A pilot-based cost reduction and efficiency improvement optimization method

Abstract

The present application relates to the technical field of intelligent optimization and pilot project process management, and particularly relates to a pilot cost reduction and efficiency improvement based optimization method, comprising: collecting historical operation data and business constraint conditions of a pilot project, preprocessing to construct a sample data set; constructing an optimization model with cost reduction and efficiency improvement as the target based on the data set and the constraint conditions; solving the optimization model by using an improved particle swarm optimization algorithm, obtaining an optimal parameter group satisfying the constraint conditions, and issuing it to a management and control system for implementation, collecting implementation effects within 7 days; evaluating cost reduction and efficiency improvement indexes according to the collected data, if both indexes reach the set threshold, determining a promotion scheme, if the indexes do not reach the threshold, backtracking and correcting a penalty coefficient in the model and re-solving until the threshold is reached. The present application realizes optimization of parameter configuration, and improves promotion efficiency and resource utilization rate of the pilot project.

Description

Technical Field

[0001] This invention relates to the technical field of intelligent optimization and pilot project process management, and in particular to an optimization method based on pilot for cost reduction and efficiency improvement. Background Technology

[0002] In new product development, process optimization, or workflow management, enterprises commonly use Pilot projects as an important verification mechanism before transformation. The Pilot model verifies the feasibility and profitability of technical solutions in a small-scale, low-risk manner, within a limited period and with controlled resources, thereby guiding subsequent large-scale promotion. With the development of big data analysis, automated control, and intelligent optimization technologies, system optimization based on the operational data of Pilot projects is gradually becoming a key means to promote cost reduction and efficiency improvement for enterprises. Especially in industries such as manufacturing, energy, construction, and environmental protection, the correlation between parameter settings, resource input, and operational results in Pilot projects is becoming increasingly strong, placing higher demands on the structured collection, modeling, and optimization analysis of process data.

[0003] CN117236680A discloses a beam factory management optimization system based on particle swarm optimization. By solving the beam manufacturing task plan and optimizing the production schedule, it effectively improves production scheduling efficiency and reduces process time and cost. However, this scheme is still mainly based on static scheduling and does not involve the data collection and iterative optimization mechanism in the pilot stage. CN110032755A discloses a multi-objective optimization method for urban sewage treatment process under multiple working conditions. It designs a multi-objective particle swarm optimization algorithm and introduces a case library to improve the algorithm's adaptability. However, its optimization objectives are only for water quality and energy consumption, and it lacks a clear periodic evaluation and promotion judgment mechanism. These existing methods usually have the following shortcomings: the optimization process is a single execution and a closed-loop feedback is not constructed; the promotion conditions are unclear and there is a lack of quantifiable indicator judgment basis; after optimization failure, there is a lack of model adaptive adjustment capability and a model update mechanism based on data feedback cannot be formed.

[0004] Therefore, existing Pilot-based optimization schemes generally suffer from the following technical bottlenecks: they cannot achieve quantitative judgment and iterative feedback mechanisms for operational effects, the optimization process lacks systematicity and stability, and it is difficult to support large-scale promotion decisions. Our invention proposes an optimization method based on pilot cost reduction and efficiency improvement. By constructing an optimization model oriented towards cost reduction and efficiency improvement indicators, combining historical data and business constraints, and using an improved particle swarm optimization algorithm to obtain the optimal parameter set, it is applied to the Pilot management system. The system collects operational data after implementation within a 7-day cycle and makes promotion judgments based on dual threshold indicators. If the promotion conditions are not met, the system automatically backtracks and corrects the penalty coefficient in the model for re-optimization, thereby realizing a complete closed-loop process of collection, modeling, solving, implementation, evaluation, and re-optimization. Summary of the Invention

[0005] The purpose of this section is to outline some aspects of the embodiments of the present invention and to briefly introduce some preferred embodiments. Some simplifications or omissions may be made in this section, as well as in the abstract and title of the present application, to avoid obscuring the purpose of this section, the abstract and title of the invention. Such simplifications or omissions shall not be used to limit the scope of the present invention.

[0006] In view of the aforementioned existing problems, the present invention is proposed.

[0007] To solve the above-mentioned technical problems, the present invention provides the following technical solution: collecting historical operation data and business constraints of the pilot project, and preprocessing the historical operation data to obtain a sample dataset;

[0008] An optimization model is constructed based on the sample dataset and the business constraints, with the goal of optimizing cost reduction and efficiency improvement indicators.

[0009] The optimization model is solved using an improved particle swarm optimization algorithm to obtain a parameter set that satisfies the business constraints.

[0010] The parameter set is sent to the pilot project's management system in the form of configuration information for operation, adaptively adjusting the pilot project's process parameters, and collecting post-implementation operation data within a preset period of 7 days.

[0011] Based on the operational data after implementation, the cost reduction and efficiency improvement indicators of the parameter group are calculated and compared with the set first and second thresholds:

[0012] When the cost reduction indicator is greater than or equal to the first threshold and the efficiency improvement indicator is greater than or equal to the second threshold, the parameter group is determined as the promotion plan;

[0013] When the cost reduction index is less than the first threshold, or the efficiency improvement index is less than the second threshold, the process returns to the optimization model solution step. By correcting the penalty coefficient in the optimization model, the solution is re-solved to obtain a new set of parameters, until the promotion plan is determined.

