Intelligent APS scheduling method and system based on multi-source heterogeneous ICS platform

Abstract

The invention provides an intelligent APS scheduling method and system based on a multi-source heterogeneous ICS platform, and relates to the technical field of intelligent manufacturing, and the method comprises the steps: obtaining production equipment, a process route and order information, generating a process propagation matrix, carrying out the tensor decomposition to obtain a process priority feature vector, mapping the process priority feature vector to a two-dimensional grid space to form a process group, and obtaining a process group; and calculating a conflict coefficient and a profit value, executing game optimization iteration, outputting an optimal scheduling scheme, and issuing and executing the optimal scheduling scheme. According to the method, the problem of production scheduling optimization in a multi-source heterogeneous data environment can be effectively solved, and the utilization rate of production resources and the order delivery timeliness rate are improved.

Description

Technical Field

[0001] This invention relates to the field of intelligent manufacturing technology, and in particular to an intelligent APS scheduling method and system based on a multi-source heterogeneous ICS platform. Background Technology

[0002] With the deepening of intelligent transformation in the manufacturing industry, Integrated Control Systems (ICS) have become an indispensable core component of modern industrial production. Especially in complex production environments, Advanced Planning and Scheduling Systems (APS) based on multi-source heterogeneous ICS platforms are of great significance for improving production efficiency, reducing costs, and optimizing resource allocation. Currently, manufacturing enterprises commonly use various information systems to collect and manage production data, including equipment status information, process routes, and order management. These data come from diverse sources and are presented in various formats, constituting a typical multi-source heterogeneous data environment. Based on this, the scientific formulation and dynamic optimization of production plans through intelligent scheduling algorithms has become one of the key technologies for intelligent manufacturing.

[0003] Traditional APS scheduling methods have significant shortcomings when facing multi-source heterogeneous ICS platform environments. Traditional scheduling methods struggle to effectively handle heterogeneous data from different sources and formats, leading to difficulties in data integration and an inability to fully utilize the production information value provided by multi-source data. Particularly, they lack scientifically effective methods for quantitatively describing process relationships. Existing technologies are relatively simple in process grouping and conflict identification, failing to comprehensively consider multi-dimensional factors such as equipment sharing, material overlap, and process overlap. This results in insufficient conflict handling capabilities of scheduling schemes in complex production environments, making them unsuitable for highly complex modern manufacturing environments. Traditional scheduling optimization strategies typically employ fixed-weight objective functions, lacking adaptability to dynamic production environments. Especially in situations with multiple orders competing for resources, they cannot dynamically adjust the optimization objective based on actual production conditions, making it difficult to achieve globally optimal scheduling results.

[0004] With the rapid development of Industry 4.0 and intelligent manufacturing, higher requirements are placed on the intelligence, adaptability and optimization capabilities of production scheduling systems. There is an urgent need for an intelligent APS scheduling method that can make full use of multi-source heterogeneous data, scientifically handle process conflicts and achieve dynamic optimization. Summary of the Invention

[0005] This invention provides an intelligent APS scheduling method and system based on a multi-source heterogeneous ICS platform, which can solve the problems in the prior art.

[0006] A first aspect of this invention provides an intelligent APS scheduling method based on a multi-source heterogeneous ICS platform, comprising:

[0007] Obtain production equipment information, process route information, and order information through a multi-source heterogeneous ICS platform;

[0008] The process propagation matrix is ​​generated based on the process route information, the process correlation is extracted, the corresponding information entropy weights are calculated, and the process propagation matrix is ​​decomposed into tensor vectors to obtain the process priority feature vectors.

[0009] The process priority feature vector is mapped to a two-dimensional grid space, the activation value of the grid node is calculated to form a process group, the equipment sharing degree of the process group is calculated according to the production equipment information, the material crossover degree and process overlap degree of the process group are calculated, a feature matrix is ​​constructed and the process group conflict coefficient is obtained based on the feature value analysis.

[0010] The expected revenue value of the process group is calculated based on the conflict coefficient of the process group and the order information. The marginal revenue decay factor is introduced into the expected revenue value to obtain the actual revenue value. Game optimization iteration is performed based on the actual revenue value. When the revenue increment between two adjacent iterations is less than the preset convergence threshold, the optimal scheduling scheme is output.

[0011] The optimal scheduling scheme is converted into a sequence of scheduling instructions and then issued for execution.

[0012] In one optional embodiment, a process propagation matrix is ​​generated based on the process route information, process relationships are extracted, corresponding information entropy weights are calculated, and tensor decomposition is performed on the process propagation matrix to obtain a process priority feature vector, including:

[0013] Construct the process topology based on the process route information, calculate the propagation strength between process nodes, and generate the initial propagation matrix;

[0014] A regulatory neural network is constructed in the initial propagation matrix to adaptively regulate the propagation intensity of process nodes;

[0015] Based on the adjusted propagation matrix, process correlation characteristics are calculated, and a process characteristic evaluation matrix is ​​constructed.

[0016] Calculate the local information entropy of the process feature evaluation matrix, generate feature weights, and feed them back to the regulatory neural network to obtain the optimized propagation matrix;

[0017] The optimized propagation matrix is ​​constructed as a feature tensor, and the Tucker decomposition method is used to perform feature decomposition on the feature tensor. Based on the feature contribution, the process priority feature vector is extracted.

[0018] In one optional embodiment, constructing a regulatory neural network in the initial propagation matrix to adaptively regulate the propagation intensity of process nodes includes:

[0019] A control neuron is configured at each process node position of the initial propagation matrix. Each control neuron contains a threshold judgment unit, a state response unit, and a feedback gain unit. A neural network topology is constructed among the control neurons.

[0020] The threshold judgment unit monitors the process operation status and outputs judgment signals. The status response unit calculates the control factor based on the judgment signal. The control factor is amplified by the feedback gain unit to form a control command. Based on the control command, the propagation intensity of the corresponding process node in the initial propagation matrix is ​​updated. Local feedback pathways are established between adjacent control neurons to form an adaptive control network.

[0021] Based on the output of the adaptive control network, the process association characteristics are calculated, including node propagation direction coefficient, node response strength and node hierarchical relationship, and a process characteristic evaluation matrix is ​​constructed.

[0022] Calculate the local information entropy of the process feature evaluation matrix, combine it with the operating parameters of the control neurons to generate entropy correction coefficients, and correct the local information entropy to obtain feature weights;

[0023] The feature weights are input into the state response unit of the regulating neuron, and the response parameters are adjusted based on the feature weights to establish a dynamic mapping relationship between the feature weights and the response parameters, thereby generating an optimized propagation matrix.

[0024] In an optional embodiment, the process priority feature vector is mapped to a two-dimensional grid space, the activation values ​​of the grid nodes are calculated to form process groups, the equipment sharing degree of the process groups is calculated based on the production equipment information, the material crossover degree and process overlap degree of the process groups are calculated, a feature matrix is ​​constructed, and the process group conflict coefficient is obtained based on eigenvalue analysis, including:

[0025] The process priority feature vector is mapped to a two-dimensional grid space, and the activation value of the grid node in the two-dimensional grid space is calculated by the hyperbolic tangent function. The input of the hyperbolic tangent function includes the mapping weight coefficient and bias term of the process priority feature vector.

[0026] The Euclidean distance between the grid nodes is calculated based on the activation value of the grid nodes, and the processes corresponding to the grid nodes whose Euclidean distance is less than a preset distance threshold are divided into the same process group.

[0027] Based on the set of application equipment for each process pair in a process group, the ratio of the intersection cardinality to the union potential of the application equipment sets is calculated to obtain the equipment sharing degree between the process pairs.

[0028] Based on the material input and output information of the process pair, the weighted sum of the material input overlap and material output overlap is calculated as a ratio to the process group size to obtain the material crossover degree between the process pairs.

[0029] Based on the process path information of the process pairs, the ratio of the intersection potential to the union potential of the process path sets is calculated to obtain the process overlap between the process pairs.

[0030] A fusion feature matrix of equipment sharing degree, material crossover degree, and process overlap degree is constructed. Based on the eigenvalue analysis and feature fusion of the fusion feature matrix, the conflict coefficient of the process group is obtained.

[0031] In one optional embodiment, a fusion feature matrix of equipment sharing, material overlap, and process overlap is constructed. Based on the eigenvalue analysis and feature fusion of the fusion feature matrix, the process group conflict coefficient is obtained, including:

[0032] The equipment sharing degree in the process group is mapped to the first quaternion. Based on the value of the equipment sharing degree, the real part of the first quaternion is determined, the fluctuation value of the equipment sharing degree is calculated, the imaginary part in the first direction is determined, and the remaining imaginary parts are zero.

[0033] The material crossover degree is mapped to a second quaternion. Based on the value of the material crossover degree, the real part of the second quaternion is determined, the fluctuation value of the material crossover degree is calculated, the imaginary part of the second direction is determined, and the rest of the imaginary parts are zero.

[0034] The process overlap is mapped to a third quaternion. Based on the value of the process overlap, the real part of the third quaternion is determined, the fluctuation value of the process overlap is calculated, the third-directed imaginary part is determined, and the remaining imaginary parts are zero.

[0035] A fusion feature matrix is ​​constructed based on the first quaternion, the second quaternion, and the third quaternion. The diagonal elements of the fusion feature matrix are the first quaternion, and the off-diagonal elements are composed of alternating arrangements of the second quaternion and the third quaternion.

[0036] The characteristic equation of the fused feature matrix is ​​solved to obtain the eigenvalues. The eigenvalues ​​are normalized and then projected into a four-dimensional space. The projection result is then weighted and fused with the original features to obtain the process group conflict coefficient.

