Intelligent production scheduling method based on artificial intelligence

Abstract

The present invention relates to the technical field of intelligent production scheduling, and discloses an intelligent production scheduling method based on artificial intelligence, which obtains consumption data of a historical time period at a consumer end, a number of production scheduling logics at a planning end, the priority of the production scheduling logic, and the capacity data at a production end; constructs a vulnerability query model of the production scheduling logic, and the vulnerability query model includes optimization constraints, pseudo-individuals, and optimization functions: based on the consumption data of the consumer end in the historical time period, generates a number of pseudo-individuals that meet the optimization constraints, and the optimization constraints are constructed based on the capacity data of the production end; constructs an optimization function based on the production scheduling logic and the priority of the production scheduling logic; screens the pseudo-individuals based on the optimization function to obtain a number of training samples; constructs an optimization model through reinforcement learning based on the training samples, and replaces the planning end with the optimization model.

Description

Technical Field

[0001] The present invention relates to the technical field of intelligent production scheduling, and more specifically, to an intelligent production scheduling method based on artificial intelligence. Background Art

[0002] In the traditional production scheduling field, it can generally be abstractly divided into the consumer side, the planning side, and the production side. The planning side collects consumption data from the consumer side, processes the consumption data according to the production capacity of the production side and several artificially formulated production scheduling logics, and obtains the formulated production scheduling plan to coordinate the supply and demand balance between the consumer side and the production side. However, with the development of the Internet economy, the uncertainty brought about by the traffic economy makes it difficult for the planning side to cope with the changes in supply and demand brought about by the traffic economy. Therefore, artificial intelligence is needed to assist the planning side in solving the imbalance problem of supply and demand. Summary of the invention

[0003] The present invention provides an intelligent production scheduling method based on artificial intelligence to solve the technical problems raised in the background technology.

[0004] The present invention provides an intelligent production scheduling method based on artificial intelligence, which is applied to the consumer end, the planning end and the production end, including:

[0005] Step 1: Obtain consumption data of the historical time period on the consumer side, several scheduling logics on the planning side, the priority of the scheduling logic, and the production capacity data on the production side;

[0006] Step 2: Build a vulnerability query model for production scheduling logic. The vulnerability query model includes optimization constraints, pseudo individuals, and optimization functions:

[0007] Based on the consumption data of the consumer side in the historical time period, several pseudo individuals that meet the optimization constraints are generated. The optimization constraints are constructed based on the production capacity data of the production side.

[0008] Build optimization functions based on scheduling logic and scheduling logic priorities;

[0009] Step 3, screening pseudo individuals based on the optimization function to obtain several training samples;

[0010] Step 4: Build an optimization model through reinforcement learning based on the training samples, and replace the planning end with the optimization model.

[0011] Furthermore, based on the consumption data of the consumer side in the historical time period, several pseudo individuals that meet the optimization constraints are generated, including:

[0012] Obtain consumption data of the consumer end in a historical time period at fixed time intervals. The consumption data at each moment includes: the type of consumer products and the quantity of each type of consumer products;

[0013] Based on the consumption data in the historical time period, several pseudo individuals are randomly generated ;

[0014] ;

[0015] in, represents the kth moment, represents the kth moment The number of consumer products of this type is , represents the kth moment The number of consumer products of this type is , represents the kth moment The number of consumer products of this type is , , Represents the number of moments in the historical time period;

[0016] Among them, based on the consumption end in the historical time period The mean number of consumer products of the type , construct the Mean range , ,in, and Respectively represent The minimum and maximum values ​​of the mean range; The mean value of In the range of values; and so on, we get The mean values ​​of Within the value range and The value range is within the range.

[0017] Furthermore, optimization constraints are constructed based on the production capacity data, including:

[0018] Capacity data includes: the type of production equipment, the maximum load per unit time of each type of production equipment, the number of each type of production equipment, and the operating cost per unit time of each type of production equipment;

[0019] Based on the capacity data, the optimization constraints are determined as follows:

[0020] The shelf life of consumer products is , the unit cost of each type of consumer product is , the unit revenue is Any type of consumer product that undergoes Then carry out harmless treatment;

[0021] The mean ;in, Indicates The maximum load per unit time of the production equipment of the type; and so on, we get The mean as well as The mean ;

[0022] For The mean range is The intersection of the value range and the corresponding optimization constraint is calculated to obtain The final value range of the mean.