[0014] As a preferred embodiment of the pilot-based cost reduction and efficiency improvement optimization method described in this invention, the historical operating data includes the set process parameters, actual operating parameters, unit product energy consumption, unit product cost, equipment operating time, equipment utilization rate, production cycle time, yield rate, and number of defective products of the target process in the historical operating stage.

[0015] The business constraints include at least the upper limit of delivery cycle, upper limit of budget, upper limit of product quality consistency, upper limit of equipment load, upper limit of personnel working hours, process compliance threshold and emission limit.

[0016] As a preferred embodiment of the pilot-based cost reduction and efficiency improvement optimization method described in this invention, the historical operating data is preprocessed to obtain a sample dataset, including:

[0017] Read the historical operation data, use an interpolation algorithm to fill in missing items, and remove outliers by three times the standard deviation;

[0018] The completed numerical data is subjected to minimum and maximum standardization, which linearly maps all parameter values ​​to the interval [0,1].

[0019] The data is segmented using a sliding method with a time window length of 7 days, and each segment forms a record unit;

[0020] For each record unit, calculate the average process parameter value, average energy consumption value, average output cycle time, average yield rate, and average equipment utilization rate;

[0021] Each record unit and its corresponding business constraints are combined to form a sample entry;

[0022] All sample entries are merged to generate the sample dataset.

[0023] As a preferred embodiment of the pilot-based cost reduction and efficiency improvement optimization method described in this invention, an optimization model is constructed based on the sample dataset and the business constraints, with the goal of optimizing the cost reduction and efficiency improvement indicators, including:

[0024] The decision variables are set as adjustable process parameters in the sample dataset;

[0025] The first objective function is defined as minimizing unit cost, and the second objective function is defined as maximizing both production cycle time and yield rate.

[0026] The upper limit of delivery cycle, upper limit of budget, upper limit of equipment load, process compliance threshold and emission limit recorded in the business constraints are used as model constraints.

[0027] The first objective function and the second objective function are combined into a weighted multi-objective function, and the model constraints are embedded therein;

[0028] Establish a mapping relationship, where the input is a set of process parameters and the output is a combination of cost and efficiency indicators;

[0029] Generate an optimization model that includes the objective function, constraint functions, and input-output relationships.

[0030] As a preferred embodiment of the pilot-based cost reduction and efficiency improvement optimization method described in this invention, the optimization model is solved using an improved particle swarm optimization algorithm to obtain a parameter set that satisfies the business constraints, including:

[0031] Set the particle swarm size, maximum number of iterations, initial value of inertia weight, and learning factor;

[0032] Randomly initialize the parameter vector for each particle within the parameter feasible region;

[0033] Calculate the fitness value for each particle, where the fitness function is a weighted sum of the objective function and constraint penalty terms;

[0034] The individual optimal solution and the global optimal solution are determined based on the fitness values.

[0035] Update the parameter vectors of all particles according to the velocity and position update formula;

[0036] When the maximum number of iterations is reached, the current global optimal solution is output as the parameter set.

[0037] As a preferred embodiment of the pilot-based cost reduction and efficiency improvement optimization method described in this invention, the parameter set specifically includes:

[0038] Read all process parameter values ​​from the parameter group;

[0039] Compare the values ​​of each process parameter with the corresponding parameters in the current pilot project process database to determine the parameter items that need to be adjusted;

[0040] For each parameter that needs adjustment, determine the target setting value, adjustment range, adjustment step size, execution order, and execution node number;

[0041] The parameters that need to be adjusted are encapsulated into an optimization plan file. Each parameter in the file contains a parameter identifier, target value, adjustment range, adjustment step size, and execution node number.

[0042] As a preferred embodiment of the pilot-based cost reduction and efficiency improvement optimization method described in this invention, adaptive adjustment of the process parameters of the pilot project includes:

[0043] The optimization scheme file is converted into a structured instruction set, in key-value pair format;

[0044] The instruction set is uploaded to the configuration receiving module of the management and control system via the application interface;

[0045] The configuration receiving module parses the contents of the instruction set and generates a control task sequence;

[0046] The execution module modifies the process parameters one by one to the corresponding target values ​​according to the control task sequence;

[0047] Then record the modified parameter status to the runtime log file.

[0048] As a preferred embodiment of the pilot-based cost reduction and efficiency improvement optimization method described in this invention, operational data after implementation is collected within a preset period of 7 days. The operational data after implementation specifically includes:

[0049] Automatically extract the operation records of each production cycle after the cycle ends;

[0050] Data collected includes actual process parameters, unit product cost, total output, yield rate, average production cycle time, resource utilization rate, total energy consumption, and number of alarm events.

[0051] The collected data from the seven cycles were merged in chronological order to generate a complete post-implementation operational data table;

[0052] Perform standardization processing on the post-implementation running data table, consistent with the sample dataset, and output a standardized running dataset.

[0053] As a preferred embodiment of the pilot-based cost reduction and efficiency improvement optimization method described in this invention, the cost reduction and efficiency improvement indicators of the parameter group are calculated based on the operational data after implementation, including:

[0054] Read the unit cost data before and after implementation from the standardized operation dataset, and calculate the cost reduction index C. The calculation formula is as follows:

[0055]

[0056] in, The unit product cost before implementation. This refers to the unit product cost after implementation.