[0037] In one optional embodiment, the expected revenue value of the process group is calculated based on the conflict coefficient of the process group and order information. A marginal revenue decay factor is introduced into the expected revenue value to obtain the actual revenue value. Game optimization iteration is performed based on the actual revenue value. When the revenue increment between two adjacent iterations is less than a preset convergence threshold, the optimal scheduling scheme is output, including:

[0038] For each process group, the conflict coefficient of the process group is multiplied by the weight coefficient and value coefficient of the corresponding order and then summed to calculate the expected revenue value of the process group.

[0039] Obtain the current resource quantity allocated to the process group, calculate the resource ratio of the process group, set a benchmark attenuation coefficient according to the importance of the resource type, and when the resource ratio exceeds a preset occupancy threshold, multiply the benchmark attenuation coefficient by the proportion exceeding the preset occupancy threshold to obtain the marginal attenuation coefficient. Use the product of the marginal attenuation coefficient and the resource ratio as the independent variable of the exponential function to calculate the marginal revenue attenuation factor.

[0040] Multiply the expected revenue value of the process group by the marginal revenue decay factor to obtain the actual revenue value of the process group;

[0041] Calculate the degree of process connection between process groups, which is determined based on the sequential connection relationship between processes in the process group; construct a competition adjustment coefficient between process groups based on the degree of process connection.

[0042] Construct a payment matrix by multiplying the actual revenue values ​​of any two process groups by their corresponding competition adjustment coefficients;

[0043] Dynamic game optimization is performed based on the payoff matrix. The scheduling scheme of each process group is iteratively updated until the profit increment between two adjacent iterations is less than the preset convergence threshold, and the optimal scheduling scheme is output.

[0044] In one optional embodiment, dynamic game optimization is performed based on the payoff matrix, iteratively updating the scheduling scheme of each process group until the payoff increment between two adjacent iterations is less than a preset convergence threshold, and the optimal scheduling scheme is output, including:

[0045] Using the priority order and resource requirements of the work groups as input parameters, the payment matrix is ​​constructed in blocks to generate an initial resource allocation and scheduling scheme.

[0046] Each process group is traversed sequentially, and the game payoff of the process group with other process groups is calculated based on the payoff matrix to obtain the optimal response scheduling scheme that maximizes the payoff.

[0047] Establish a historical revenue path feature library for process groups, extract the temporal patterns of revenue changes, and generate the scheduling scheme distribution for other process groups in the next iteration;

[0048] Adjust the response scheduling scheme of the current work group according to the distribution of the scheduling scheme, and generate an adaptive step size factor by combining the change gradient of the revenue value of the work group and the competition intensity between the work groups.

[0049] The current scheduling scheme of the work group is updated towards the adjusted response scheduling scheme according to the adaptive step size factor to obtain the updated scheduling scheme; the updated scheduling scheme that exceeds the resource capacity constraint is projected into the feasible strategy space.

[0050] The revenue value of each process group is recalculated based on the feasible strategy space, the revenue increment between two adjacent iterations is obtained, and the historical revenue path feature library is updated at the same time.

[0051] The convergence threshold is dynamically adjusted using a logarithmic function of the number of iterations. The optimal scheduling scheme is determined when the profit increment of all process groups is less than the corresponding convergence threshold; otherwise, the process is re-executed.

[0052] A second aspect of this invention provides an intelligent APS scheduling system based on a multi-source heterogeneous ICS platform, comprising:

[0053] The first unit is used to obtain production equipment information, process route information, and order information through the multi-source heterogeneous ICS platform;

[0054] The second unit is used to generate a process propagation matrix based on the process route information, extract process correlations, calculate the corresponding information entropy weights, and perform tensor decomposition on the process propagation matrix to obtain the process priority feature vector.

[0055] The third unit is used to map the process priority feature vector to a two-dimensional grid space, calculate the activation value of the grid node to form a process group, calculate the equipment sharing degree of the process group according to the production equipment information, calculate the material crossover degree and process overlap degree of the process group, construct a feature matrix, and obtain the process group conflict coefficient based on the feature value analysis.

[0056] The fourth unit is used to calculate the expected revenue value of the process group based on the conflict coefficient of the process group and the order information. The marginal revenue decay factor is introduced into the expected revenue value to obtain the actual revenue value. Game optimization iteration is performed based on the actual revenue value. When the revenue increment between two adjacent iterations is less than the preset convergence threshold, the optimal scheduling scheme is output.

[0057] The fifth unit is used to convert the optimal scheduling scheme into a sequence of scheduling instructions and issue them for execution.

[0058] A third aspect of the present invention provides an electronic device, comprising:

[0059] processor;

[0060] Memory used to store processor-executable instructions;

[0061] The processor is configured to invoke instructions stored in the memory to execute the aforementioned method.

[0062] A fourth aspect of the present invention provides a computer-readable storage medium having stored thereon computer program instructions that, when executed by a processor, implement the aforementioned method.

[0063] In this embodiment of the invention, production data is acquired through a multi-source heterogeneous ICS platform, and process priority features are extracted using tensor decomposition and mapped to a two-dimensional grid space to form process groups. This effectively solves the problem of low scheduling efficiency of traditional APS systems in complex production environments and achieves optimized allocation of production resources. By calculating the equipment sharing degree, material crossover degree, and process overlap degree of process groups, a feature matrix is ​​constructed and the conflict coefficient is analyzed, providing a scientific basis for order scheduling, avoiding resource conflicts, and improving the utilization rate and capacity output of the production line. A game optimization mechanism of expected return value and marginal return decay factor is introduced. When the incremental return of iteration is less than a preset threshold, the optimal solution is output, making the scheduling results more in line with actual production needs, enhancing the executability and robustness of the production plan, and reducing the production cycle and operating costs. Attached Figure Description

[0064] Figure 1 This is a flowchart illustrating the intelligent APS scheduling method based on a multi-source heterogeneous ICS platform according to an embodiment of the present invention.

[0065] Figure 2 This is a flowchart illustrating the calculation of the conflict coefficient of the process group in an embodiment of the present invention. Detailed Implementation

[0066] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0067] The technical solution of the present invention will be described in detail below with reference to specific embodiments. These specific embodiments can be combined with each other, and the same or similar concepts or processes may not be described again in some embodiments.

[0068] Figure 1 This is a flowchart illustrating the intelligent APS scheduling method based on a multi-source heterogeneous ICS platform according to an embodiment of the present invention. Figure 1 As shown, the method includes:

[0069] Obtain production equipment information, process route information, and order information through a multi-source heterogeneous ICS platform;

[0070] The process propagation matrix is ​​generated based on the process route information, the process correlation is extracted, the corresponding information entropy weights are calculated, and the process propagation matrix is ​​decomposed into tensor vectors to obtain the process priority feature vectors.

[0071] The process priority feature vector is mapped to a two-dimensional grid space, the activation value of the grid node is calculated to form a process group, the equipment sharing degree of the process group is calculated according to the production equipment information, the material crossover degree and process overlap degree of the process group are calculated, a feature matrix is ​​constructed and the process group conflict coefficient is obtained based on the feature value analysis.

[0072] The expected revenue value of the process group is calculated based on the conflict coefficient of the process group and the order information. The marginal revenue decay factor is introduced into the expected revenue value to obtain the actual revenue value. Game optimization iteration is performed based on the actual revenue value. When the revenue increment between two adjacent iterations is less than the preset convergence threshold, the optimal scheduling scheme is output.

[0073] The optimal scheduling scheme is converted into a sequence of scheduling instructions and then issued for execution.

[0074] In one optional implementation, a process propagation matrix is ​​generated based on the process route information, process relationships are extracted, corresponding information entropy weights are calculated, and tensor decomposition is performed on the process propagation matrix to obtain a process priority feature vector, including:

[0075] Construct the process topology based on the process route information, calculate the propagation strength between process nodes, and generate the initial propagation matrix;

[0076] A regulatory neural network is constructed in the initial propagation matrix to adaptively regulate the propagation intensity of process nodes;

[0077] Based on the adjusted propagation matrix, process correlation characteristics are calculated, and a process characteristic evaluation matrix is ​​constructed.

[0078] Calculate the local information entropy of the process feature evaluation matrix, generate feature weights, and feed them back to the regulatory neural network to obtain the optimized propagation matrix;

[0079] The optimized propagation matrix is ​​constructed as a feature tensor, and the Tucker decomposition method is used to perform feature decomposition on the feature tensor. Based on the feature contribution, the process priority feature vector is extracted.

[0080] In one specific implementation, during the actual process implementation, when constructing the process topology based on the process route information, process nodes need to be represented as nodes in a graph structure, and the relationships between processes need to be represented as edges. For example, the production of a product includes 10 processes, labeled P1, P2, P3...P10. By analyzing the process flow table, the dependencies between processes are determined, such as P1 must be completed before P3, and P2 and P4 having the possibility of parallel processing. Based on these dependencies, a directed graph G = (V, E) is constructed, where V represents the set of processes and E represents the dependencies between processes. The propagation strength between nodes is then calculated, determined by a combination of material flow, information flow, and time dependency between processes. For example, if material transfer between P1 and P3 is frequent, a higher weight of 0.85 is assigned; if P2 and P5 only have information dependency, a medium weight of 0.63 is assigned. In this way, a 10×10 initial propagation matrix M is constructed, where each element M(i, j) represents the propagation strength from process i to process j.