[0023] Furthermore, an optimization function is constructed based on the scheduling logic and the priority of the scheduling logic, including:

[0024] The production scheduling logic is abstracted into a logic flow chart according to the corresponding priority, and the logic flow chart is used as the optimization function; the logic flow chart includes nodes, directed edges and Boolean logic gates, as follows:

[0025] Map a production scheduling logic to a node in the logic flow chart, construct a directed edge from the high-priority node to the low-priority node between adjacent priority nodes, establish Boolean logic gates for the directed edges of the two child nodes of the high-priority node with the same priority, and activate the corresponding directed edges based on the Boolean logic gates.

[0026] Furthermore, the pseudo data individuals are screened based on the optimization function to obtain training samples, including:

[0027] Step 51, input pseudo individuals from the starting node of the logic flow chart in chronological order. arrive The consumption data at each moment is collected, and the final cost and benefit are calculated. The difference between the cost and benefit is used as the optimization function value of the corresponding pseudo-individual.

[0028] Step 52, retain the pseudo-individuals whose optimization function values ​​are greater than the preset difference, and use the remaining pseudo-individuals as parents for cyclic update until a preset number of pseudo-individuals have optimization function values ​​greater than the preset difference, then stop updating, and use the pseudo-individuals whose optimization function values ​​are greater than the preset difference as training samples.

[0029] Furthermore, the remaining pseudo-individuals are used as parents for cyclic update, including:

[0030] For pseudo-individuals The elements in are updated, and the updates include a first update, a second update, and a third update;

[0031] The calculation formula for the first update is as follows:

[0032] ;

[0033] in, represents the update of pseudo-individuals, represents the parent pseudo-individual, Indicates that any pseudo individual whose optimization function value is greater than the preset difference value, The mean is 0 and the standard deviation is A random value sampled from a normal distribution, represents the noise weight, Represents a random number from 0 to 1, Represents a random number from 1 to 2;

[0034] The calculation formula for the second update is as follows:

[0035] ;

[0036] in, represents the natural base, Represents the operation of finding the Euclidean distance;

[0037] The calculation formula for the third update is as follows:

[0038] ;

[0039] in, Represents any one of all pseudo-individuals.

[0040] Furthermore, an optimization model is constructed through reinforcement learning based on training samples, including:

[0041] Step 71, delete the directed edges and Boolean logic gates from the logic flow graph to obtain m nodes;

[0042] Step 72, establish R wandering factors, and drive the R wandering factors to reconstruct several update flow charts based on the strengthened constraints; wherein the strengthened constraints include:

[0043] The positions of the first node and the mth node in the logic flow graph are fixed;

[0044] Traverse the remaining nodes in random series or parallel to construct directed edges;

[0045] Step 73, if the non-pointing ends of the directed edges of any multiple nodes are the same node, then Boolean logic gates are constructed for the directed edges of any multiple nodes respectively;

[0046] Step 74, according to step 51, respectively calculating the average difference between the cost and the benefit of all training samples for the R update flow charts;

[0047] Step 75, obtaining the median of the average difference, pausing the wandering factors whose average difference is less than the median, and performing binary fission on the wandering factors whose average difference is greater than the median;

[0048] Step 76, repeat step 75 until the average difference between the cost and benefit of at least one update flowchart for all training samples is less than a preset difference, then a final update flowchart is obtained, and the final update flowchart is set as the optimization model.