[0057] Read the average production cycle time, yield rate, and resource utilization rate data before and after implementation from the standardized operation dataset, and calculate the efficiency improvement index E. The calculation formula is as follows:

[0058]

[0059] in, and These represent the average production cycle time before and after implementation. and The yield rates before and after implementation are respectively. and These represent resource utilization rates before and after implementation. , , These are the weighting coefficients for average production cycle time, yield rate, and resource utilization rate, respectively.

[0060] The cost reduction index C and efficiency improvement index E are output as input results for threshold comparison.

[0061] As a preferred embodiment of the pilot-based cost reduction and efficiency improvement optimization method described in this invention, the method for setting the first threshold and the second threshold includes:

[0062] Retrieve statistical data on cost reduction and efficiency improvement indicators from at least 10 similar pilot projects from the historical project database;

[0063] Calculate the average of historical cost reduction indicators and multiply it by a correction factor of 1.05 to obtain the first threshold.

[0064] Calculate the upper quartile of the historical efficiency improvement index and multiply it by a correction factor of 1 to obtain the second threshold;

[0065] The first threshold and the second threshold are recorded in the threshold configuration table as fixed inputs to the threshold comparison module.

[0066] The beneficial effects of this invention are as follows: This invention not only realizes the transformation from empirical data to optimization models, but also ensures the efficiency, controllability and promotion value of the optimization results by introducing an improved swarm intelligence algorithm and quantitative judgment mechanism. At the same time, the system's adaptive ability in actual working condition fluctuations is further enhanced by the penalty coefficient callback, which significantly improves the stability and cost-effectiveness of the parameter configuration scheme in the actual implementation process. Attached Figure Description

[0067] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort. Wherein:

[0068] Figure 1 This is a flowchart illustrating the cost reduction and efficiency improvement optimization method based on pilot as shown in this invention. Detailed Implementation

[0069] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments.

[0070] Based on the embodiments of this invention, all other embodiments obtained by those skilled in the art without inventive effort should fall within the scope of protection of this invention.

[0071] Many specific details are set forth in the following description in order to provide a full understanding of the invention. However, the invention may also be practiced in other ways different from those described herein, and those skilled in the art can make similar extensions without departing from the spirit of the invention. Therefore, the invention is not limited to the specific embodiments disclosed below.

[0072] According to an embodiment of the present invention, in combination Figure 1 The flowchart shown illustrates a cost reduction and efficiency improvement method based on pilot technology, which specifically includes the following steps:

[0073] S1. Collect historical operational data and business constraints of the pilot project, and preprocess the historical operational data to obtain a sample dataset. Note the following in this step:

[0074] The historical operation data files of the pilot project are automatically read from the enterprise's production execution system. To ensure the integrity and time consistency of the data, database query commands are used for batch export. The data fields include: set process parameters (such as temperature, pressure, speed, and flow rate), actual operating parameters (actual values ​​collected in real time by sensors), unit product energy consumption and unit product cost, equipment running time and equipment utilization rate, production cycle time, yield rate and number of defective products.

[0075] Next, read the business constraint file from the project management system, and read the various constraint indicators in the project configuration table through the configuration interface. Extract parameters including: upper limit of delivery cycle (e.g., 10 days), upper limit of budget (e.g., 50,000 yuan), upper limit of product quality consistency (e.g., 5%), upper limit of equipment load (e.g., 85%), upper limit of personnel working hours (e.g., 40 hours), process compliance threshold (e.g., 95%), and emission limit (e.g., 3.5 kg CO2 / piece).

[0076] The two types of data are encoded and standardized according to a unified field format to form a structured input dataset. Each record is uniquely identified by its index key using a timestamp and process number. The storage structure adopts a two-dimensional table format. This structured input dataset serves as the input for subsequent data preprocessing and provides a unified data interface for model training and optimization.

[0077] It should be noted that by establishing a data interface between the production execution system and the project management system, the automated extraction and synchronous formatting of runtime data and constraints were achieved, avoiding delays and errors caused by manual processing. This ensured the integrity and consistency of the input data for subsequent optimization modeling. This process achieved the standardized integration of data standards, significantly improving the data applicability and calculation accuracy in the model building phase.

[0078] In a preferred embodiment, preprocessing is performed on the extracted historical running data, specifically including the following steps:

[0079] For data with missing items, a linear interpolation algorithm is used to complete the numerical values; for example, assuming two consecutive time points... , The observed values ​​are respectively , If in the middle of the time If missing, use:

[0080]

[0081] Outliers were removed using the three-standard-deviation method, i.e., if a parameter value... satisfy If it is an abnormal item, it will be removed. The mean, Standard deviation;

[0082] Minimum and maximum normalization are performed on all numerical parameters to linearly map the data to the [0,1] interval. This normalization process eliminates scale differences between different units and ensures the comparability of subsequent multi-objective optimization calculations.

[0083] The data is then segmented using a sliding window with a time window length of 7 days. Continuous data within each time period is divided into a record unit to reflect short-term operational characteristics. For each record unit, the following average characteristics are calculated:

[0084]

[0085]

[0086]

[0087]

[0088]

[0089] Where n represents the number of data entries within the current time window, and i represents the i-th data record. For the process parameter value of the i-th record, This is the average value of all process parameter values ​​within the current time window. Let i be the unit product energy consumption of the i-th record. This represents the average energy consumption level within the current time window. Let i be the production cycle time of the i-th record. This represents the average production cycle time within the current time window. Let be the yield rate of the i-th record. This represents the average yield rate within the current time window. Let i be the equipment utilization rate of the i-th record. This represents the average equipment utilization rate within the current time window.