[0081] When constructing the control neural network based on the initial propagation matrix, a three-layer network structure is adopted: the input layer corresponds to the initial process node information, the hidden layer realizes the adaptive adjustment of the propagation intensity, and the output layer generates the adjusted propagation intensity value. In the specific implementation, the number of neurons in the hidden layer is set to 15, and the activation function is a linear rectified function. Through forward propagation calculation, each element M(i,j) in the initial propagation matrix is ​​input into the network, and the hidden layer adjusts the weight coefficients through the learning process. For example, the propagation intensity of P1 to P3 in the propagation matrix is ​​0.85, which becomes 0.92 after adjustment by the neural network, indicating that the correlation between the two processes is enhanced; while the propagation intensity of P2 to P7 is adjusted from 0.32 to 0.25, indicating that the correlation is weakened. The training data of the control network comes from the historical process execution results, including actual production data such as process execution time and quality indicators. After 50 iterations of the control process, the propagation intensity value tends to stabilize, and the adjusted propagation matrix M' is obtained.

[0082] When calculating process association characteristics based on the adjusted propagation matrix, the association between processes needs to be evaluated from multiple dimensions. The propagation distance feature fd represents the logical distance between processes in the process flow; the propagation strength feature fs represents the ability of processes to transfer information or materials; the propagation stability feature ft represents the reliability of the process association; and the propagation importance feature fi represents the degree of impact of the process association on the overall production process. Taking the association between P3 and P7 as an example, the calculated propagation distance feature value is 0.65, the propagation strength feature value is 0.78, the propagation stability feature value is 0.82, and the propagation importance feature value is 0.71. Based on these four features, a process feature evaluation matrix E is constructed. This matrix is ​​a 10×10×4 three-dimensional matrix, where E(i,j,k) represents the evaluation value of the k-th feature from process i to process j.

[0083] When calculating the local information entropy of the process feature evaluation matrix, the information entropy value is calculated for each feature dimension. In the local information entropy calculation, the distribution of each feature value in the whole is statistically analyzed, and the uncertainty of this distribution is calculated. For example, the information entropy of the propagation distance feature is 0.78, the information entropy of the propagation intensity feature is 0.65, the information entropy of the propagation stability feature is 0.82, and the information entropy of the propagation importance feature is 0.91. Based on the information entropy values, the weights of each feature are calculated, and the weights are inversely proportional to the information entropy value. Specifically, the calculated feature weights are: propagation distance feature weight 0.22, propagation intensity feature weight 0.26, propagation stability feature weight 0.21, and propagation importance feature weight 0.31. These weight values ​​are fed back to the regulatory neural network to further optimize the propagation matrix. After 5 iterations of optimization, the final optimized propagation matrix M is obtained.

[0084] When constructing the optimized propagation matrix as a feature tensor, the influence of the time dimension is considered. The propagation matrices at different time points are merged to construct a 10×10×T three-dimensional tensor, where T is the number of time sampling points. In the implementation case, T=6. When performing feature decomposition on the feature tensor using the Tucker decomposition method, the three-dimensional tensor is decomposed into the product of a core tensor and three factor matrices. The decomposition rank is set to (5, 5, 3), indicating that 5, 5, and 3 main features are retained in the three dimensions, respectively. The core tensor G and the three factor matrices A, B, and C are obtained through calculation. Factor matrices A and B correspond to the sending and receiving features of the process nodes, respectively, and matrix C corresponds to the time dimension features. Based on the norm of the eigenvectors corresponding to each process in factor matrix A, the feature contribution of the process is calculated. For example, the feature contribution of P1 is 0.87, the feature contribution of P2 is 0.73, and so on. After normalizing the feature contribution, the process priority feature vector V = (0.12, 0.10, 0.15, 0.08, 0.11, 0.14, 0.09, 0.07, 0.06, 0.08) is obtained. This vector represents the relative importance of the 10 processes, with larger values ​​indicating higher priority. In actual process optimization, the allocation of process resources and execution order can be adjusted based on this feature vector to effectively improve overall production efficiency.

[0085] The above implementation method realizes the conversion of process route information into process priority feature vectors, providing data support and decision-making basis for process optimization in intelligent manufacturing. This method fully utilizes the topological information contained in the process route, achieves adaptive control through neural networks, calculates feature weights using information entropy theory, and extracts key features through tensor decomposition, forming a complete process optimization analysis framework.

[0086] In one optional implementation, constructing a regulatory neural network in the initial propagation matrix to adaptively regulate the propagation intensity of process nodes includes:

[0087] A control neuron is configured at each process node position of the initial propagation matrix. Each control neuron contains a threshold judgment unit, a state response unit, and a feedback gain unit. A neural network topology is constructed among the control neurons.

[0088] The threshold judgment unit monitors the process operation status and outputs judgment signals. The status response unit calculates the control factor based on the judgment signal. The control factor is amplified by the feedback gain unit to form a control command. Based on the control command, the propagation intensity of the corresponding process node in the initial propagation matrix is ​​updated. Local feedback pathways are established between adjacent control neurons to form an adaptive control network.

[0089] Based on the output of the adaptive control network, the process association characteristics are calculated, including node propagation direction coefficient, node response strength and node hierarchical relationship, and a process characteristic evaluation matrix is ​​constructed.

[0090] Calculate the local information entropy of the process feature evaluation matrix, combine it with the operating parameters of the control neurons to generate entropy correction coefficients, and correct the local information entropy to obtain feature weights;

[0091] The feature weights are input into the state response unit of the regulating neuron, and the response parameters are adjusted based on the feature weights to establish a dynamic mapping relationship between the feature weights and the response parameters, thereby generating an optimized propagation matrix.

[0092] In one specific implementation, an initial propagation matrix containing multiple process nodes is constructed. This matrix represents the connection relationship and initial propagation strength between processes in the manufacturing process. For example, for an assembly process containing 5 process nodes, a 5×5 initial propagation matrix M can be constructed, where M(i,j) represents the propagation strength from process i to process j. The initial value can be set based on historical production data, such as M(1,2) = 0.8 indicating that process 1 has a strong influence on process 2.

[0093] A control neuron is configured at each process node position in the initial propagation matrix to form a neural network topology. Each control neuron consists of three functional units: a threshold judgment unit, a state response unit, and a feedback gain unit. Taking the i-th process node in the matrix as an example, the threshold judgment unit of its control neuron Ni is set with a warning threshold δi = 0.75, the initial response coefficient of the state response unit is αi = 0.5, and the gain factor of the feedback gain unit is gi = 1.2. These parameter values ​​can be adjusted according to actual production needs.

[0094] The workflow of the control neurons is as follows: The threshold judgment unit continuously monitors the process operation status data, such as the processing time, quality pass rate, and equipment load of process i. When a certain indicator is detected to exceed the preset threshold, a judgment signal Si is output. For example, when the processing time of process 2 exceeds 20% of the standard time, its judgment unit outputs Si = 1; otherwise, it outputs Si = 0. The state response unit receives the judgment signal and calculates the control factor θi = αi × Si + βi × Vi, where Vi is the feedback signal from the adjacent control neurons, and βi is the weighting coefficient with a value of 0.3. The feedback gain unit amplifies the control factor and generates the control instruction Ci = gi × θi. Finally, the propagation strength of the corresponding node in the propagation matrix is ​​updated based on the control instruction, with the update rule being M'(i, j) = M(i, j) × (1 + Ci).

[0095] Local feedback pathways are established between adjacent regulatory neurons, forming an adaptive regulatory network. For example, the regulatory neuron N2 in step 2 transmits its state information to the adjacent regulatory neurons in steps 1 and 3. This local feedback mechanism enables the entire network to respond to local perturbations and dynamically adjust the propagation intensity.

[0096] Based on the output of the adaptive control network, the process association characteristics are calculated. For process node i, its node propagation direction coefficient Di is calculated, representing the node's ability to influence downstream processes. The Di value is obtained by calculating the weighted sum of the propagation intensities M'(i,j) from node i to all reachable nodes j. In the actual calculation, for the example of 5 nodes, the propagation direction coefficient D1 = 0.85 for process 1, indicating that it has a strong influence on subsequent processes. The node response intensity Ri reflects the sensitivity of the process node to changes in upstream processes, and is obtained by calculating the weighted sum of the propagation intensities M'(j,i) from all reachable nodes j to node i. For example, the response intensity R3 = 0.72 for process 3 indicates that it has a certain degree of response to changes in preceding processes. The node hierarchy Li represents the importance of the process in the entire process, and is calculated by the in-degree and out-degree of the nodes. For example, the hierarchy Li of process 2 = 0.68 indicates that it occupies a relatively important position in the process.

[0097] Construct a process feature evaluation matrix E, where E(i,j) represents the comprehensive influence of process i on process j, calculated as E(i,j) = w1×Di + w2×Rj + w3×|Li-Lj|, where w1, w2, and w3 are weight coefficients with values ​​of 0.4, 0.4, and 0.2, respectively. In the example, E(1,3) = 0.75, indicating that process 1 has a strong comprehensive influence on process 3.

[0098] The local information entropy H(i) of the process feature evaluation matrix is ​​calculated, representing the degree of uncertainty of process i. For process 2, its local information entropy H(2) = 0.65, indicating that there is a certain degree of uncertainty in this process. Combining the operating parameters of the regulating neuron, the entropy correction coefficient λi = f(αi, gi, H(i)) is generated, which associates the neuron parameters with the entropy value. For example, the entropy correction coefficient λ2 = 0.82 for process 2. The local information entropy is corrected by applying the correction coefficient, resulting in the feature weight Wi = H(i) × λi. In the example, the feature weight W2 = 0.53 for process 2.