[0049] The beneficial effects of the present invention are as follows: by collecting historical data from the consumer side, the scheduling logic and priority of the planning side, and the production capacity data from the production side, a scheduling logic vulnerability query model is constructed, pseudo-individuals are generated using optimization constraints, and they are screened through optimization functions to generate training samples. Subsequently, through reinforcement learning, step-by-step optimization and iteration, an optimized scheduling model that can dynamically adapt to changes in supply and demand is constructed. The present invention improves the accuracy and flexibility of production plans, can effectively cope with complex supply and demand imbalances, and significantly reduces production costs and resource waste, improves the operating efficiency of the production side, and at the same time achieves a rapid response to consumer demand, thereby optimizing the overall supply and demand coordination efficiency and economic benefits. BRIEF DESCRIPTION OF THE DRAWINGS

[0050] Figure 1 It is a flow chart of the intelligent production scheduling method based on artificial intelligence of the present invention. DETAILED DESCRIPTION

[0051] The subject matter described herein will now be discussed with reference to example embodiments. It should be understood that the discussion of these embodiments is only to enable those skilled in the art to better understand and implement the subject matter described herein, and the functions and arrangements of the elements discussed may be changed without departing from the scope of protection of the contents of this specification. Each example may omit, replace or add various processes or components as needed. In addition, the features described relative to some examples may also be combined in other examples.

[0052] like Figure 1 As shown in the figure, the intelligent scheduling method based on artificial intelligence is applied to the consumption end, planning end and production end, including:

[0053] Step 1: Obtain consumption data of the historical time period on the consumer side, several scheduling logics on the planning side, the priority of the scheduling logic, and the production capacity data on the production side;

[0054] Step 2: Build a vulnerability query model for production scheduling logic. The vulnerability query model includes optimization constraints, pseudo individuals, and optimization functions:

[0055] Based on the consumption data of the consumer side in the historical time period, several pseudo individuals that meet the optimization constraints are generated. The optimization constraints are constructed based on the production capacity data of the production side.

[0056] Build optimization functions based on scheduling logic and scheduling logic priorities;

[0057] Step 3, screening pseudo individuals based on the optimization function to obtain several training samples;

[0058] Step 4: Build an optimization model through reinforcement learning based on the training samples, and replace the planning end with the optimization model.

[0059] In one embodiment of the present invention, a number of pseudo individuals that meet the optimization constraints are generated based on the consumption data of the consumer end in the historical time period, including:

[0060] Obtain consumption data of the consumer end in a historical time period at fixed time intervals. The consumption data at each moment includes: the type of consumer products and the quantity of each type of consumer products;

[0061] Based on the consumption data in the historical time period, several pseudo individuals are randomly generated ;

[0062] ;

[0063] in, represents the kth moment, represents the kth moment The number of consumer products of this type is , represents the kth moment The number of consumer products of this type is , represents the kth moment The number of consumer products of this type is , , Represents the number of moments in the historical time period;

[0064] Among them, based on the consumption end in the historical time period The mean number of consumer products of the type , construct the Mean range , ,in, and Respectively represent The minimum and maximum values ​​of the mean range; The mean value of In the range of values; and so on, we get The mean values ​​of Within the value range and The value range is within the range.

[0065] Specifically, in order to generate pseudo-individuals that meet the optimization constraints, it is necessary to simulate the consumption pattern using the historical data of the consumer end. This process includes the following key steps. First, collect the consumption data of the consumer end over a period of historical time at fixed time intervals. These data include different types of consumer products and their specific consumption quantities at each time point. These data are used to fully understand the fluctuations in consumer demand at different time points. Secondly, based on the collected historical consumption data, generate pseudo-individuals randomly. Specifically, for a certain type of consumer product in the historical data, the mean of its historical consumption is used to construct a reasonable value range, defining the minimum and maximum possible values ​​of the consumption of this type of product. On this basis, the consumption of the pseudo-individual is limited to this value range. Similarly, repeat this process for other categories of consumer products to ensure that the generated pseudo-individuals can simulate real consumption scenarios in each consumption category. The pseudo-individuals generated by the above process can more accurately reflect the regularity in the historical consumption data while retaining a certain degree of randomness and flexibility.