[0090] For example, if the pilot project runs for 21 days from March 1st to March 21st, 2024, then it will slide with a 7-day window to form three record units:

[0091] Record Unit 1 (March 1st to March 7th);

[0092] Record Unit 2 (March 8th - March 14th);

[0093] Record Unit 3 (March 15th - March 21st);

[0094] Each record stores the average process characteristics and statistical indicators such as energy consumption, cycle time, yield rate, and equipment utilization rate for the corresponding time period.

[0095] It should be noted that the preprocessing method combining interpolation completion and anomaly removal ensures the continuity and reliability of the samples. Normalization enhances the comparability of different parameters, and sliding window segmentation makes the sample data temporally representative and dynamically sensitive, which is beneficial for the subsequent optimization model to capture changes in operating trends and improve the optimization model's ability to respond to periodic fluctuations.

[0096] In a preferred embodiment, each record unit is further combined with its corresponding time period's business constraints to form independent sample entries. The combination method is as follows: using the feature vector of the record unit... With business constraint vector Concatenate them into a composite sample vector:

[0097]

[0098] in, This represents a vector concatenation operation, where sample entries s store the mapping relationship between process characteristics and constraint parameters for that time period;

[0099] The sample entries generated from all time periods are merged in batches to finally generate the sample dataset. ,in The number of sample entries is [number]. The dataset is stored in CSV format and used as input for optimizing model training and validation.

[0100] Preferably, this sample item combination mechanism integrates process variables and management constraints to form a multi-dimensional sample that reflects the coupling relationship between production efficiency, energy consumption cost and process constraints. Through a unified structure storage format, it facilitates rapid indexing, matrix input and objective function solving of subsequent algorithm models, thereby improving the overall efficiency of data control and optimization calculation.

[0101] S2. Construct an optimization model based on the sample dataset and business constraints, aiming to achieve the optimal cost reduction and efficiency improvement indicators. Note the following in this step:

[0102] The decision variables (i.e., parameters that can be adjusted to affect the operating cost and production efficiency of the pilot project) are set as adjustable process parameters in the sample dataset;

[0103] It should be noted that the range of values ​​for each decision variable is jointly limited by historical operating data and business constraints. For example, the upper limit of temperature is determined by equipment safety constraints, and the lower limit of flow rate is limited by emission standards. This decision variable constitutes the input vector of the optimization model. After normalization, all variables are mapped to the interval [0,1]. By selecting and defining decision variables from the sample dataset, a controllable search area can be achieved in the multidimensional parameter space, ensuring that the subsequent algorithm solution is carried out under the premise of meeting the process boundary conditions, thereby improving the stability and solution efficiency of the model calculation.

[0104] The first objective function is defined as minimizing unit cost, and the second objective function is defined as maximizing both production cycle time and yield rate.

[0105] For example, the mathematical expression of the first objective function is as follows:

[0106]

[0107] For example, the mathematical expression of the second objective function is as follows:

[0108]

[0109] Using the delivery cycle limit, budget limit, equipment load limit, process compliance threshold, and emission limit recorded in the business constraints as model constraints, the constraint function is defined as follows:

[0110]

[0111] The first objective function and the second objective function are merged into a weighted multi-objective function, and model constraints are embedded in it. The mathematical expression of this weighted multi-objective function is as follows:

[0112]

[0113] Establish a mapping relationship: the input is a set of process parameters, and the output is a combination of cost and efficiency indicators;

[0114]

[0115] Where Y is the output indicator vector, which includes the cost reduction indicator C and the efficiency improvement indicator E;

[0116] Generate an optimization model that includes the objective function, constraint functions, and input-output relationships. Its mathematical expression is as follows:

[0117]

[0118] in, For an adjustable process parameter vector, Let the first objective function be... The second objective function is... , , These represent the costs of raw materials, energy consumption, and labor, respectively. Output per unit time To improve the yield rate, To produce a production rhythm, Let α, β, and γ be the energy efficiency ratio, α, β, and γ be the weights of the efficiency function, and λ be the energy consumption penalty coefficient. , , , , These include delivery cycle, budget, equipment load, process compliance, and emission function. , , , , For the corresponding upper limit of constraints, , For multi-objective weighting coefficients, , , , These are the normalized upper and lower bounds of the corresponding objective function in the sample dataset, respectively. To constrain the penalty function, This is a penalty factor.

[0119] It should be noted that the comprehensive objective function After normalization, its range is [0,1], where, When the value approaches 0, it indicates that the system achieves minimum cost and maximum efficiency while satisfying all constraints, representing the highest degree of optimization completion; when... When the value approaches 1, it indicates that the objective function has deviated significantly, and the decision variables or penalty coefficients need to be further adjusted.

[0120] Specifically, within the feasible solution space, the nonlinear mapping between input X and output Y is approximated using sample data, and this is achieved by solving... The minimum value yields the optimal parameter combination, since the model contains a penalty function term. It can automatically correct the search direction when constraints are violated, and achieve convergence within the feasible region.

[0121] Preferably, this step implements a multi-objective optimization modeling method that integrates Pilot runtime data with business constraints, enabling the model to not only ensure the achievement of cost reduction goals but also improve the stability of efficiency indicators, providing a clear mathematical solution basis for subsequent improvements to the particle swarm optimization algorithm.

