CN117408832B

CN117408832B - Abnormality analysis method and system applied to environment-friendly glue production control system

Info

Publication number: CN117408832B
Application number: CN202311729627.6A
Authority: CN
Inventors: 林聪荣
Original assignee: Taichang Resin Foshan Co ltd
Current assignee: Taichang Resin Foshan Co ltd
Priority date: 2023-12-15
Filing date: 2023-12-15
Publication date: 2024-04-02
Anticipated expiration: 2043-12-15
Also published as: CN117408832A

Abstract

The embodiment of the application provides an anomaly analysis method and an anomaly analysis system applied to an environmental protection glue production control system, which relate to the technical field of intelligent manufacturing, and are characterized in that firstly, a production scheduling event sequence of the environmental protection glue production control system is obtained, a target neural network for making an abnormal resource occupation decision is obtained, the target neural network is configured according to a preset network parameter knowledge base, a target network parameter is generated, knowledge learning is conducted on the target neural network based on the production scheduling event sequence and the target network parameter, and the target neural network for completing the knowledge learning is generated, so that the target neural network for completing the knowledge learning is loaded into a system computing terminal of the environmental protection glue production control system, and a resource occupation optimization scheme is determined by utilizing the target neural network, so that operation of corresponding production control nodes is optimized. Therefore, the resource occupation in the production process of the environment-friendly adhesive can be effectively optimized, the production efficiency is improved, and the cost is reduced.

Description

Abnormality analysis method and system applied to environment-friendly glue production control system

Technical Field

The application relates to the technical field of intelligent manufacturing, in particular to an anomaly analysis method and system applied to an environment-friendly glue production control system.

Background

The environmental glue manufacturing process is a complex manufacturing process involving a plurality of production control nodes, each having respective operating parameters. These operating parameters play an important role in resource occupation, production efficiency, cost control, etc. Traditional production control methods mainly rely on manual setting and adjustment of these parameters, however, due to complexity and uncertainty of the production environment, it is often difficult to achieve optimal production effect.

Therefore, a new method is urgently needed, the resource occupation in the production process of the environment-friendly glue can be effectively optimized, the production efficiency is improved, the cost is reduced, the dependence on professional knowledge is reduced, and the self-adaptability and the stability of the system are improved.

Disclosure of Invention

In order to at least overcome the above-mentioned shortcomings in the prior art, the present application aims to provide an anomaly analysis method and system applied to an environmental protection glue production control system.

In a first aspect, the present application provides an anomaly analysis method applied to an environmental protection glue production control system, and applied to an intelligent production monitoring system, the method comprising:

acquiring a production scheduling event sequence of an environmental protection glue production control system, wherein the environmental protection glue production control system comprises a plurality of production control nodes, each production control node correspondingly defines at least one production control node field, the production scheduling event sequence comprises a node state value corresponding to the node field correspondingly defined by each production control node, and the environmental protection glue production control system comprises at least one node field correspondingly defined by the production control node and comprises an adjustable node field;

Acquiring a target neural network for making an abnormal resource occupation decision on the environment-friendly glue production control system, and carrying out network parameter configuration on the target neural network according to a preset network parameter knowledge base to generate target network parameters called by the target neural network when making the resource occupation decision;

performing knowledge learning on the target neural network based on the production scheduling event sequence and the target network parameters, and generating a target neural network for completing knowledge learning;

loading the target neural network with knowledge learning into a system computing terminal of the environment-friendly glue production control system, and enabling the system computing terminal to determine a resource occupation optimization scheme of the environment-friendly glue production control system by using the target neural network with knowledge learning, wherein the resource occupation optimization scheme comprises node state values corresponding to adjustable node fields of at least one production control node, and optimizing operation of corresponding production control nodes in the environment-friendly glue production control system according to the node state values corresponding to each adjustable node field in the resource occupation optimization scheme.

