress); 2. Algorithm processing layer: KF processes linear task data to output progress estimation value and deviation; 3. Early warning display layer: through the park "industrial coordination and scheduling platform", the progress is displayed, and when the deviation exceeds the threshold, the early warning (SMS + platform notification) is triggered.

[0112] 7.2 Kalman Filter (KF) Monitoring Midstream Equipment Update Progress (High Priority Shortcoming) 7.2.1 Model Construction (State and Observation Definition) 1. Problem definition: A midstream chip company needs to complete the upgrade of 12 lithography equipment within 6 months (total goal =12 units), monitor the monthly update progress and provide early warning of delay risks.

[0113] 2. State equation and observation equation: State equation (describing progress over time): ; (7-1) : The actual number of units updated at time t (state variable, t=1-6, unit: unit); A=1: state transition coefficient (progress is linearly cumulative, for example, 2 units are updated in January, and 2 more units are added in February); B=2: control coefficient (2 units are planned to be updated each month, 12 units / 6 months = 2 units / month); : Resource investment ratio at time t-1 (0≤u≤1, for example, u=1 means full investment, u=0.8 means 80% investment); : Process noise (equipment procurement delays, installation failures, etc., with mean 0 and covariance Q=0.1, estimated based on historical data).

[0114] Observation equation (describing the relationship between the observed value and the true state): ; (7-2) : The number of units updated at time t (actual data collected, e.g., 2 units have been updated by the end of January); H=1: Observation coefficient (the observed value directly reflects the actual number of units); : Observation noise (statistical errors, data entry errors, etc., with mean 0 and covariance R=0.05, estimated based on data collection accuracy).

[0115] 3. KF iteration formula (prediction-update two steps): Prediction step (predicting the current state based on the previous state): ; (7-3) ; (7-4) : Forecast progress at time t; : Forecast covariance, reflecting forecast uncertainty.

[0116] Update step (correct the prediction based on the observations to get the optimal estimate): ; (7-5) ; (7-6) ; (7-7) : Kalman gain, balance prediction and observation weight; : optimal estimation progress at time t; : updated covariance; I: identity matrix).

[0117] 7.2.2 Model training (initial parameter determination) 1. Initial state setting: t = 0 (before the start of the program): no updated equipment, so = 0 units; Initial covariance = 0.5: high initial uncertainty (there are more unknown risks in equipment procurement).

[0118] 2. Control input Setting: According to the progress of the program, the full amount of resources is invested every month, so = 1 (t = 1 uses = 1, t = 2 uses = 1, and so on).

[0119] 3. Noise covariance Q, R estimation: Q = 0.1: According to the past 3 equipment update projects in the park, the process noise standard deviation is about 0.3, so Q ≈ 0.1; R = 0.05: Data collection is responsible by the park industrial and information bureau, statistical error ≤ 0.2, so R = 0.04 ≈ 0.05.

[0120] 7.2.3 Model application (6-month monitoring process) Premise: S6 program plans to update 12 units of equipment in 6 months, plans to update 2 units per month ( = 2 × 1 = 2 units / month), warning threshold: actual progress ≤ 80% of planned progress (e.g. 1 month plan 2 units, warning threshold 1.6 units).

[0121] Table 7-1. Monitoring progress example

[0122]

[0123] Warning processing flow (take t = 2 as an example): 1. Trigger warning: end of February = 3.3 units > planned 4 units × 90% = 3.6 units, actual deviation rate -17.5%, trigger warning; 2. Cause analysis: The park industrial and information technology bureau and the enterprise jointly investigated and found that the supplier (ASML) had insufficient capacity and the equipment delivery was delayed for 15 days; 3. Rectification measures: Increase resource input in March (u=1.2), coordinate ASML to deliver in priority, and start bidding for backup suppliers (Nikon) at the same time; 4. Effect verification: Progress returned to normal in April ( =7.9 units, close to the planned 8 units), and the warning was lifted.