[0099] Feature weights are input into the state response unit of the regulating neuron to establish a dynamic mapping relationship between feature weights and response parameters. Specifically, the response coefficient αi of the state response unit is adjusted according to the feature weights as αi' = αi × (1 + γ × Wi), where γ is an adjustment factor with a value of 0.5. For process 2, the adjusted response coefficient α2' = 0.63. This dynamic adjustment mechanism enables the system to adaptively adjust the response parameters according to the importance of process features.

[0100] Based on the updated parameters of the regulating neurons, the propagation intensity is recalculated to generate an optimized propagation matrix M. In the example, the optimized propagation intensity M(1,2) = 0.92 from process 1 to process 2, which is a significant improvement compared to the initial value of 0.8, indicating that the system recognizes the importance of this correlation. In this way, the system can dynamically adjust the propagation intensity between processes according to the process operation status, improving the adaptability and stability of the manufacturing process.

[0101] In existing technologies, process correlation analysis mainly employs static statistical models or simple linear propagation models, such as common Markov chain models and Bayesian network models. These models typically construct fixed process correlation matrices based on historical data, and once the correlation strength is determined, it is no longer adjusted with changes in the production environment. For example, the correlation matrix M(i,j) used in traditional methods only represents the fixed probability or weight from process i to process j, which is difficult to reflect the dynamic characteristics of actual production.

[0102] Existing technologies primarily rely on offline data analysis and periodic manual parameter adjustments. Traditional methods involve collecting vast amounts of historical production data, using statistical methods to determine the correlation coefficients between processes, and then constructing a process propagation network model that remains constant throughout a production cycle. When the production environment changes or performance deteriorates, technicians need to recollect data and manually adjust the model parameters. This approach has the following drawbacks: first, the model is slow to respond to dynamic changes in the production environment; second, it lacks adaptive adjustment capabilities; third, it cannot capture complex nonlinear relationships between processes; and fourth, it is difficult to achieve system-level collaborative optimization.

[0103] This embodiment addresses the problems of static and rigid process association models and poor adaptability in existing technologies by introducing neural networks and information entropy theory to achieve dynamic adaptive optimization of process association relationships. First, a regulatory neural network is introduced based on the traditional process propagation matrix, enabling each process node to have autonomous perception and response capabilities. Second, a ternary neuron structure including threshold judgment, state response, and feedback gain is designed to enhance the system's sensitivity to abnormal states. Third, an adaptive regulation mechanism based on local feedback is established, allowing the system to adjust the propagation intensity between processes in real time. Fourth, information entropy theory is introduced to assess the uncertainty of process characteristics, and entropy value correction improves the robustness of the model. Fifth, a dynamic mapping relationship between feature weights and neuron response parameters is constructed, realizing self-optimization adjustment of system parameters.

[0104] In one optional implementation, the process priority feature vector is mapped to a two-dimensional grid space, the activation values ​​of the grid nodes are calculated to form process groups, the equipment sharing degree of the process groups is calculated based on the production equipment information, the material crossover degree and process overlap degree of the process groups are calculated, a feature matrix is ​​constructed, and the process group conflict coefficient is obtained based on eigenvalue analysis, including:

[0105] The process priority feature vector is mapped to a two-dimensional grid space, and the activation value of the grid node in the two-dimensional grid space is calculated by the hyperbolic tangent function. The input of the hyperbolic tangent function includes the mapping weight coefficient and bias term of the process priority feature vector.

[0106] The Euclidean distance between the grid nodes is calculated based on the activation value of the grid nodes, and the processes corresponding to the grid nodes whose Euclidean distance is less than a preset distance threshold are divided into the same process group.

[0107] Based on the set of application equipment for each process pair in a process group, the ratio of the intersection cardinality to the union potential of the application equipment sets is calculated to obtain the equipment sharing degree between the process pairs.

[0108] Based on the material input and output information of the process pair, the weighted sum of the material input overlap and material output overlap is calculated as a ratio to the process group size to obtain the material crossover degree between the process pairs.

[0109] Based on the process path information of the process pairs, the ratio of the intersection potential to the union potential of the process path sets is calculated to obtain the process overlap between the process pairs.

[0110] A fusion feature matrix of equipment sharing degree, material crossover degree, and process overlap degree is constructed. Based on the eigenvalue analysis and feature fusion of the fusion feature matrix, the conflict coefficient of the process group is obtained.

[0111] like Figure 2 As shown, the method includes:

[0112] In one specific implementation, a process priority feature vector is obtained and mapped to a two-dimensional grid space. During the mapping process, let the process priority feature vector be V, the node coordinates in the two-dimensional grid space be (x, y), the mapping weight coefficient be W, and the bias term be b. For each node in the grid space, its activation value A(x, y) is calculated using the hyperbolic tangent function: A(x, y) = tanh(V·W+b). For example, for a certain process, its priority feature vector V is [0.8, 0.6, 0.4], the mapping weight coefficient W is [[0.3, 0.2], [0.5, 0.4], [0.1, 0.3]], and the bias term b is [0.1, 0.2]. Then, the activation value of this process on the two-dimensional grid node can be calculated using the aforementioned hyperbolic tangent function.

[0113] After calculating the activation values ​​of the grid nodes, calculate the Euclidean distance between the grid nodes. Let the activation values ​​of two grid nodes be A1 and A2, then the Euclidean distance between them is D = [(A1x - A2x)]. 2 +(A1y-A2y) 2 ] 1 / 2 The processes corresponding to grid nodes whose Euclidean distance is less than a preset distance threshold T are grouped into the same process group. For example, if the distance threshold T = 0.5, and the activation values ​​of the grid nodes corresponding to process 1 and process 2 are [0.6, 0.7] and [0.8, 0.9] respectively, the calculated Euclidean distance is 0.28, which is less than the threshold of 0.5. Therefore, process 1 and process 2 are grouped into the same process group.

[0114] After determining the process group, the equipment sharing degree between the process pairs is calculated based on the set of equipment used in each process pair within the process group. Let the set of equipment used in process i and process j be Si and Sj, respectively. Then the equipment sharing degree ESij = |Si∩Sj| / |Si∪Sj|, where |Si∩Sj| represents the cardinality of the intersection of the equipment sets, and |Si∪Sj| represents the cardinality of the union of the equipment sets. For example, if the set of equipment used in process 1 is {equipment A, equipment B, equipment C}, and the set of equipment used in process 2 is {equipment B, equipment C, equipment D}, then their intersection is {equipment B, equipment C}, and their union is {equipment A, equipment B, equipment C, equipment D}. The equipment sharing degree is 2 / 4 = 0.5.

[0115] Next, we calculate the material overlap between process pairs. Let the material input sets for process i and process j be Ii and Ij, respectively, and the material output sets be Oi and Oj, respectively. The input overlap is |Ii∩Ij| / |Ii∪Ij|, and the output overlap is |Oi∩Oj| / |Oi∪Oj|. The material overlap MCij = (α·input overlap + β·output overlap) / N, where α and β are weighting coefficients, and N is the process group size. For example, a process group contains 3 processes. The material input sets for process 1 and process 2 are {material 1, material 2} and {material 2, material 3}, respectively, and the output sets are {material 4, material 5} and {material 5, material 6}, respectively. Let α = 0.6 and β = 0.4, then the input overlap is 1 / 3 = 0.33, the output overlap is 1 / 3 = 0.33, and the material crossover is (0.6 × 0.33 + 0.4 × 0.33) / 3 = 0.066.

[0116] Calculate the process overlap between process pairs. Let the process path sets for process i and process j be Pi and Pj, respectively. Then the process overlap PRij = |Pi∩Pj| / |Pi∪Pj|. For example, if the process path set for process 1 is {path 1, path 2, path 3} and the process path set for process 2 is {path 2, path 3, path 4}, then the process overlap is 2 / 5 = 0.4.

[0117] Construct a fusion feature matrix M, where the matrix element Mij = w1·ESij + w2·MCij + w3·PRij, and w1, w2, and w3 are weight coefficients. For example, if w1 = 0.5, w2 = 0.3, w3 = 0.2, the equipment sharing degree between process 1 and process 2 is 0.5, the material overlap degree is 0.066, and the process overlap degree is 0.4, then the corresponding element value of the fusion feature matrix is ​​0.5×0.5 + 0.3×0.066 + 0.2×0.4 = 0.35.

[0118] Perform eigenvalue analysis on the fused feature matrix. Calculate the maximum eigenvalue λmax of the matrix and its corresponding eigenvector V. The conflict coefficient CF of the process group is λmax / n, where n is the matrix dimension. In practical applications, if a process group contains 4 processes, and the calculated maximum eigenvalue of the constructed 4×4 fused feature matrix is ​​3.2, then the conflict coefficient is 3.2 / 4 = 0.8.

[0119] This method can effectively identify conflict relationships between processes in a production system, providing a basis for production scheduling optimization. For example, in a certain manufacturing system, the conflict coefficients of three process groups calculated by this method are 0.8, 0.5, and 0.3, respectively. This indicates that the first process group has the highest degree of internal conflict, and the coordination of processes within this process group should be prioritized during scheduling to reduce resource competition and production bottlenecks.

[0120] Through the above steps, this invention realizes a method for mapping process priority feature vectors to a two-dimensional grid space and calculating the conflict coefficient of process groups, providing technical support for production scheduling optimization in complex manufacturing environments. This method considers the influencing factors of three dimensions: equipment sharing, material crossover, and process overlap, comprehensively assessing the degree of conflict within process groups, and has high practical value.