In a possible implementation manner of the first aspect, the target neural network includes a plurality of neural network branches corresponding to responses of each production control node in the environmental protection glue production control system, and the target network parameters include: each neural network branch in the target neural network invokes a resource occupation decision function;

Wherein, each neural network branch in the target neural network is respectively based on respective resource occupation decision functions to jointly decide the resource occupation state of the environmental protection glue production control system;

the knowledge learning is performed on the target neural network based on the production scheduling event sequence and the target network parameter, and the target neural network for completing knowledge learning is generated, which comprises the following steps:

performing knowledge learning on the target neural network based on the production scheduling event sequence and the target network parameters to generate a temporary neural network;

performing application index verification on the temporary neural network by using a production control simulation process to generate an application index verification result;

if the application index verification result reflects that the temporary neural network passes the application index verification, the temporary neural network is used as a target neural network for completing knowledge learning;

and if the application index verification result reflects that the temporary neural network fails the application index verification, carrying out knowledge learning on the temporary neural network to generate a target neural network for completing the knowledge learning.

In a possible implementation manner of the first aspect, the temporary neural network includes temporary neural network branches obtained by knowledge learning on each neural network branch;

The step of verifying the application index of the temporary neural network by using the production control simulation process to generate an application index verification result comprises the following steps:

generating verification output data of an ith temporary neural network branch in the temporary neural networks by using the production control simulation process for the ith temporary neural network branch; i is [1, M ], M is the number of temporary neural network branches included in the temporary neural network;

and verifying the decision accuracy of the ith temporary neural network branch based on the verification output data, and generating an application index verification result.

In a possible implementation manner of the first aspect, the environmental protection glue production control system is located in a global production line control system;

operating each production control node in the production control simulation process;

determining a resource occupation optimization scheme of the production control simulation process by using the temporary neural network at intervals of a set time period in a target time domain range;

each time a group of resource occupation optimization schemes are determined, controlling the operation of corresponding production control nodes in the production control simulation process according to node state values corresponding to all adjustable node fields in the resource occupation optimization schemes determined by the round;

After the target time domain range is terminated, determining a target resource occupation state of the global production line control system in the target time domain range, wherein the global production line control system generates a plurality of production control scene data in the target time domain range, and the target resource occupation state comprises: the global production line control system characterizes the characteristic value of the resource occupation under each production control scene data in the plurality of production control scene data;

obtaining a comparison resource occupation state of the global production line control system, wherein the comparison resource occupation state comprises the following steps: when the temporary neural network is not called, the global production line control system occupies a characteristic value of resource occupation under each production control scene data in the plurality of production control scene data;

verifying the influence of the occupation of the resources of the temporary neural network based on the occupation state of the target resources and the occupation state of the control resources, and generating a target occupation influence weight of the temporary neural network;

and verifying the resource occupation evaluation index of the temporary neural network according to the target resource occupation influence weight of the temporary neural network and the set resource occupation requirement, and generating an application index verification result.

In a possible implementation manner of the first aspect, the resource occupancy characterization feature value is a ratio between a quantized value of a computational resource used by an occupied resource of the global production line control system and a quantized value of all computational resources used by the global production line control system;

the verifying the influence of the occupation of the resources on the temporary neural network based on the occupation state of the target resources and the occupation state of the control resources, generating the influence weight of the occupation of the target resources on the temporary neural network, includes:

the global production line control system walks a plurality of production control scene data generated in the target time domain range, the production control scene data walked by the round is output as the production control scene data of the round, the maximum resource occupation value and the minimum resource occupation value corresponding to the occupation resource range in the production control scene data of the round are determined, the average value calculation is carried out on the maximum resource occupation value and the minimum resource occupation value, and the average quantized value of occupation resources corresponding to the production control scene data of the round is generated;

calculating a difference value between a resource occupation representation characteristic value corresponding to the production control scene data of the present round in the target resource occupation state and a resource occupation representation characteristic value corresponding to the production control scene data of the present round in the comparison resource occupation state, multiplying the corresponding difference value and an average quantized value of occupied resources corresponding to the production control scene data of the present round, and generating a resource occupation optimization completion degree of the temporary neural network under the production control scene data of the present round;

Calculating a resource optimization effect value of the temporary neural network under the current round of production control scene data based on the determined resource occupation optimization completion degree and the existence time of the current round of production control scene data in the target time domain range;

and after the plurality of production control scene data are all walked, fusing the resource optimization effect values of the temporary neural network under each production control scene data to generate a target resource optimization effect value of the temporary neural network.