[0124] 7.3 Other short board monitoring 7.3.1 Monitoring of upstream supply chain connection of new energy vehicles (high priority) 1. Supply cycle monitoring (KF application): Target: Reduce the average supply cycle from 14 days to 10 days within 6 months, with a planned reduction of 0.67 days per month (4 days / 6 months); Monitoring data: Collect the supply cycle of 10 suppliers every month, calculate the average value (e.g. January =13.5 days, February =12.8 days); KF result: June =10.2 days (close to the target of 10 days), deviation rate 2%, monitoring normal.

[0125] 2. Supplier assessment monitoring (manual statistics): Each month, score according to "cycle 40% + quality 30% + price 30%", e.g. Supplier 1 scored 92 points in January (cycle 10 days, quality 98%, reasonable price), ranked first, and obtained 10% order tilt; Each quarter, the scores are summarized, and the suppliers who have ranked last for two consecutive times (e.g. Supplier 8 has ranked last for two consecutive times in March) are eliminated.

[0126] 3. Safety stock monitoring (threshold warning): Set safety stock threshold: 50 tons of lithium material (warning line 45 tons), 10 tons of cobalt material (warning line 9 tons); Collect stock data daily, e.g. on May 10, lithium material inventory is 44 tons < 45 tons, triggering warning, immediately start replenishment (within 2 days replenish to 50 tons).

[0127] 7.3.2 Monitoring of collaborative innovation of electronic information supporting services (high priority) 1. Software iteration cycle monitoring (milestone monitoring): Milestone nodes: Complete signing of school-enterprise cooperation agreement in the 3rd month, and reduce iteration cycle to 4 weeks in the 6th month; Monitoring results: signed the agreement on time in the 3rd month, actual iteration cycle of 4.2 weeks in the 6th month (close to the target of 4 weeks), deviation reason is new employee training delay, need to increase training frequency.

[0128] 2. Talent introduction monitoring (manual statistics): Target: introduce 20 software engineers (including 5 architects) in 6 months; Monitoring data: 8 people (including 2 architects) were introduced in the 3rd month, 15 people (including 4 architects) were introduced in the 5th month, and 21 people (including 5 architects) were introduced in the 6th month, exceeding the target.

[0129] 3. Tool application monitoring (effect evaluation): Monthly statistics of tool usage efficiency: after enterprise 1 uses automated testing tools, the test time is shortened from 8 hours / time to 3 hours / time, the efficiency is improved by 62.5%, and the expected effect is achieved.

[0130] 7.3.3 New energy vehicle downstream comprehensive capacity monitoring (medium priority) 1. Production line expansion monitoring (progress bar monitoring): Progress bar: complete planning and design in the 3rd month (3 / 12 months = 25%), complete construction in the 10th month (10 / 12 ≈ 83%), and complete acceptance in the 12th month (100%); Monitoring results: construction progress of 80% in the 10th month (lagging 3%), reason is rain season influence, need to extend construction time for 15 days, complete acceptance on time in the 12th month.

[0131] 2. Assembly efficiency monitoring (KF application): Target: increase assembly efficiency from 35 units / person / day to 41 units / person / day within 12 months, plan to increase 0.5 units per month; KF result: 40.8 units at the end of December (close to the target of 41 units), deviation rate -0.5%, monitoring is normal.

[0132] 3. Order coordination monitoring (platform data): Platform data: after the order sharing platform is put into operation in the 8th month, the capacity utilization rate difference of 3 enterprises is reduced from 20% to 8% (such as enterprise A utilization rate 85%, enterprise B 77%, enterprise C 82%), reaching the target.