[0121] In one optional implementation, a fusion feature matrix of equipment sharing, material overlap, and process overlap is constructed. Based on the eigenvalue analysis and feature fusion of the fusion feature matrix, the process group conflict coefficient is obtained, including:

[0122] The equipment sharing degree in the process group is mapped to the first quaternion. Based on the value of the equipment sharing degree, the real part of the first quaternion is determined, the fluctuation value of the equipment sharing degree is calculated, the imaginary part in the first direction is determined, and the remaining imaginary parts are zero.

[0123] The material crossover degree is mapped to a second quaternion. Based on the value of the material crossover degree, the real part of the second quaternion is determined, the fluctuation value of the material crossover degree is calculated, the imaginary part of the second direction is determined, and the rest of the imaginary parts are zero.

[0124] The process overlap is mapped to a third quaternion. Based on the value of the process overlap, the real part of the third quaternion is determined, the fluctuation value of the process overlap is calculated, the third-directed imaginary part is determined, and the remaining imaginary parts are zero.

[0125] A fusion feature matrix is ​​constructed based on the first quaternion, the second quaternion, and the third quaternion. The diagonal elements of the fusion feature matrix are the first quaternion, and the off-diagonal elements are composed of alternating arrangements of the second quaternion and the third quaternion.

[0126] The characteristic equation of the fused feature matrix is ​​solved to obtain the eigenvalues. The eigenvalues ​​are normalized and then projected into a four-dimensional space. The projection result is then weighted and fused with the original features to obtain the process group conflict coefficient.

[0127] In one specific implementation, in a production scheduling system, there are multi-dimensional conflict relationships between work groups, including equipment sharing, material overlap, and process overlap. This embodiment maps these conflict relationships to a quaternion space and analyzes the conflict characteristics of work groups by fusing feature matrices.

[0128] For the equipment sharing degree within a process group, this embodiment maps it to a first quaternion. Assuming the equipment sharing degree of a certain process group is 0.75, it means that 75% of the equipment in this process group is shared by other process groups. This value is taken as the real part of the first quaternion, i.e., 0.75. The fluctuation value of the equipment sharing degree is calculated, reflecting the changes in the degree of equipment sharing. For example, in the past 5 production cycles, the equipment sharing degree of this process group was 0.70, 0.72, 0.75, 0.76, and 0.77, respectively. Its standard deviation is calculated to be 0.028. This fluctuation value is set as the imaginary part in the first direction, and the other two imaginary parts are set to zero. Therefore, the first quaternion is represented as (0.75, 0.028, 0, 0).

[0129] For material crossover, this embodiment maps it to a second quaternion. Assuming the material crossover of a process group is 0.45, it means that 45% of the materials in this process group are used interchangeably with those in other process groups. This value is taken as the real part of the second quaternion, i.e., 0.45. The fluctuation value of the material crossover is calculated. In the past 5 production cycles, the material crossover of this process group was 0.40, 0.42, 0.45, 0.47, and 0.46, respectively. Its standard deviation is calculated to be 0.029. This fluctuation value is set as the imaginary part in the second direction, and the other two imaginary parts are set to zero. Therefore, the second quaternion is represented as (0.45, 0, 0.029, 0).

[0130] For process overlap, this embodiment maps it to a third quaternion. Assuming the process overlap of a process group is 0.60, it means that 60% of the process flow in this process group overlaps with other process groups. This value is taken as the real part of the third quaternion, i.e., 0.60. The fluctuation value of the process overlap is calculated. In the past 5 production cycles, the process overlap of this process group was 0.58, 0.59, 0.60, 0.61, and 0.62, respectively. Its standard deviation is calculated to be 0.016. This fluctuation value is set as the third directed imaginary part, and the other two imaginary parts are set to zero. Therefore, the third quaternion is represented as (0.60, 0, 0, 0.016).

[0131] Based on the three quaternions mentioned above, this embodiment constructs a fusion feature matrix. This matrix is ​​a 3×3 matrix, with the diagonal elements all being the first quaternion (0.75, 0.028, 0, 0), and the off-diagonal elements consisting of alternating arrangements of the second quaternion (0.45, 0, 0.029, 0) and the third quaternion (0.60, 0, 0, 0.016). The constructed fusion feature matrix is ​​as follows: the (1,1)th element is (0.75, 0.028, 0, 0), the (1,2)th element is (0.45, 0, 0.029, 0), the (1,3)th element is (0.60, 0, 0, 0.016), the (2,1)th element is (0.60, 0, 0, 0.016), the (2,2)th element is (0.75, 0.028, 0, 0), the (2,3)th element is (0.45, 0, 0.029, 0), the (3,1)th element is (0.45, 0, 0.029, 0), the (3,2)th element is (0.60, 0, 0, 0.016), and the (3,3)th element is (0.75, 0.028, 0, 0).

[0132] The eigenvalues ​​are obtained by solving the characteristic equation of the fused characteristic matrix. Since the matrix elements are quaternions, the eigenvalue calculation needs to be performed in the quaternion domain. The specific calculation methods include solving the quaternion determinant and finding the roots of the characteristic equation. In this example, three eigenvalues ​​are obtained through iterative calculation: λ1 = (1.42, 0.035, 0.022, 0.012), λ2 = (0.62, 0.018, 0.031, 0.009), and λ3 = (0.21, 0.003, 0.005, 0.007).

[0133] These eigenvalues ​​are normalized by dividing each eigenvalue by the sum of the moduli of all eigenvalues. The moduli of eigenvalue λ1 are 1.42, λ2 are 0.62, and λ3 are 0.21, with a sum of 2.25. Therefore, the normalized eigenvalues ​​are: λ1' = (0.631, 0.016, 0.010, 0.005), λ2' = (0.276, 0.008, 0.014, 0.004), λ3' = (0.093, 0.001, 0.002, 0.003).

[0134] The normalized eigenvalues ​​are projected onto a four-dimensional space, resulting in four-dimensional vectors (0.631, 0.016, 0.010, 0.005), (0.276, 0.008, 0.014, 0.004), and (0.093, 0.001, 0.002, 0.003).

[0135] The projected results are weighted and fused with the original features to obtain the process group conflict coefficient. The weighting coefficients are set as α = 0.7, β = 0.3, and the original features are the average of equipment sharing (0.75), material overlap (0.45), and process overlap (0.60), with a value of 0.60. The weighted fusion calculation is: Conflict coefficient = α × principal component of the projected result + β × average value of the original features = 0.7 × 0.631 + 0.3 × 0.60 = 0.622.

[0136] The conflict coefficient represents the overall degree of conflict between work groups. The higher the conflict coefficient, the more serious the conflict between work groups, and the more careful the allocation of related resources and the sequencing of work processes need to be in production scheduling.

[0137] In existing technologies, planar matrices or simple vector models are typically used to characterize the conflict relationships between process groups, such as Petri net models, conflict graph models, and binary relation matrices. These traditional methods mainly focus on single-dimensional conflict relationships, such as considering only equipment conflicts or analyzing only material conflicts, and are difficult to comprehensively reflect the complex interrelationships of multi-dimensional conflicts.

[0138] Existing technologies primarily rely on constructing two-dimensional conflict matrices. Matrix elements are typically binary (0-1) or simple scalar values, representing the existence and degree of conflict between process groups. For example, traditional methods usually establish independent conflict matrices for each conflict type (equipment sharing, material crossing, process overlap). The equipment sharing conflict degree between process group i and process group j is denoted as SD(i,j), with a value range of [0,1]; the material crossing conflict degree is denoted as MC(i,j); and the process overlap conflict degree is denoted as PO(i,j). These independent conflict indices are then combined using a simple linear weighting method, such as Total(i,j) = w1 × SD(i,j) + w2 × MC(i,j) + w3 × PO(i,j), where w1, w2, and w3 are weighting coefficients. This approach has drawbacks: it struggles to capture the intrinsic correlation between conflict indices and ignores their dynamic fluctuation characteristics; the linear weighting method is overly simplistic and cannot express the nonlinear interactions of multi-dimensional conflicts; and it lacks mathematical rigor, making in-depth feature analysis difficult.

[0139] This embodiment addresses the shortcomings of existing technologies in analyzing multidimensional conflicts within work processes by introducing quaternion theory and matrix feature analysis to achieve a unified expression and deep feature extraction of multidimensional conflict relationships. Quaternions are used to represent multidimensional conflict indices, simultaneously encoding the degree of conflict and its volatility in the real and imaginary parts of the quaternions, thus enriching the dimensions of conflict representation. A quaternion-based fusion feature matrix is ​​constructed, integrating conflict indices from different dimensions within a unified mathematical framework. Through solving characteristic equations and eigenvalue analysis, the intrinsic characteristics of work process group conflicts are mathematically extracted. A feature projection and weighted fusion mechanism is designed to preserve original feature information while extracting deeper conflict features.

[0140] In one optional implementation, the expected revenue value of the process group is calculated based on the conflict coefficient and order information. A marginal revenue decay factor is introduced into the expected revenue value to obtain the actual revenue value. Game optimization iteration is performed based on the actual revenue value. When the revenue increment between two adjacent iterations is less than a preset convergence threshold, the optimal scheduling scheme is output, including:

[0141] For each process group, the conflict coefficient of the process group is multiplied by the weight coefficient and value coefficient of the corresponding order and then summed to calculate the expected revenue value of the process group.