[0066] In one embodiment of the present invention, constructing optimization constraints based on production capacity data at the production end includes:

[0067] Capacity data includes: the type of production equipment, the maximum load per unit time of each type of production equipment, the number of each type of production equipment, and the operating cost per unit time of each type of production equipment;

[0068] Based on the capacity data, the optimization constraints are determined as follows:

[0069] The shelf life of consumer products is , the unit cost of each type of consumer product is , the unit revenue is Any type of consumer product that undergoes Then carry out harmless treatment;

[0070] The mean ;in, Indicates The maximum load per unit time of the production equipment of the type; and so on, we get The mean as well as The mean ;

[0071] For The mean range is The intersection of the value range and the corresponding optimization constraint is calculated to obtain The final value range of the mean.

[0072] Specifically, in order to construct optimization constraints based on the production capacity data of the production end, it is necessary to comprehensively consider the characteristics of the production equipment and the actual operating capacity of the production end to ensure that the generated pseudo-individuals are realistic and feasible, including the following aspects. Type and load capacity of production equipment: Each type of production equipment has its maximum load capacity per unit time. For example, different production equipment may have different performance parameters, which determine the maximum number of products it can produce in a certain period of time. When constructing optimization constraints, it is necessary to take these maximum load capacities as the basis to ensure that the production plan does not exceed the actual carrying capacity of the equipment. Equipment quantity restriction: The number of each type of production equipment in the factory is limited. Therefore, when generating optimization constraints, it is necessary to comprehensively consider the quantity constraints of different equipment types to ensure that the pseudo-individuals will not arrange production needs that exceed the total amount of equipment supported. This design can avoid the problem of unrealistic scheduling plans. Operating cost: The operating cost of each type of production equipment per unit time also needs to be included in the optimization constraints. By taking the operating cost as one of the constraints, it can be ensured that the generated pseudo-individuals are reasonable in terms of economy and avoid overly expensive production plans. Product shelf life: For consumer products, shelf life is an important factor. The time from production to consumption of each type of product must not exceed its shelf life. Therefore, when generating optimization constraints, it is necessary to use the shelf life as the basis to ensure that the production scheduling plan of the pseudo-individual can meet this basic requirement. Harmless treatment: If certain products cannot be sold within their shelf life, they need to be harmlessly treated. This means that in the optimization constraints, it is necessary to introduce the cost of processing these expired products to balance the relationship between production volume and actual sales volume to avoid waste of resources. Determination of intersection and reasonable range: Based on the above data, by gradually analyzing the intersection of each optimization constraint condition, the reasonable range of production quantity of each product in each time period is finally obtained. This method can ensure that the generated pseudo-individuals not only meet the capacity constraints of the production end, but also have practical feasibility.

[0073] In one embodiment of the present invention, an optimization function is constructed based on the production scheduling logic and the priority of the production scheduling logic, including:

[0074] The production scheduling logic is abstracted into a logic flow chart according to the corresponding priority, and the logic flow chart is used as the optimization function; the logic flow chart includes nodes, directed edges and Boolean logic gates, as follows:

[0075] Map a production scheduling logic to a node in the logic flow chart, construct a directed edge from the high-priority node to the low-priority node between adjacent priority nodes, establish Boolean logic gates for the directed edges of the two child nodes of the high-priority node with the same priority, and activate the corresponding directed edges based on the Boolean logic gates.