[0122] S3. Solve the optimization model using the improved particle swarm optimization algorithm to obtain the parameter set that satisfies the business constraints. Note the following in this step:

[0123] In this preferred embodiment, in order to efficiently solve the optimization model constructed in step S2, it is necessary to set the basic parameters of the particle swarm optimization algorithm and initialize the swarm. Specifically, the system determines the particle dimension based on the number of process parameters to be optimized, and determines the upper and lower limits of each parameter based on historical operating data and process boundary conditions. Combining the project scale and computing power, the system sets the values ​​of the number of particles (e.g., 40), the maximum number of iterations (e.g., 150), the inertia weight range (e.g., 0.4~0.9), and the learning factor (e.g., 1.9). It is easy to understand that the inertia weight adopts a linear decreasing method during the algorithm operation, so as to enhance the global search capability in the early stage and improve the local convergence accuracy in the later stage.

[0124] After setting the parameters, within the allowable range of each process parameter, an initial particle swarm is randomly generated using Latin hypercube sampling. This ensures that the initial distribution of all particles uniformly covers the feasible solution space, thereby avoiding the problem of the initial solution set being too dense or biased towards the boundary. Each particle contains a parameter vector and a velocity vector. The initial velocity is randomly generated within a safe range and is used to control the step size of parameter adjustment. After the above settings, the initial state of the algorithm is established, laying a stable foundation for subsequent iterative optimization.

[0125] In each iteration, the fitness value of all particles needs to be calculated to evaluate their performance. The fitness function is composed of the objective function and the constraint penalty term, and its mathematical expression is as follows:

[0126]

[0127] in, This represents the fitness value of a particle in the current iteration. To synthesize the objective function value, This represents the calculation result of the k-th constraint. It takes a positive value when the constraint is violated, where K is the total number of constraints. This is a dynamic penalty coefficient;

[0128] The fitness assessment process includes the following two scenarios:

[0129] (1) Particles that satisfy the constraints

[0130] When all constraint functions of the particle When the condition is met, it indicates that all constraints are satisfied. In this case, the fitness value is determined by the comprehensive objective function. The direct decision involves inputting the cost reduction and efficiency improvement indicators of each particle into the objective function, and then performing a weighted sum using pre-defined weights to obtain the performance score for each particle. A lower score indicates lower cost and higher efficiency, giving the particle a better ranking in the ranking process. For example, in a pilot project, when the unit cost is 92 yuan / piece and the overall efficiency score is 0.88, the fitness value obtained after weighting through the objective function is 0.12. Therefore, this particle is considered better than a particle with a fitness value of 0.25.

[0131] (2) There are particles that violate the constraints.

[0132] When any constraint function When this happens, the corresponding penalty term is automatically added; the specific method is as follows: calculate the excess value of each violated constraint term, square it, and multiply it by the dynamic penalty coefficient of that constraint. All penalty terms are summed to form a total penalty value, which is then added to the fitness value. For example, when the delivery cycle exceeds the allowable limit by 3% and the budget exceeds the limit by 2%, the excess portion is squared and then weighted and added to significantly increase the fitness value of the particle. The penalty coefficient increases with the number of iterations to enhance the enforceability of constraint convergence. In this way, the penalized particle is at a disadvantage in the ranking and will be guided to move towards the feasible solution region in the next iteration.

[0133] Through this mechanism, the algorithm always maintains the search principle of prioritizing feasible solutions throughout the optimization process, that is, it prioritizes selecting particles that meet the constraints and have good target performance to enter the next generation of the population, thereby achieving continuous approximation of the feasible solution space.

[0134] Through the above optimization design, the particle swarm algorithm in this embodiment can achieve efficient solution in the multi-constraint and multi-objective environment of the pilot project. The reasonable parameter setting and balanced initialization distribution enable the algorithm to have good global exploration ability. The combination of fitness function design and dynamic penalty mechanism ensures compliance with constraints and fast convergence of objective function.

[0135] After calculating the fitness value of each particle, the historical best solution (i.e., individual best) of each particle and the global best solution of the current swarm will be automatically updated. Specifically, the current fitness value of each particle is compared with its historical best fitness value. If the current value is better, the current position of the particle is recorded as the new individual best position. At the same time, the particle with the best fitness value among all particles is selected as the global best solution for the current iteration. If there are multiple particles with the same fitness, the particle with lower cost and lower energy consumption is selected as the global best to meet the business orientation of cost reduction and efficiency improvement.

[0136] Through this two-layer update mechanism, the algorithm forms a dynamic balance between local and global search. The individual optimal solution ensures that the particle continues to explore along its own historical optimal direction, while the global optimal solution guides the group as a whole to move towards the global optimal region, thereby effectively preventing the search from getting trapped in local minima.

[0137] In each iteration, the particle's velocity and position need to be updated, and the update rule is expressed by the following formula:

[0138]

[0139]

[0140] in, The particle's current velocity. This is the particle's current position. The optimal position for an individual particle in its history. The optimal position for the entire group. The inertial weight decreases with the number of iterations. and These are individual learning factors and group learning factors, respectively. and A random number between 0 and 1, used to enhance the randomness of the search. This is a contraction factor used to prevent excessive speed from causing instability in the solution;

[0141] It should be noted that, in order to prevent particles from exceeding the feasible region of parameters, when a particle's position exceeds the boundary, the system will adopt a mirror repair strategy to map the out-of-bounds position back to within the boundary. At the same time, if a particle does not improve within several generations, the system will implement a small perturbation on some particles with poor fitness to increase population diversity and avoid premature convergence. Through the above mechanisms, the algorithm achieves a balance between global optimization and local refinement while ensuring feasibility.