[0133] 7.4 Monitoring report and S8 connection 7.4.1 Monitoring report output ​Output reports according to the "daily summary, weekly analysis, monthly evaluation" model: 1. Daily monitoring progress chart: automatically summarizes the real-time data of each short board (such as the number of equipment updates and the supply cycle) and sends it to the person in charge of the responsible entity; 2. Weekly monitoring and analysis meeting minutes: a regular meeting is held every Friday to analyze progress deviations (such as the reasons for delayed equipment updates) and formulate corrective measures; 3. Monthly monitoring and evaluation report: generated at the end of each month, including: KF optimal estimated progress of each short board , Observation Progress ; Deviation rate and early warning handling situation; Next month’s monitoring focus (e.g. the third month’s focus will be on monitoring the signing of the school-enterprise cooperation agreement).

[0134] 7.4.2 Connection Logic with S8 Monthly monitoring and evaluation reports are sent to S8, and the core delivers KF optimal estimation progress : 1. “Actual completion value” calculated as the S8 target achievement rate (e.g. midstream equipment update for 6 months =12 units, target 12 units, achievement rate 100%); 2. If There is a large deviation from the S5 target value (e.g., a software iteration cycle of 6 months =4.2 weeks > target 4 weeks), then S8 needs to analyze the reasons and adjust the S5 target or S6 plan; 3. Deviation reasons in the monitoring data (e.g., equipment delivery delays) will serve as the basis for S8 adjustment measures (e.g., adding backup suppliers), forming a closed loop of “monitoring-assessment-adjustment.”

[0135] S8: Program effectiveness evaluation and feedback adjustment (based on S7 monitoring) 8.1 Core Evaluation Indicators and Formulas The core of the evaluation is the "target achievement rate", which is the ratio of the actual completion value to the S5 target value. The formula is: ; (8-1) : Target achievement rate of link g and indicator X (0< ≤120%, if it exceeds 120%, it will be calculated as 120%); : The best estimated progress at the end of the S7 monitoring period (e.g. 6-month equipment update =12); : The target value at the end of the cycle set by S5 (such as the equipment update target of 12 units).

[0136] according to The evaluation results are divided into three categories (Table 8-1): Table 8-1. Example of target achievement rate evaluation results

[0137] 8.2 Evaluation Results 8.2.1 New Energy Vehicle Upstream Supply Chain Shortage (Cycle 6 months, Target Delivery Cycle 10 days) 1. Core Index Evaluation: S7 Monitoring Results: Average Delivery Cycle by the end of June = 10.2 days; S5 Target Value: = 10 days; Achievement Rate Calculation (Formula 8-1): (Up to standard); 2. Auxiliary Target Evaluation: Supplier Optimization: Eliminate 2, Add 3, 100% Achievement Rate; Safety Inventory: Establish 15-day inventory, 100% achievement rate; Collaboration Platform: On schedule, functional up to standard, 100% achievement rate.

[0138] 3. Adjustment Strategy: Continuous Optimization: Monthly evaluation of supplier scores, increase order proportion of top 3 suppliers from 10% to 15%; Expand Cooperation: Explore "VMI (Vendor Managed Inventory)" model with upstream suppliers, further shorten response cycle.

[0139] 8.3 Overall Evaluation Report and Adjustment Mechanism 8.3.1 Evaluation Results Summary (Table 8-2) Table 8-2. Evaluation Results Summary Example

[0140] 8.3.2 Feedback Adjustment Mechanism (Three-level Response) 1. Immediate Adjustment (Triggered when deviation rate > 5%): Responsible Subject: Submit deviation reason analysis within 24 hours (e.g. equipment update lag, ASML capacity shortage); Executive Department: Develop remedial measures within 48 hours (e.g. activate backup supplier Nikon); Case: Electronic information midstream equipment update deviation 17.5% in February, immediately increase 20% resource input in March, restore progress in April.

[0141] 2. Periodic Adjustment (After monthly evaluation): For shortfalls that meet the target: Increase target value by 5% to 10% (e.g. software iteration cycle from 4 weeks to 3.5 weeks); For basic shortfalls that meet the target: Strengthen auxiliary targets that do not meet the target (e.g. downstream employee retraining); Output monthly adjustment list for record by the park administrative committee.