[0076] Specifically, in order to construct an optimization function based on the scheduling logic and its priority, the scheduling logic is abstracted and visualized using the logic flow chart method to ensure that the optimization function can effectively reflect the relationship and priority weight between different logics in the scheduling process. Specifically, it includes the following aspects. Construction of logic flow chart: The scheduling logic is abstracted into a logic flow chart, which contains multiple nodes, directed edges, and Boolean logic gates. Among them, each node represents a specific scheduling logic or condition, and the directed edge represents the priority relationship between logics. For example, a node with a high priority will point to a node with a lower priority through a directed edge, thereby logically clarifying the execution order of different conditions. Expression of priority: Priority is a key factor in scheduling logic. In the logic flow chart, the directed edges between adjacent nodes are used to indicate the priority level, and the logic with a high priority will be processed first. In this way, it can be ensured that the key scheduling logic is given enough attention during the optimization process. For example, for a condition that must be met, the corresponding node has a higher priority, so it is located at a higher position in the flow chart. Introduction of Boolean logic gates: When there are multiple logic conditions with the same priority, in order to simplify the logic flow chart, Boolean logic gates are introduced to process these conditions. For example, Boolean logic gates can represent the relationship of "and" or "or", so as to uniformly represent the logical combination of multiple nodes in the flowchart. This design can effectively avoid complex logical conflicts and ensure that the flowchart is clear and easy to understand. Generation of optimization function: The optimization function is constructed based on the logic flowchart. The optimization function not only contains the relationship between nodes and edges, but also comprehensively considers the priority of nodes and their position in the logic flow. For example, the importance of nodes and their priority weights will directly affect the results of the optimization function. In addition, the introduction of Boolean logic gates further enhances the expression ability of the optimization function for complex logical conditions. Combination of target cost and benefit: The ultimate goal of the optimization function is to balance cost and benefit. During the construction process, each pseudo-individual needs to be input into the starting node of the logic flowchart, simulate the flow of its consumption data according to the time sequence, calculate its total cost and total benefit in the entire logic flowchart, and use the difference as the target value of optimization. This method ensures that the optimization function can reflect the actual economic effect of the scheduling logic.

[0077] In one embodiment of the present invention, the pseudo data individuals are screened based on the optimization function to obtain training samples, including:

[0078] Step 51, input pseudo individuals from the starting node of the logic flow chart in chronological order. arrive The consumption data at each moment is collected, and the final cost and benefit are calculated. The difference between the cost and benefit is used as the optimization function value of the corresponding pseudo-individual.

[0079] Step 52, retain the pseudo-individuals whose optimization function values ​​are greater than the preset difference, and use the remaining pseudo-individuals as parents for cyclic update until a preset number of pseudo-individuals have optimization function values ​​greater than the preset difference, then stop updating, and use the pseudo-individuals whose optimization function values ​​are greater than the preset difference as training samples.

[0080] Specifically, training samples are selected by using pseudo-individuals to impact loopholes in the existing scheduling logic. This design ensures that the selected pseudo-individuals can represent the potential improvement direction of the scheduling logic on the planning side. In the screening stage, the pseudo-individuals are input into the starting node of the logic flow chart one by one, and the flow of consumption data is simulated in chronological order. In this process, the pseudo-individual needs to pass through each node of the logic flow chart in turn and trigger the scheduling logic related to it. Through this simulation, the adaptability of the pseudo-individual to the scheduling logic and its potential impact ability can be verified. In the process of pseudo-individuals flowing through the logic flow chart, the system will count the production costs and benefits related to the pseudo-individuals in real time. The core purpose of this process is to quantify the impact effect of pseudo-individuals on the existing scheduling logic. The retained pseudo-individuals usually have the following characteristics: they can trigger unforeseen loopholes in the existing scheduling logic; they can successfully break through the constraints in some scheduling logic. These pseudo-individuals represent the potential improvement direction of the existing scheduling logic, and are therefore used as training samples.

[0081] In one embodiment of the present invention, the remaining pseudo individuals are used as parents for cyclic updating, including:

[0082] For pseudo-individuals The elements in are updated, and the updates include a first update, a second update, and a third update;

[0083] The calculation formula for the first update is as follows:

[0084] ;

[0085] in, represents the update of pseudo-individuals, represents the parent pseudo-individual, Indicates that any pseudo individual whose optimization function value is greater than the preset difference value, The mean is 0 and the standard deviation is A random value sampled from a normal distribution, represents the noise weight, Represents a random number from 0 to 1, Represents a random number from 1 to 2;

[0086] The calculation formula for the second update is as follows:

[0087] ;

[0088] in, represents the natural base, Represents the operation of finding the Euclidean distance;

[0089] The calculation formula for the third update is as follows:

[0090] ;

[0091] in, Represents any one of all pseudo-individuals.