[0142] When the iteration process meets the termination condition, the final global optimal solution is output as the optimal parameter set. Then, all process parameter values ​​are read from this parameter set and compared with the existing parameters in the process database of the pilot project to find the parameter items that need to be adjusted. For each parameter that needs to be adjusted, the system determines its target set value, allowable adjustment range, adjustment step size, execution order and execution node number.

[0143] For example, when the optimal parameter set indicates that the reaction temperature needs to be adjusted from 175 ℃ to 182 ℃, the system will generate the target value of 182 ℃, the allowable range of 170 ℃ to 190 ℃, and the single adjustment step size of 1 ℃, and set it as the first step according to the process logic; if the flow rate parameter needs to be adjusted from 50 L / min to 52 L / min, then the step size is set to 0.5 L / min and it is scheduled to be executed in the second step; and if the equipment speed parameter is increased from 840 rpm to 860 rpm, then it is scheduled to be executed in the third step.

[0144] The system encapsulates the above parameters and corresponding information into an optimization plan file. Each parameter includes fields such as parameter name, target value, adjustment range, adjustment step size, execution order, and node number. This file is stored in a structured format and can be directly recognized and executed by the control system, realizing automated optimization management of process parameters for pilot projects.

[0145] Preferably, the particle swarm optimization solution process of the present invention can achieve a global optimal search for process parameters while satisfying complex business constraints. Specifically, by controlling dynamic weights and shrinkage factors, global exploration in the early stage of the search and local refinement in the later stage are achieved; by using dynamic penalty mechanisms and feasible priority rules, the optimization process is ensured to converge stably to the feasible solution domain; and by using boundary repair and perturbation mechanisms, the algorithm is prevented from getting trapped in local optima. Therefore, it not only significantly improves the solution efficiency and stability of cost reduction and efficiency improvement schemes, but also ensures the feasibility of implementing the pilot project in a real production environment.

[0146] S4. Send the optimal parameter set as configuration information to the pilot project's management system for adaptive adjustment of the pilot project's process parameters, and collect post-implementation operational data within a preset 7-day period. Note the following in this step:

[0147] Convert the optimization scheme file into a structured instruction set, in key-value pair format;

[0148] The instruction set is uploaded to the configuration receiving module of the management and control system through the application interface;

[0149] The configuration receiving module parses the contents of the instruction set and generates a control task sequence;

[0150] The execution module modifies the process parameters one by one to the corresponding target values ​​according to the control task sequence;

[0151] Then record the modified parameter status to the runtime log file.

[0152] Furthermore, the specific operational data after implementation includes:

[0153] Automatically extract the operation records of each production cycle after the cycle ends;

[0154] Data collected includes actual process parameters, unit product cost, total output, yield rate, average production cycle time, resource utilization rate, total energy consumption, and number of alarm events.

[0155] The collected data from the seven cycles were merged in chronological order to generate a complete post-implementation operational data table;

[0156] Perform standardization processing on the implemented running data table to be consistent with the sample dataset, and output the standardized running dataset.

[0157] It should be noted that in this embodiment, by smoothly mapping the optimal parameter set to the parameter tuning actions that the device can perform, auditable and traceable data is provided for subsequent threshold judgment, ensuring the consistency and comparability of data before and after implementation, thereby reliably supporting the threshold judgment in the next step.

[0158] Preferably, compared to the extensive mode of manual parameter adjustment or one-time centralized distribution in existing technologies, this step has the advantages of being replicable, rollbackable, and low-risk: First, it significantly reduces the safety and compliance risks in the parameter implementation process; second, it ensures stable data quality within the 7-day observation window, improving the statistical credibility of cost reduction and efficiency improvement assessments; and third, it provides high-quality back-collection data and detailed operation trajectories for further optimization, increasing the success rate of promotion.

[0159] S5. Based on the operational data after implementation, calculate the cost reduction and efficiency improvement indicators of the selected optimization scheme, and compare them with the set first and second thresholds. Note that the following should be noted in this step:

[0160] Read the unit cost data before and after implementation from the standardized operation dataset, and calculate the cost reduction index C. The calculation formula is as follows:

[0161]

[0162] in, The unit product cost before implementation. This refers to the unit product cost after implementation.

[0163] Read the average production cycle time, yield rate, and resource utilization rate data before and after implementation from the standardized operation dataset, and calculate the efficiency improvement index E. The calculation formula is as follows:

[0164]

[0165] in, and These represent the average production cycle time before and after implementation. and The yield rates before and after implementation are respectively. and These represent resource utilization rates before and after implementation. , , These are the weighting coefficients for average production cycle time, yield rate, and resource utilization rate, respectively.

[0166] The cost reduction index C and efficiency improvement index E are output as input results for threshold comparison;

[0167] When the cost reduction indicator is greater than or equal to the first threshold and the efficiency improvement indicator is greater than or equal to the second threshold, the preferred optimization scheme is determined as the promotion scheme.

[0168] When the cost reduction indicator is less than the first threshold, or the efficiency improvement indicator is less than the second threshold, return to the optimization model solution step, and re-solve the model by correcting the penalty coefficient in the optimization model to obtain a new optimal parameter set, until the promotion plan is determined.