[0142] Obtain the current resource quantity allocated to the process group, calculate the resource ratio of the process group, set a benchmark attenuation coefficient according to the importance of the resource type, and when the resource ratio exceeds a preset occupancy threshold, multiply the benchmark attenuation coefficient by the proportion exceeding the preset occupancy threshold to obtain the marginal attenuation coefficient. Use the product of the marginal attenuation coefficient and the resource ratio as the independent variable of the exponential function to calculate the marginal revenue attenuation factor.

[0143] Multiply the expected revenue value of the process group by the marginal revenue decay factor to obtain the actual revenue value of the process group;

[0144] Calculate the degree of process connection between process groups, which is determined based on the sequential connection relationship between processes in the process group; construct a competition adjustment coefficient between process groups based on the degree of process connection.

[0145] Construct a payment matrix by multiplying the actual revenue values ​​of any two process groups by their corresponding competition adjustment coefficients;

[0146] Dynamic game optimization is performed based on the payoff matrix. The scheduling scheme of each process group is iteratively updated until the profit increment between two adjacent iterations is less than the preset convergence threshold, and the optimal scheduling scheme is output.

[0147] In one specific implementation, the expected revenue value is calculated for each process group. Taking a production system as an example, assume there are three process groups G1, G2, and G3, and two orders O1 and O2. The conflict coefficients of process group G1 with respect to orders O1 and O2 are 0.8 and 0.5, respectively; the weight coefficient of order O1 is 0.7, and its value coefficient is 80; the weight coefficient of order O2 is 0.5, and its value coefficient is 60. Based on these parameters, the expected revenue value of G1 is calculated as: 0.8 × 0.7 × 80 + 0.5 × 0.5 × 60 = 59.5. Similarly, the expected revenue value of G2 is calculated to be 45.2, and the expected revenue value of G3 is 52.8.

[0148] A marginal revenue decay factor is introduced to adjust the expected revenue value of the process group. Assume there are three resource types R1, R2, and R3 in the system, with baseline decay coefficients of 0.3, 0.2, and 0.1 respectively, and a preset occupancy threshold of 0.4 for each. Taking process group G1 as an example, the currently allocated quantities of R1, R2, and R3 resources are 12, 8, and 5 respectively, and the total system resources are 20, 20, and 20 respectively. Therefore, the resource proportions of G1 are 0.6, 0.4, and 0.25 respectively. Since the R1 resource proportion of 0.6 exceeds the preset occupancy threshold of 0.4, the excess is (0.6-0.4) / 0.4 = 0.5, so the marginal revenue decay coefficient of R1 is 0.3 × 0.5 = 0.15. The R2 resource proportion is exactly equal to the preset occupancy threshold, and the R3 resource proportion is lower than the preset occupancy threshold; therefore, their marginal revenue decay coefficients are all 0. Substituting the product of the marginal revenue decay coefficient and the resource proportion into the exponential function, the marginal revenue decay factor of R1 is calculated as e. (-0.15×0.6) =0.914, and the marginal revenue decay factors of R2 and R3 are both 1. The minimum value of the marginal revenue decay factors of the three resources is taken as the comprehensive marginal revenue decay factor of G1, which is 0.914. Multiplying the expected revenue of G1 by this decay factor, we get the actual revenue of G1 as 59.5 × 0.914 = 54.4.

[0149] Similarly, the actual profit values ​​of G2 and G3 were calculated to be 43.1 and 52.8, respectively.

[0150] To consider the connection between process groups, it is necessary to calculate the degree of connection between them. Assume G1 is a prerequisite process for G2, with a connection degree of 0.8; G2 is a prerequisite process for G3, with a connection degree of 0.6; G1 and G3 have no direct connection, so their connection degree is 0. Based on the connection degree, a competition adjustment coefficient is constructed between process groups, which is proportional to the connection degree. For example, the competition adjustment coefficient between G1 and G2 is 1.5, between G2 and G3 is 1.2, and between G1 and G3 is 1.0.

[0151] Construct the payment matrix. Multiply the actual revenue values ​​of any two process groups by their corresponding competition adjustment coefficients to obtain the elements of the payment matrix. For example, the elements of the payment matrix between G1 and G2 are (54.4×1.5, 43.1×1.5) = (81.6, 64.7), indicating that when G1 and G2 compete simultaneously, the payment value of G1 is 81.6 and the payment value of G2 is 64.7. Similarly, the elements of the payment matrix between G2 and G3 can be calculated as (43.1×1.2, 52.8×1.2) = (51.7, 63.4), and the elements of the payment matrix between G1 and G3 are (54.4×1.0, 52.8×1.0) = (54.4, 52.8).

[0152] Dynamic game optimization iterations are performed based on the payoff matrix. Initially, it is assumed that the resource allocation ratio for each of the three work groups is 1 / 3 of the total system resources. After the first iteration, the resource allocation is adjusted according to the payoff matrix, with the resource allocation ratios for G1, G2, and G3 adjusted to 0.4, 0.25, and 0.35, respectively. The adjusted total system revenue is calculated to be 54.4 × 0.4 + 43.1 × 0.25 + 52.8 × 0.35 = 50.9, an increase of 2.6 compared to the initial revenue of 48.3.

[0153] After the second iteration, the resource allocation ratios for the process groups were further adjusted to 0.45, 0.2, and 0.35, increasing the total system revenue to 51.7, with a revenue increment of 0.8. After the third iteration, the resource allocation ratios were adjusted to 0.47, 0.18, and 0.35, resulting in a total system revenue of 52.0 and a revenue increment of 0.3.

[0154] Assuming the preset convergence threshold is 0.5, the iteration stops because the profit increment of 0.3 in the third iteration is less than the preset convergence threshold of 0.5, and the optimal scheduling scheme is output: the resource allocation ratios of process groups G1, G2 and G3 are 0.47, 0.18 and 0.35, respectively.

[0155] This dynamic game-theoretic scheduling method, based on the conflict coefficient of work groups and the decay of marginal revenue, can effectively balance resource competition among multiple orders and improve the overall system revenue. Experimental results show that, compared with traditional scheduling methods, this method can increase the throughput of the production system and shorten the average order completion time by approximately [percentage missing], making it particularly suitable for multi-order collaborative production scheduling problems in complex manufacturing environments.

[0156] In one optional implementation, dynamic game optimization is performed based on the payoff matrix, iteratively updating the scheduling scheme of each process group until the payoff increment between two adjacent iterations is less than a preset convergence threshold, and the optimal scheduling scheme is output, including:

[0157] Using the priority order and resource requirements of the work groups as input parameters, the payment matrix is ​​constructed in blocks to generate an initial resource allocation and scheduling scheme.

[0158] Each process group is traversed sequentially, and the game payoff of the process group with other process groups is calculated based on the payoff matrix to obtain the optimal response scheduling scheme that maximizes the payoff.

[0159] Establish a historical revenue path feature library for process groups, extract the temporal patterns of revenue changes, and generate the scheduling scheme distribution for other process groups in the next iteration;

[0160] Adjust the response scheduling scheme of the current work group according to the distribution of the scheduling scheme, and generate an adaptive step size factor by combining the change gradient of the revenue value of the work group and the competition intensity between the work groups.

[0161] The current scheduling scheme of the work group is updated towards the adjusted response scheduling scheme according to the adaptive step size factor to obtain the updated scheduling scheme; the updated scheduling scheme that exceeds the resource capacity constraint is projected into the feasible strategy space.

[0162] The revenue value of each process group is recalculated based on the feasible strategy space, the revenue increment between two adjacent iterations is obtained, and the historical revenue path feature library is updated at the same time.

[0163] The convergence threshold is dynamically adjusted using a logarithmic function of the number of iterations. The optimal scheduling scheme is determined when the profit increment of all process groups is less than the corresponding convergence threshold; otherwise, the process is re-executed.

[0164] In one specific implementation, the priority order of the work groups and their resource requirements are used as input parameters to construct the payoff matrix in blocks, generating an initial resource allocation and scheduling scheme. Specifically, assume there are 5 work groups in the production system, with priorities of 0.3, 0.25, 0.2, 0.15, and 0.1, and resource requirements of 10, 8, 6, 5, and 4 units, respectively. The payoff matrix is ​​divided into 5×5 sub-matrices, each representing a game relationship between a pair of work groups. For example, the sub-matrix between work group 1 and work group 2 is a 4×3 matrix, indicating that work group 1 has 4 scheduling schemes and work group 2 has 3 scheduling schemes. Based on the priority order, the initial scheduling scheme allocates 5 resource units to work group 1, 4 resource units to work group 2, 3 resource units to work group 3, and 2 resource units each to work groups 4 and 5.

[0165] Each process group is iterated sequentially, and the game payoff between the process group and other process groups is calculated based on the payoff matrix to obtain the optimal response scheduling scheme corresponding to maximizing the payoff. Taking process group 1 as an example, the game payoff between it and process groups 2, 3, 4, and 5 is calculated. Assume that process group 1 has 4 resource allocation schemes: allocating 3, 4, 5, or 6 resource units. For each scheme, the payoff value of process group 1 is calculated through the payoff matrix. In the payoff matrix, the element value(i,j) represents the payoff obtained by process group 1 when process group 1 chooses scheme i and process group 2 chooses scheme j. The specific calculation formula is: value(i,j)=α·P1·R1(i)-β·C(R1(i),R2(j)), where P1 is the priority of process group 1 (0.3), R1(i) is the resource allocation amount corresponding to scheme i, C(R1(i),R2(j)) is the conflict cost between the two process group resource allocation schemes, and α and β are the payoff coefficient (2.5) and conflict penalty coefficient (1.2), respectively. For example, when process group 1 chooses to allocate 5 resource units (Scheme 3) and process group 2 chooses to allocate 4 resource units (Scheme 2), the calculated value(3,2) = 2.5 × 0.3 × 5 - 1.2 × 0.4 = 3.75 - 0.48 = 3.27. Considering the game payoff of process group 1 with all other process groups, when allocating 5 resource units, process group 1 obtains a total payoff of 25 units. Therefore, this scheme is selected as the optimal response scheduling scheme.