[0142] 3. Annual Optimization (Every December): Recalculate entropy weight of S3 shortfalls identification based on annual data (e.g. "software iteration cycle" weight reduced from 0.4 to 0.35 due to reduced dispersion); Update industry benchmark value (e.g. chip yield industry benchmark increased from 95% to 96% due to technological progress); Revise S5 targets and S6 plans for the next year to form an "annual closed loop".

[0143] The technical solution can be popularized to similar industrial parks across the country, and provides quantifiable and landable methodological support for "complementing short boards and forging long boards" of the industrial chain.

[0144] The industrial collaborative scheduling system based on the short board analysis of the park industrial chain applies the industrial collaborative scheduling method based on the short board analysis of the park industrial chain, as shown in Figure 4 The industrial collaborative scheduling system based on the short board analysis of the park industrial chain comprises: A collection module is configured to collect park industrial chain original data to form a park industrial chain original data set; A preprocessing module is configured to preprocess the park industrial chain original data set to form a park industrial chain standardized data set and average standardized values of each industrial chain link; An identification module is configured to, based on the park industrial chain standardized data set and the average standardized values of each industrial chain link, in combination with industry benchmark data, calculate objective weights of each index in each industrial chain link by using an entropy weight method, and then calculate weighted gap rates of each index and comprehensive gap rates of each link by using the objective weights, the average standardized values, and the industry benchmark data, to identify single-link short boards and cross-link bottlenecks, and form a park industrial chain short board list; An ordering module is configured to, based on the objective weights in the park industrial chain short board list, in combination with subjective weights of each evaluation index determined by an analytic hierarchy process, calculate a fusion coefficient of the analytic hierarchy process and the entropy weight method by using a coefficient of variation method, and then fuse the subjective weights and the objective weights to obtain combined weights by using the fusion coefficient, to calculate comprehensive influence scores of the single-link short boards in combination with scores of the single-link short boards in value-added influence degree, stability influence degree, and repair difficulty dimensions, and form a park industrial chain short board priority ranking table; A setting module is configured to, based on the comprehensive influence scores in the park industrial chain short board priority ranking table, set core index target values, target achievement periods, and auxiliary targets corresponding to the single-link short boards, and form a park industrial collaborative scheduling target table; An optimization module is configured to, based on the core index target values and the auxiliary targets in the park industrial collaborative scheduling target table, in combination with the park industrial chain original data set, optimize production task allocation, supplier selection, or technology transformation project ordering by using a genetic algorithm, design specific solutions to the single-link short boards, and form a park industrial collaborative scheduling implementation scheme.

Claims

1. The industrial collaborative scheduling method based on the analysis of the shortcomings of the industrial chain of the park is characterized by: include: S1: Collect the original data of the industrial chain of the park to form the original data set of the industrial chain of the park; S2: Preprocess the original data set of the industrial chain of the park to form a standardized data set of the industrial chain of the park and the average standardized value of each industrial chain link; S3: Based on the standardized data set of the industrial chain of the park and the average standardized value of each industrial chain link, combined with industry benchmark data, the entropy weight method is used to calculate the objective weight of each indicator in each industrial chain link. Then, the weighted gap rate of each indicator and the comprehensive gap rate of each link are calculated by combining the objective weight, the average standardized value and the industry benchmark data. The shortcomings of single links and cross-link bottlenecks are identified to form a list of shortcomings of the industrial chain of the park; S4: Based on the objective weights in the list of shortcomings in the industrial chain of the park, combined with the subjective weights of each evaluation indicator determined by the hierarchical analysis method, the coefficient of variation method is used to calculate the fusion coefficient of the hierarchical analysis method and the entropy weight method. The subjective weight and the objective weight are integrated through the fusion coefficient to obtain the combined weight. Then, combined with the scores of the shortcomings in the output value impact, stability impact and repair difficulty dimensions of the single link shortcomings, the comprehensive impact score of the single link shortcomings is calculated to form a priority ranking table of the shortcomings in the industrial chain of the park; S5: Based on the comprehensive impact scores in the priority ranking table of the park's industrial chain shortcomings, set the core indicator target values, target achievement cycles, and auxiliary targets corresponding to the shortcomings of each link to form a park industry collaborative scheduling target table; S6: Based on the core indicator target values ​​and auxiliary targets in the park's industrial collaborative scheduling target table, combined with the original data set of the park's industrial chain, a genetic algorithm is used to optimize the production task allocation, supplier selection or technology transformation project sorting, and specific solutions to the shortcomings of single links are designed to form an implementation plan for the park's industrial collaborative scheduling.