[0169] Furthermore, the method for setting the first threshold and the second threshold includes:

[0170] Retrieve statistical data on cost reduction and efficiency improvement indicators from at least 10 similar pilot projects from the historical project database;

[0171] Calculate the average of historical cost reduction indicators and multiply it by a correction factor of 1.05 to obtain the first threshold;

[0172] Calculate the upper quartile of the historical efficiency improvement index and multiply it by a correction factor of 1 to obtain the second threshold;

[0173] The first threshold and the second threshold are recorded in the threshold configuration table, which serves as the fixed input for the threshold comparison module.

[0174] In this preferred embodiment, when the cost reduction or efficiency improvement indicators fail to reach a preset threshold, the system will enter a correction process for the optimization model penalty coefficient to guide the algorithm to more effectively balance the objective function and constraints in the next iteration, thereby achieving adaptive adjustment of the optimization direction.

[0175] Extract the execution status and deviation data of the constraint functions from the previous solution results, including the frequency of constraint violations, the extent of exceeding limits, and the duration of violations. Based on the constraint type, they are divided into three categories: safety and compliance, economic performance, and stability. Safety and compliance includes hard constraints such as emission limits, equipment load, and personnel hour limits; economic performance includes cost-related conditions such as budget, energy consumption, and delivery cycle; and stability covers conditions such as process compliance, quality consistency, and production line balance.

[0176] For each type of constraint, the system adjusts the penalty weight according to the degree and importance of its violation. When a safety compliance constraint is violated, its penalty coefficient is directly increased to a higher level, forcing the algorithm to satisfy the constraint in the next round of solving. For economic performance constraints, the penalty coefficient is increased in stages according to the actual deviation ratio: smaller deviations are increased gradually, while larger deviations are increased rapidly. The penalty adjustment for stability constraints is based on volatility. If the same type of unstable behavior is detected in two consecutive solutions, the corresponding penalty value is doubled.

[0177] A rolling window is used to record the historical penalty change trend of each constraint. An inertial adjustment mechanism is formed through a trend weighting method. When a constraint remains stable and has no violation records in multiple iterations, the system automatically reduces its penalty intensity to expand the search space and restore the flexibility of the model. If out-of-bounds behavior occurs frequently in multiple iterations, the penalty intensity of the constraint is continuously increased until it reaches the preset upper limit. Through this mechanism, the model can dynamically tighten or loosen the constraint intensity according to historical performance, thereby improving the stability of convergence.

[0178] In penalty correction, the system adopts a risk-stratified strategy. For safety and compliance constraints, the penalty weight is adjusted with the highest magnitude and takes effect immediately. For economic performance constraints, the adjustment magnitude is moderate, and a verification solution is required before it takes effect to confirm the correct direction. For stability constraints, the penalty coefficient is fine-tuned through a secondary adjustment mechanism to prevent convergence stagnation due to excessive tightening. All correction rules are prioritized as follows: safety and compliance > economic performance > stability. When different types of constraints conflict, safety constraints take precedence, and other constraints decrease in order.

[0179] After each penalty coefficient adjustment, the system generates a penalty adjustment record table, which includes: the adjusted constraint, the penalty value before and after the adjustment, the adjustment range, the triggering reason, the affected round, and the effective range. All records are distinguished by version number and are automatically archived in the model version library to ensure subsequent traceability. If subsequent solutions show that the penalty value is too high, causing excessive shrinkage of the feasible region, the system will automatically reduce the penalty coefficient according to the decay rule to restore the model's activity.

[0180] After the correction is completed, the system writes the updated penalty coefficient into the optimization model configuration file and triggers a new round of solution tasks. The initial solution state uses the global optimal parameters of the previous round as the hot start point. At the same time, the objective function and constraint balance ratio are recalculated, so that the algorithm can explore better feasible solutions while maintaining continuity.

[0181] Through the above implementation methods, the correction of the penalty coefficient realizes intelligent adaptive adjustment at the constraint level, enabling the optimization process to have the dynamic evolution capabilities of self-learning, self-constraint, and self-recovery. This method can automatically adjust the search weights according to historical operation and constraint violation, thereby effectively avoiding constraint failure caused by excessive relaxation and solution stagnation caused by excessive tightening.

[0182] Furthermore, the correction process is combined with the model versioning mechanism, making each optimization and adjustment traceable and ensuring the security and controllability of the implementation. Through this mechanism, the algorithm gradually approaches the balanced solution that takes into account both cost optimization and efficiency optimization in multiple iterations, thereby achieving long-term stability of the goals of cost reduction and efficiency improvement.

[0183] The aforementioned preprocessing methods for historical operational data can be carried out using existing technologies and methods, and will not be elaborated further in this example.