[0166] A historical revenue path feature library is established for each work group to extract the temporal patterns of revenue changes and generate the scheduling scheme distribution for other work groups in the next iteration. The historical revenue path feature library records the scheduling schemes and corresponding revenue values ​​for each work group in past iterations. For example, work group 2 selected allocations of 3, 4, and 4 resource units in the first three iterations, with corresponding revenue values ​​of 18, 22, and 23 units. By analyzing this data, it is extracted that work group 2 tends to choose a scheduling scheme with a resource allocation of 4 units, and it is predicted that in the next iteration, it has a 70% probability of choosing to allocate 4 resource units, a 20% probability of choosing to allocate 5 resource units, and a 10% probability of choosing to allocate 3 resource units.

[0167] The response scheduling scheme for the current work group is adjusted based on the distribution of scheduling schemes. An adaptive step size factor is generated by combining the gradient of the work group's revenue change and the competition intensity between work groups. For work group 1, considering the predicted scheduling scheme distribution of other work groups, its optimal response scheduling scheme is adjusted. Assuming work group 1 is currently allocated 5 resource units, but considering the increased resource demand of work group 2, the adjusted optimal response is to allocate 6 resource units. The revenue change gradient of work group 1 is +3 units / resource unit, and its competition intensity with work group 2 is 0.8. Based on this, the adaptive step size factor is calculated to be 0.6.

[0168] The current scheduling scheme for each work group is updated towards the adjusted response scheduling scheme using an adaptive step factor, resulting in an updated scheduling scheme. Updated scheduling schemes exceeding resource capacity constraints are then projected into the feasible policy space. The current scheduling scheme for work group 1 allocates 5 resource units, the adjusted response scheduling scheme allocates 6 resource units, and with a step factor of 0.6, the updated scheduling scheme allocates 5.6 resource units. However, since resource units must be integers, they are rounded down to 5 resource units. If the total system resource capacity is 20 units, the total demand for all work groups after the update is 22 units, exceeding the capacity constraint, requiring projection. Following priority, the allocation for work group 5 is reduced from 2 to 1 resource unit, and the allocation for work group 4 is reduced from 3 to 2 resource units, reducing the total demand to 20 units, satisfying the capacity constraint.

[0169] The revenue values ​​for each process group are recalculated based on the feasible strategy space, and the revenue increment between two adjacent iterations is obtained. Simultaneously, the historical revenue path feature library is updated. After adjustment, the revenue of process group 1 is 25 units, process group 2 is 23 units, process group 3 is 20 units, process group 4 is 15 units, and process group 5 is 8 units. Compared to the previous iteration, the revenue increment for process group 1 is 0 units, for process group 2 it is 1 unit, for process group 3 it is 2 units, for process group 4 it is -1 unit, and for process group 5 it is -2 units. This information is then updated in the historical revenue path feature library.

[0170] The convergence threshold is dynamically adjusted using a logarithmic function of the number of iterations. The optimal scheduling scheme is determined when the revenue increment of all process groups is less than the corresponding convergence threshold; otherwise, the process is restarted. Assuming the current iteration is the 10th iteration, the convergence threshold is set to 2 / log(10)≈0.87 units of revenue. Checking the revenue increment of each process group, it is found that the revenue increments of process group 2 and process group 3 are 1 and 2 units respectively, which are greater than the convergence threshold of 0.87 units, so it is necessary to continue iterating. After multiple iterations, in the 15th iteration, the revenue increment of all process groups is less than the convergence threshold of 2 / log(15)≈0.76 units. At this time, the optimal scheduling scheme is determined as follows: process group 1 is allocated 6 resource units, process group 2 is allocated 5 resource units, process group 3 is allocated 4 resource units, process group 4 is allocated 3 resource units, and process group 5 is allocated 2 resource units, for a total of 20 resource units. The revenue of each process group is 27, 25, 22, 17 and 10 units respectively.

[0171] Existing technologies typically employ centralized optimization methods based on mathematical programming, such as linear programming, integer programming, and heuristic algorithms; or simple allocation strategies based on priority rules, such as Highest Priority First (HPF) and Shortest Processing Time First (SPT). These traditional methods have limitations in handling dynamic resource allocation problems in complex production environments.

[0172] Existing technologies typically rely on establishing a global optimization model or using fixed allocation rules. Centralized optimization methods construct objective functions and constraints, such as MaxΣ(w i ·U i ), where w i U represents the weight of process group i. i The method involves optimizing resource utilization and then using a solver to find the optimal solution. However, this approach has high computational complexity and struggles to handle large-scale real-time scheduling problems. Rule-based methods allocate resources directly based on preset priorities or resource demands, such as first satisfying all demands of high-priority process groups and then sequentially satisfying other process groups. However, this method is overly simplistic and struggles to balance the interests of different process groups. Existing technologies suffer from the following drawbacks: they ignore the strategic interactions between process groups; they are difficult to adapt to dynamically changing production environments; they cannot effectively capture historical behavior patterns of process groups; and they lack adaptive adjustment mechanisms, resulting in low convergence efficiency.

[0173] This embodiment addresses the lack of strategic interaction and dynamic adaptability in existing resource allocation methods by introducing a game theory framework and an adaptive learning mechanism to achieve collaborative competitive optimization among process groups. It introduces a block payoff matrix, transforming the resource allocation problem into a multi-process group strategic game, more accurately describing the conflicts of interest and collaborative relationships among process groups. A historical payoff path feature library is designed to improve the foresight of resource allocation by recording and analyzing the historical decision-making patterns of process groups. An adaptive step-size factor mechanism is proposed to dynamically adjust the update rate of resource allocation based on payoff gradients and competition intensity, balancing exploration and utilization. A dynamic convergence threshold based on the number of iterations is designed to improve the convergence efficiency and stability of the algorithm. A feasible strategy space projection mechanism is introduced to ensure that the resource allocation scheme always meets capacity constraints.

[0174] The intelligent APS scheduling system based on a multi-source heterogeneous ICS platform, as described in this embodiment of the invention, includes:

[0175] The first unit is used to obtain production equipment information, process route information, and order information through the multi-source heterogeneous ICS platform;

[0176] The second unit is used to generate a process propagation matrix based on the process route information, extract process correlations, calculate the corresponding information entropy weights, and perform tensor decomposition on the process propagation matrix to obtain the process priority feature vector.

[0177] The third unit is used to map the process priority feature vector to a two-dimensional grid space, calculate the activation value of the grid node to form a process group, calculate the equipment sharing degree of the process group according to the production equipment information, calculate the material crossover degree and process overlap degree of the process group, construct a feature matrix, and obtain the process group conflict coefficient based on the feature value analysis.

[0178] The fourth unit is used to calculate the expected revenue value of the process group based on the conflict coefficient of the process group and the order information. The marginal revenue decay factor is introduced into the expected revenue value to obtain the actual revenue value. Game optimization iteration is performed based on the actual revenue value. When the revenue increment between two adjacent iterations is less than the preset convergence threshold, the optimal scheduling scheme is output.

[0179] The fifth unit is used to convert the optimal scheduling scheme into a sequence of scheduling instructions and issue them for execution.

[0180] A third aspect of the present invention provides an electronic device, comprising:

[0181] processor;

[0182] Memory used to store processor-executable instructions;

[0183] The processor is configured to invoke instructions stored in the memory to execute the aforementioned method.

[0184] A fourth aspect of the present invention provides a computer-readable storage medium having stored thereon computer program instructions that, when executed by a processor, implement the aforementioned method.

[0185] This invention can be a method, apparatus, system, and / or computer program product. The computer program product may include a computer-readable storage medium having computer-readable program instructions loaded thereon for performing various aspects of the invention.

[0186] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. An intelligent APS scheduling method based on a multi-source heterogeneous ICS platform, characterized in that, include: Obtain production equipment information, process route information, and order information through a multi-source heterogeneous ICS platform; The process propagation matrix is ​​generated based on the process route information, the process correlation is extracted, the corresponding information entropy weights are calculated, and the process propagation matrix is ​​decomposed into tensor vectors to obtain the process priority feature vectors. The process priority feature vector is mapped to a two-dimensional grid space, the activation value of the grid node is calculated to form a process group, the equipment sharing degree of the process group is calculated according to the production equipment information, the material crossover degree and process overlap degree of the process group are calculated, a feature matrix is ​​constructed and the process group conflict coefficient is obtained based on the feature value analysis. The expected revenue value of the process group is calculated based on the conflict coefficient of the process group and the order information. The marginal revenue decay factor is introduced into the expected revenue value to obtain the actual revenue value. Game optimization iteration is performed based on the actual revenue value. When the revenue increment between two adjacent iterations is less than the preset convergence threshold, the optimal scheduling scheme is output. The optimal scheduling scheme is converted into a sequence of scheduling instructions and then issued for execution.