[0092] Specifically, the update mechanism of pseudo-individuals is divided into three different update methods, namely the first update, the second update and the third update. In addition, it is ensured that enough pseudo-individuals are obtained as training samples. The first update focuses on randomness: by introducing random perturbation factors, part of the data of the pseudo-individual is changed to simulate different consumption or production scenarios. This randomness makes the pseudo-individual more exploratory and able to try more possibilities. The second update focuses on accuracy: by analyzing the similarities between pseudo-individuals and optimizing the differences between them and excellent pseudo-individuals, the updated pseudo-individuals are closer to the samples that successfully impact logical vulnerabilities. The third update focuses on globality: all pseudo-individuals are adjusted from a global perspective, and the characteristics of multiple pseudo-individuals are combined to generate new pseudo-individuals to explore new logical vulnerabilities.

[0093] In one embodiment of the present invention, an optimization model is constructed through reinforcement learning based on training samples, including:

[0094] Step 71, delete the directed edges and Boolean logic gates from the logic flow graph to obtain m nodes;

[0095] Step 72, establish R wandering factors, and drive the R wandering factors to reconstruct several update flow charts based on the strengthened constraints; wherein the strengthened constraints include:

[0096] The positions of the first node and the mth node in the logic flow graph are fixed;

[0097] Traverse the remaining nodes in random series or parallel to construct directed edges;

[0098] Step 73, if the non-pointing ends of the directed edges of any multiple nodes are the same node, then Boolean logic gates are constructed for the directed edges of any multiple nodes respectively;

[0099] Step 74, according to step 51, respectively calculating the average difference between the cost and the benefit of all training samples for the R update flow charts;

[0100] Step 75, obtaining the median of the average difference, pausing the wandering factors whose average difference is less than the median, and performing binary fission on the wandering factors whose average difference is greater than the median;

[0101] Step 76, repeat step 75 until the average difference between the cost and benefit of at least one update flowchart for all training samples is less than a preset difference, then a final update flowchart is obtained, and the final update flowchart is set as the optimization model.

[0102] Specifically, through the reinforcement learning-based approach, we build and optimize the updated flowchart to further improve the adaptability and efficiency of the scheduling logic. The core of the reinforcement learning mechanism is to reconstruct the logic flowchart using training samples. Through the reinforcement learning mechanism, the scheduling logic can be deeply optimized in combination with the actual performance of the training samples. The optimization model has the following significant advantages. Improve the adaptability of the scheduling logic to cope with diverse and complex production scenarios. Reduce resource waste and optimize the balance between production costs and benefits. Improve the flexibility and stability of the process to adapt to demand fluctuations and changes.

[0103] The above describes an embodiment of the present embodiment, but the present embodiment is not limited to the above-mentioned specific implementation methods. The above-mentioned specific implementation methods are merely illustrative and not restrictive. Under the guidance of the present embodiment, ordinary technicians in this field can also make many forms, all of which are protected by the present embodiment.