[0184] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. An optimization method for cost reduction and efficiency improvement based on pilot testing, characterized in that, include: Collect historical operational data and business constraints of the pilot project, and preprocess the historical operational data to obtain a sample dataset; including: Read the historical operation data, use an interpolation algorithm to fill in missing items, and remove outliers by three times the standard deviation; The completed numerical data is subjected to minimum and maximum standardization, which linearly maps all parameter values ​​to the interval [0,1]. The data is segmented using a sliding method with a time window length of 7 days, and each segment forms a record unit; For each record unit, calculate the average process parameter value, average energy consumption value, average output cycle time, average yield rate, and average equipment utilization rate; Each record unit and its corresponding business constraints are combined to form a sample entry; Merge all sample entries to generate the sample dataset; The historical operating data includes the set process parameters, actual operating parameters, unit product energy consumption, unit product cost, equipment operating time, equipment utilization rate, production cycle time, yield rate, and number of defective products for the target process during the historical operating phase. The business constraints include at least the upper limit of delivery cycle, upper limit of budget, upper limit of product quality consistency, upper limit of equipment load, upper limit of personnel working hours, process compliance threshold and emission limit; An optimization model is constructed based on the sample dataset and the business constraints, with the goal of optimizing cost reduction and efficiency improvement indicators; including: The decision variables are set as adjustable process parameters in the sample dataset; The first objective function is defined as minimizing unit cost, and the second objective function is defined as maximizing both production cycle time and yield rate. The upper limit of delivery cycle, upper limit of budget, upper limit of equipment load, process compliance threshold and emission limit recorded in the business constraints are used as model constraints. The first objective function and the second objective function are combined into a weighted multi-objective function, and the model constraints are embedded therein; Establish a mapping relationship, where the input is a set of process parameters and the output is a combination of cost and efficiency indicators; Generate an optimization model that includes the objective function, constraint functions, and input-output relationships; The optimization model is solved using an improved particle swarm optimization algorithm to obtain a parameter set that satisfies the business constraints; including: Set the particle swarm size, maximum number of iterations, initial value of inertia weight, and learning factor; Randomly initialize the parameter vector for each particle within the parameter feasible region; Calculate the fitness value for each particle, where the fitness function is a weighted sum of the objective function and constraint penalty terms; The individual optimal solution and the global optimal solution are determined based on the fitness values. Update the parameter vectors of all particles according to the velocity and position update formula; When the maximum number of iterations is reached, output the current global optimal solution as the parameter set; The parameter set is sent to the pilot project's management system in the form of configuration information for operation, adaptively adjusting the pilot project's process parameters, and collecting post-implementation operation data within a preset period of 7 days. Based on the operational data after implementation, the cost reduction and efficiency improvement indicators of the parameter group are calculated and compared with the set first and second thresholds: When the cost reduction indicator is greater than or equal to the first threshold and the efficiency improvement indicator is greater than or equal to the second threshold, the parameter group is determined as the promotion plan; When the cost reduction index is less than the first threshold, or the efficiency improvement index is less than the second threshold, the process returns to the optimization model solution step. By correcting the penalty coefficient in the optimization model, the solution is re-solved to obtain a new set of parameters, until the promotion plan is determined.

2. The optimization method for cost reduction and efficiency improvement based on pilot according to claim 1, characterized in that, The parameter set specifically includes: Read all process parameter values ​​from the parameter group; Compare the values ​​of each process parameter with the corresponding parameters in the current pilot project process database to determine the parameter items that need to be adjusted; For each parameter that needs adjustment, determine the target setting value, adjustment range, adjustment step size, execution order, and execution node number; The parameters that need to be adjusted are encapsulated into an optimization plan file. Each parameter in the file contains a parameter identifier, target value, adjustment range, adjustment step size, and execution node number.

3. The optimization method for cost reduction and efficiency improvement based on pilot according to claim 2, characterized in that, Adaptive adjustment of the process parameters for the pilot project, including: The optimization scheme file is converted into a structured instruction set, in key-value pair format; The instruction set is uploaded to the configuration receiving module of the management and control system via the application interface; The configuration receiving module parses the contents of the instruction set and generates a control task sequence; The execution module modifies the process parameters one by one to the corresponding target values ​​according to the control task sequence; Then record the modified parameter status to the runtime log file.

4. The optimization method for cost reduction and efficiency improvement based on pilot according to claim 3, characterized in that, The operational data after implementation will be collected within a preset period of 7 days. The operational data after implementation specifically includes: Automatically extract the operation records of each production cycle after the cycle ends; Data collected includes actual process parameters, unit product cost, total output, yield rate, average production cycle time, resource utilization rate, total energy consumption, and number of alarm events. The collected data from the seven cycles were merged in chronological order to generate a complete post-implementation operational data table; Perform standardization processing on the post-implementation running data table, consistent with the sample dataset, and output a standardized running dataset.

5. The optimization method for cost reduction and efficiency improvement based on pilot according to claim 4, characterized in that, Based on the operational data after implementation, the cost reduction and efficiency improvement indicators of the parameter group are calculated, including: Read the unit cost data before and after implementation from the standardized operation dataset, and calculate the cost reduction index C. The calculation formula is as follows: in, The unit product cost before implementation. This refers to the unit product cost after implementation. Read the average production cycle time, yield rate, and resource utilization rate data before and after implementation from the standardized operation dataset, and calculate the efficiency improvement index E. The calculation formula is as follows: in, and These represent the average production cycle time before and after implementation. and The yield rates before and after implementation are respectively. and These represent resource utilization rates before and after implementation. , , These are the weighting coefficients for average production cycle time, yield rate, and resource utilization rate, respectively. The cost reduction index C and efficiency improvement index E are output as input results for threshold comparison.

6. The optimization method for cost reduction and efficiency improvement based on pilot according to claim 5, characterized in that, The method for setting the first threshold and the second threshold includes: Retrieve statistical data on cost reduction and efficiency improvement indicators from at least 10 similar pilot projects from the historical project database; Calculate the average of historical cost reduction indicators and multiply it by a correction factor of 1.05 to obtain the first threshold. Calculate the upper quartile of the historical efficiency improvement index and multiply it by a correction factor of 1 to obtain the second threshold; The first threshold and the second threshold are recorded in the threshold configuration table as fixed inputs to the threshold comparison module.

Images