2. The method according to claim 1, characterized in that, Based on the process route information, a process propagation matrix is ​​generated, process relationships are extracted, corresponding information entropy weights are calculated, and tensor decomposition of the process propagation matrix is ​​performed to obtain the process priority feature vector, including: Construct the process topology based on the process route information, calculate the propagation strength between process nodes, and generate the initial propagation matrix; A regulatory neural network is constructed in the initial propagation matrix to adaptively regulate the propagation intensity of process nodes; Based on the adjusted propagation matrix, process correlation characteristics are calculated, and a process characteristic evaluation matrix is ​​constructed. Calculate the local information entropy of the process feature evaluation matrix, generate feature weights, and feed them back to the regulatory neural network to obtain the optimized propagation matrix; The optimized propagation matrix is ​​constructed as a feature tensor, and the Tucker decomposition method is used to perform feature decomposition on the feature tensor. Based on the feature contribution, the process priority feature vector is extracted.

3. The method according to claim 2, characterized in that, Constructing a regulatory neural network within the initial propagation matrix to adaptively regulate the propagation intensity of process nodes includes: A control neuron is configured at each process node position of the initial propagation matrix. Each control neuron contains a threshold judgment unit, a state response unit, and a feedback gain unit. A neural network topology is constructed among the control neurons. The threshold judgment unit monitors the process operation status and outputs judgment signals. The status response unit calculates the control factor based on the judgment signal. The control factor is amplified by the feedback gain unit to form a control command. Based on the control command, the propagation intensity of the corresponding process node in the initial propagation matrix is ​​updated. Local feedback pathways are established between adjacent control neurons to form an adaptive control network. Based on the output of the adaptive control network, the process association characteristics are calculated, including node propagation direction coefficient, node response strength and node hierarchical relationship, and a process characteristic evaluation matrix is ​​constructed. Calculate the local information entropy of the process feature evaluation matrix, combine it with the operating parameters of the control neurons to generate entropy correction coefficients, and correct the local information entropy to obtain feature weights; The feature weights are input into the state response unit of the regulating neuron, and the response parameters are adjusted based on the feature weights to establish a dynamic mapping relationship between the feature weights and the response parameters, thereby generating an optimized propagation matrix.

4. The method according to claim 1, characterized in that, The process priority feature vector is mapped to a two-dimensional grid space, and the activation values ​​of the grid nodes are calculated to form process groups. The equipment sharing degree of the process groups is calculated based on the production equipment information. The material overlap and process overlap of the process groups are also calculated. A feature matrix is ​​constructed, and the process group conflict coefficient is obtained based on eigenvalue analysis, including: The process priority feature vector is mapped to a two-dimensional grid space, and the activation value of the grid node in the two-dimensional grid space is calculated by the hyperbolic tangent function. The input of the hyperbolic tangent function includes the mapping weight coefficient and bias term of the process priority feature vector. The Euclidean distance between the grid nodes is calculated based on the activation value of the grid nodes, and the processes corresponding to the grid nodes whose Euclidean distance is less than a preset distance threshold are divided into the same process group. Based on the set of application equipment for each process pair in a process group, the ratio of the intersection cardinality to the union potential of the application equipment sets is calculated to obtain the equipment sharing degree between the process pairs. Based on the material input and output information of the process pair, the weighted sum of the material input overlap and material output overlap is calculated as a ratio to the process group size to obtain the material crossover degree between the process pairs. Based on the process path information of the process pairs, the ratio of the intersection potential to the union potential of the process path sets is calculated to obtain the process overlap between the process pairs. A fusion feature matrix of equipment sharing degree, material crossover degree, and process overlap degree is constructed. Based on the eigenvalue analysis and feature fusion of the fusion feature matrix, the conflict coefficient of the process group is obtained.

5. The method according to claim 4, characterized in that, A fusion feature matrix is ​​constructed to represent equipment sharing, material overlap, and process overlap. Based on eigenvalue analysis and feature fusion of the fusion feature matrix, the process group conflict coefficient is obtained, including: The equipment sharing degree in the process group is mapped to the first quaternion. Based on the value of the equipment sharing degree, the real part of the first quaternion is determined, the fluctuation value of the equipment sharing degree is calculated, the imaginary part in the first direction is determined, and the remaining imaginary parts are zero. The material crossover degree is mapped to a second quaternion. Based on the value of the material crossover degree, the real part of the second quaternion is determined, the fluctuation value of the material crossover degree is calculated, the imaginary part of the second direction is determined, and the rest of the imaginary parts are zero. The process overlap is mapped to a third quaternion. Based on the value of the process overlap, the real part of the third quaternion is determined, the fluctuation value of the process overlap is calculated, the third-directed imaginary part is determined, and the remaining imaginary parts are zero. A fusion feature matrix is ​​constructed based on the first quaternion, the second quaternion, and the third quaternion. The diagonal elements of the fusion feature matrix are the first quaternion, and the off-diagonal elements are composed of alternating arrangements of the second quaternion and the third quaternion. The characteristic equation of the fused feature matrix is ​​solved to obtain the eigenvalues. The eigenvalues ​​are normalized and then projected into a four-dimensional space. The projection result is then weighted and fused with the original features to obtain the process group conflict coefficient.

6. The method according to claim 1, characterized in that, The expected revenue value of the process group is calculated based on the conflict coefficient and order information. A marginal revenue decay factor is introduced into the expected revenue value to obtain the actual revenue value. Game-theoretic optimization iterations are performed based on the actual revenue value. When the revenue increment between two adjacent iterations is less than a preset convergence threshold, the optimal scheduling scheme is output, including: For each process group, the conflict coefficient of the process group is multiplied by the weight coefficient and value coefficient of the corresponding order and then summed to calculate the expected revenue value of the process group. Obtain the current resource quantity allocated to the process group, calculate the resource ratio of the process group, set a benchmark attenuation coefficient according to the importance of the resource type, and when the resource ratio exceeds a preset occupancy threshold, multiply the benchmark attenuation coefficient by the proportion exceeding the preset occupancy threshold to obtain the marginal attenuation coefficient. Use the product of the marginal attenuation coefficient and the resource ratio as the independent variable of the exponential function to calculate the marginal revenue attenuation factor. Multiply the expected revenue value of the process group by the marginal revenue decay factor to obtain the actual revenue value of the process group; Calculate the degree of process connection between process groups, which is determined based on the sequential connection relationship between processes in the process group; construct a competition adjustment coefficient between process groups based on the degree of process connection. Construct a payment matrix by multiplying the actual revenue values ​​of any two process groups by their corresponding competition adjustment coefficients; Dynamic game optimization is performed based on the payoff matrix. The scheduling scheme of each process group is iteratively updated until the profit increment between two adjacent iterations is less than the preset convergence threshold, and the optimal scheduling scheme is output.

7. The method according to claim 6, characterized in that, Dynamic game optimization is performed based on the payoff matrix. The scheduling scheme of each process group is iteratively updated until the payoff increment between two adjacent iterations is less than a preset convergence threshold. The optimal scheduling scheme is output as follows: Using the priority order and resource requirements of the work groups as input parameters, the payment matrix is ​​constructed in blocks to generate an initial resource allocation and scheduling scheme. Each process group is traversed sequentially, and the game payoff of the process group with other process groups is calculated based on the payoff matrix to obtain the optimal response scheduling scheme that maximizes the payoff. Establish a historical revenue path feature library for process groups, extract the temporal patterns of revenue changes, and generate the scheduling scheme distribution for other process groups in the next iteration; Adjust the response scheduling scheme of the current work group according to the distribution of the scheduling scheme, and generate an adaptive step size factor by combining the change gradient of the revenue value of the work group and the competition intensity between the work groups. The current scheduling scheme of the work group is updated towards the adjusted response scheduling scheme according to the adaptive step size factor to obtain the updated scheduling scheme; the updated scheduling scheme that exceeds the resource capacity constraint is projected into the feasible strategy space. The revenue value of each process group is recalculated based on the feasible strategy space, the revenue increment between two adjacent iterations is obtained, and the historical revenue path feature library is updated at the same time. The convergence threshold is dynamically adjusted using a logarithmic function of the number of iterations. The optimal scheduling scheme is determined when the profit increment of all process groups is less than the corresponding convergence threshold; otherwise, the process is re-executed.

8. An intelligent APS scheduling system based on a multi-source heterogeneous ICS platform, used to implement the method of any one of claims 1-7, characterized in that, include: The first unit is used to obtain production equipment information, process route information, and order information through the multi-source heterogeneous ICS platform; The second unit is used to generate a process propagation matrix based on the process route information, extract process correlations, calculate the corresponding information entropy weights, and perform tensor decomposition on the process propagation matrix to obtain the process priority feature vector. The third unit is used to map the process priority feature vector to a two-dimensional grid space, calculate the activation value of the grid node to form a process group, calculate the equipment sharing degree of the process group according to the production equipment information, calculate the material crossover degree and process overlap degree of the process group, construct a feature matrix, and obtain the process group conflict coefficient based on the feature value analysis. The fourth unit is used to calculate the expected revenue value of the process group based on the conflict coefficient of the process group and the order information. The marginal revenue decay factor is introduced into the expected revenue value to obtain the actual revenue value. Game optimization iteration is performed based on the actual revenue value. When the revenue increment between two adjacent iterations is less than the preset convergence threshold, the optimal scheduling scheme is output. The fifth unit is used to convert the optimal scheduling scheme into a sequence of scheduling instructions and issue them for execution.

9. An electronic device, characterized in that, include: processor; Memory used to store processor-executable instructions; The processor is configured to invoke instructions stored in the memory to execute the method according to any one of claims 1 to 7.

10. A computer-readable storage medium having computer program instructions stored thereon, characterized in that, When the computer program instructions are executed by the processor, they implement the method described in any one of claims 1 to 7.

Images