Claims

1. An intelligent production scheduling method based on artificial intelligence, applied to the consumer, planning and production ends, characterized by: include: Step 1: Obtain consumption data of the historical time period on the consumer side, several scheduling logics on the planning side, the priority of the scheduling logic, and the production capacity data on the production side; Step 2: Build a vulnerability query model for production scheduling logic. The vulnerability query model includes optimization constraints, pseudo individuals, and optimization functions: Based on the consumption data of the consumer side in the historical time period, several pseudo individuals that meet the optimization constraints are generated, specifically: Obtain consumption data of the consumer end in a historical time period at fixed time intervals. The consumption data at each moment includes: the type of consumer products and the quantity of each type of consumer products; Based on the consumption data in the historical time period, several pseudo individuals S are randomly generated; ； in, represents the kth moment, represents the kth moment The number of consumer products of this type is , represents the kth moment The number of consumer products of this type is , represents the kth moment The number of consumer products of this type is , , Represents the number of moments in the historical time period; Among them, based on the consumption end in the historical time period The mean number of consumer products of the type , construct the Mean range , ,in, and Respectively represent The minimum and maximum values ​​of the mean range; The mean value of In the range of values; and so on, we get The mean values ​​of Within the value range and Within the value range; The optimization constraints are constructed based on the production capacity data; The optimization function is constructed based on the scheduling logic and the priority of the scheduling logic, specifically: The production scheduling logic is abstracted into a logic flow chart according to the corresponding priority, and the logic flow chart is used as the optimization function; the logic flow chart includes nodes, directed edges and Boolean logic gates, as follows: Map a production scheduling logic to a node of a logic flow chart, construct a directed edge from a high-priority node to a low-priority node between adjacent priority nodes, establish Boolean logic gates for the directed edges of two child nodes with the same priority of the high-priority node, and activate the corresponding directed edge based on the Boolean logic gate; Step 3: Screen the pseudo-individuals based on the optimization function to obtain several training samples, specifically: Step 51, input pseudo individuals from the starting node of the logic flow chart in chronological order. arrive The consumption data at each moment is collected, and the final cost and benefit are calculated. The difference between the cost and benefit is used as the optimization function value of the corresponding pseudo-individual. Step 52, retaining pseudo individuals whose optimization function values ​​are greater than the preset difference, and taking the remaining pseudo individuals as parents for cyclic update, until a preset number of pseudo individuals have optimization function values ​​greater than the preset difference, then stop updating, and take the pseudo individuals whose optimization function values ​​are greater than the preset difference as training samples; Step 4: Build an optimization model through reinforcement learning based on the training samples, and replace the planning end with the optimization model. Specifically: Step 71, delete the directed edges and Boolean logic gates from the logic flow graph to obtain m nodes; Step 72, establish R wandering factors, and drive the R wandering factors to reconstruct several update flow charts based on the strengthened constraints; wherein the strengthened constraints include: The positions of the first node and the mth node in the logic flow graph are fixed; Traverse the remaining nodes by random series or parallel connection to construct directed edges; Step 73, if the non-pointing ends of the directed edges of any multiple nodes are the same node, then Boolean logic gates are constructed for the directed edges of any multiple nodes respectively; Step 74, according to step 51, respectively calculating the average difference between the cost and the benefit of all training samples for the R update flow charts; Step 75, obtaining the median of the average difference, pausing the wandering factors whose average difference is less than the median, and performing binary fission on the wandering factors whose average difference is greater than the median; Step 76, repeat step 75 until the average difference between the cost and benefit of at least one update flowchart for all training samples is less than a preset difference, then a final update flowchart is obtained, and the final update flowchart is set as the optimization model.

2. The intelligent production scheduling method based on artificial intelligence according to claim 1 is characterized in that: Construct optimization constraints based on production capacity data, including: The capacity data includes: the type of production equipment, the maximum load per unit time of each type of production equipment, the number of each type of production equipment and the operating cost per unit time of each type of production equipment; Based on the capacity data, the optimization constraints are determined as follows: The shelf life of consumer products is , the unit cost of each type of consumer product is , the unit revenue is ; Any type of consumer product Then carry out harmless treatment; The mean ;in, Indicates The maximum load per unit time of the production equipment of the type; and so on, we get The mean as well as The mean ; For The mean range is The intersection of the value range and the corresponding optimization constraint is calculated to obtain The final value range of the mean.

3. The intelligent production scheduling method based on artificial intelligence according to claim 2 is characterized in that: The remaining pseudo-individuals are used as parents to perform cyclic updates, including: For pseudo-individuals The elements in are updated, and the updates include a first update, a second update, and a third update; The calculation formula for the first update is as follows: ； in, represents the update of pseudo-individuals, represents the parent pseudo-individual, Indicates that any pseudo individual whose optimization function value is greater than the preset difference value, The mean is 0 and the standard deviation is A random value sampled from a normal distribution, represents the noise weight, Represents a random number from 0 to 1, Represents a random number from 1 to 2; The calculation formula for the second update is as follows: ； in, represents the natural base, Represents the operation of finding the Euclidean distance; The calculation formula for the third update is as follows: ； in, Represents any one of all pseudo-individuals.

Images