2. The industrial collaborative scheduling method based on park industrial chain short board analysis according to claim 1 is characterized in that: It also includes S7: Based on the genetic algorithm optimization results in the industrial collaborative scheduling implementation plan of the park as the initial state, the Kalman filter algorithm is used to make real-time predictions on the implementation progress of the plan. Combined with the real-time collected plan execution data as the observation value, the progress prediction results are dynamically corrected. If the progress lag reaches the preset warning threshold, an early warning is triggered to form a monthly monitoring and evaluation report.

3. The industrial collaborative scheduling method based on park industrial chain short board analysis according to claim 2 is characterized in that: It also includes S8: Based on the progress forecast results in the monthly monitoring and evaluation report and the core indicator target values ​​in the park industry collaborative scheduling target table, calculate the target achievement rate of the shortcomings of a single link, conduct attribution analysis on the achievement from the dimensions of measure effectiveness, resource input, external environment and data quality, adjust the park industry collaborative scheduling target table or the park industry collaborative scheduling implementation plan according to the analysis results, and feed back to S5 or S6, forming a closed loop of data collection-preprocessing-shortcoming identification-impact assessment-goal setting-plan formulation-execution monitoring-effect evaluation-adjustment and optimization.

4. The industrial collaborative scheduling method based on park industrial chain short board analysis according to claim 1 is characterized in that: In step S3, the entropy weight method interacts with industry benchmark data. The process is: first, the objective weight of each indicator is calculated by the entropy weight method to highlight the importance of indicators with high data discreteness, and then the objective weight is substituted into the indicator weighted gap rate calculation formula, so that indicators with higher weights have a greater impact on the identification of shortcomings, ensuring that the identified single-link shortcomings and cross-link bottlenecks are more in line with the core needs of the industrial chain, and the information of each short board in the short board list is directly used as the basic data for calculating the comprehensive impact score in step S4.

5. The industrial collaborative scheduling method based on park industrial chain short board analysis according to claim 1 is characterized in that: In step S4, the analytic hierarchy process and the entropy weight method are integrated, specifically: the coefficient of variation of the subjective weight of the analytic hierarchy process and the coefficient of variation of the objective weight of the entropy weight method are calculated by the coefficient of variation method, and the inverse ratio of the coefficient of variation is used as the fusion coefficient, so that the weights with smaller discreteness and higher credibility account for a higher proportion during fusion, and the combined weights are directly used to calculate the comprehensive impact score of each short board, and the comprehensive impact score is used as the basis for setting the target coefficient in step S5.

6. The industrial collaborative scheduling method based on park industrial chain short board analysis according to claim 3 is characterized in that: In step S6, the interaction process between the genetic algorithm and the target table is: the core indicator target value in the target table is used as the fitness function constraint condition of the genetic algorithm. During the iterative process of the genetic algorithm, the solution that meets the target value requirements and has optimal resource consumption is screened out through selection, crossover, and mutation operators. The optimization result of the genetic algorithm is not only used as the initial state of the Kalman filter in step S7, but also as the execution basis for production task allocation or supplier selection in the implementation plan.

7. The industrial collaborative scheduling method based on park industrial chain short board analysis according to claim 6 is characterized in that: In step S7, the interactive process of the Kalman filter's progress prediction and early warning mechanism is as follows: the Kalman filter's state equation sets the state transfer matrix and control matrix based on the planned progress in the implementation plan, and the observation equation sets the observation matrix based on the execution data collected in real time. After each iteration, the Kalman gain is used to balance the predicted value and the observed value to obtain the optimal progress estimate. When the deviation between the optimal progress estimate and the planned progress in the target table exceeds 10%, a yellow warning is triggered, and when it exceeds 20%, a red warning is triggered. The warning information is fed back to step S8 in real time for attribution analysis.

8. The industrial collaborative scheduling method based on park industrial chain short board analysis according to claim 7 is characterized in that: In the closed loop formed by step S8, the output data of each link is used as the input data of the corresponding link in the next round. Specifically: the target table adjusted by S8 is fed back to S5 as the basis for updating the next round of target setting; the implementation plan adjusted by S8 is fed back to S6 as the basis for optimizing the next round of plans; the attribution analysis results of S8 are also fed back to S3 to update the gap rate calculation parameters for short-board identification, ensuring that the closed loop continues to adapt to changes in the industrial chain of the park.

9. The industrial collaborative scheduling method based on park industrial chain short board analysis according to claim 8 is characterized in that: In step S8, the end condition of the closed loop is: within two consecutive target achievement cycles, the target achievement rate of all extremely high priority and high priority shortcomings is ≥95%, and the target achievement rate of medium and low priority shortcomings is ≥80%, or the expert committee evaluates and confirms that the core shortcomings of the park's industrial chain have been eliminated. At this time, the next round of full process cycle can be entered, otherwise the closed loop optimization will continue to be promoted by adjusting the goals or plans.

10. The industrial collaborative scheduling system based on the analysis of the shortcomings of the industrial chain of the park is characterized by: The industrial collaborative scheduling method based on the analysis of the shortcomings of the industrial chain of a park according to any one of claims 1 to 9 is applied, and the industrial collaborative scheduling system based on the analysis of the shortcomings of the industrial chain of a park comprises: The collection module is used to collect the original data of the industrial chain of the park and form the original data set of the industrial chain of the park; The preprocessing module is used to preprocess the original data set of the industrial chain of the park to form a standardized data set of the industrial chain of the park and the average standardized value of each industrial chain link; The identification module is used to calculate the objective weight of each indicator in each industrial chain link based on the standardized data set of the industrial chain of the park and the average standardized value of each industrial chain link, combined with industry benchmark data, using the entropy weight method. Then, the weighted gap rate of each indicator and the comprehensive gap rate of each link are calculated by the objective weight, average standardized value and industry benchmark data, to identify the shortcomings of single links and cross-link bottlenecks, and form a list of shortcomings of the industrial chain of the park; The ranking module is used to calculate the fusion coefficient of the analytic hierarchy process and the entropy weight method based on the objective weights in the list of shortcomings in the industrial chain of the park, combined with the subjective weights of each evaluation indicator determined by the hierarchical analysis method. The subjective weight and the objective weight are integrated through the fusion coefficient to obtain the combined weight. The combined weight is then combined with the scores of the shortcomings in the output value impact, stability impact and repair difficulty dimensions of the single link shortcomings to calculate the comprehensive impact score of the single link shortcomings and form a priority ranking table for the shortcomings of the industrial chain of the park; The setting module is used to set the core indicator target value, target achievement cycle and auxiliary targets corresponding to the shortcomings of a single link based on the comprehensive impact score in the priority ranking table of the park's industrial chain shortcomings, thereby forming a park industry collaborative scheduling target table; The optimization module is used to optimize the production task allocation, supplier selection or technology transformation project sorting based on the core indicator target values ​​and auxiliary targets in the park's industrial collaborative scheduling target table, combined with the original data set of the park's industrial chain, using genetic algorithms, design specific solutions to the shortcomings of single links, and form an implementation plan for the park's industrial collaborative scheduling.

Images