In a possible implementation manner of the first aspect, the setting a resource occupancy requirement is used to reflect: the target resource occupation influence weight of the temporary neural network is not smaller than the preset resource occupation influence weight;

the step of testing the resource occupation evaluation index of the temporary neural network according to the target resource occupation influence weight of the temporary neural network and the set resource occupation requirement to generate an application index verification result comprises the following steps:

outputting an application index verification result for reflecting the temporary neural network passing the application index verification if the target resource occupation influence weight of the temporary neural network is not less than the preset resource occupation influence weight;

And if the target resource occupation influence weight of the temporary neural network is smaller than the preset resource occupation influence weight, outputting an application index verification result for reflecting that the temporary neural network fails application index verification.

In a possible implementation manner of the first aspect, the environmental protection glue production control system correspondingly defines a structural relation model, where the structural relation model includes Q monitoring area members, W feature members, R network members and P function members;

one monitoring area member corresponds to one monitoring area, and the monitoring area is used for monitoring the node state value of a production control node in the environment-friendly glue production control system; for a kth feature member, there are H members in the structural relation model, and each member in the H members is mapped to the kth feature member according to a mapping link, where the feature data sequence corresponding to the kth feature member is generated based on the feature data of the member corresponding to the H member, and the H member includes at least one of the following: at least one monitoring zone member and at least one feature member other than the kth feature member; k is [1, W ], H is a positive integer;

One network member corresponds to one neural network branch, for the f-th network member, G members exist in the structural relation model, each member in the G members is mapped to the f-th network member according to a mapping link, the neural network branch corresponding to the f-th network member performs network operation based on data corresponding to the G members, and the G members comprise at least one of the following: at least one feature member and at least one network member other than the f-th network member; f is [1, R ], G is a positive integer;

for a t-th function member, u network members exist in the structural relation model, each network member in the u network members is mapped to the t-th function member according to a mapping link, the resource occupation decision function corresponding to the t-th function member is configured by the neural network corresponding to each network member in the u network members, t belongs to [1, P ], and u belongs to [1, W ];

the training step of each neural network branch in the target neural network specifically comprises the following steps:

acquiring an input sample of each neural network branch based on node state values corresponding to each member field of each production control node in the production scheduling event sequence and model architecture data of a structural relation model of the environmental protection glue production control system;

Training each neural network branch based on the input samples.

In a possible implementation manner of the first aspect, the performing network parameter configuration on the target neural network according to a preset network parameter knowledge base, generating a target network parameter called by the target neural network when making a resource occupation decision, includes:

for each neural network branch in the target neural network, determining a network member corresponding to the each neural network branch from the structural relationship model;

based on the model architecture data of the structural relation model, the resource occupation decision function corresponding to the function member mapped to the determined network member through the mapping link is used as the resource occupation decision function called by each neural network branch according to a preset network parameter knowledge base.

In a possible implementation manner of the first aspect, the step of determining, by the system computing terminal, a resource occupation optimization scheme of the environmental protection glue production control system by using the target neural network for completing knowledge learning specifically includes:

generating a plurality of groups of strategy data based on a simulated annealing algorithm, wherein each group of strategy data comprises node state values corresponding to each adjustable node field;

Utilizing the target neural network for completing knowledge learning to respectively carry out resource occupation decision on the environmental protection glue production control system based on each group of strategy data, and generating a resource occupation decision result; the resource occupation decision result comprises: the environment-friendly glue production control system predicts the resource occupation state under each group of strategy data;

and selecting one group of strategy data from the plurality of groups of strategy data based on the resource occupation decision result as a resource occupation optimization scheme of the environment-friendly glue production control system.

For example, in a possible implementation manner of the first aspect, the loading the target neural network for completing knowledge learning into a system computing terminal of the environmental protection glue production control system includes:

loading the target neural network for completing knowledge learning into a neural network resource library, so that the neural network resource library loads the target neural network for completing knowledge learning into a system computing terminal of the environment-friendly glue production control system;

when the neural network resource library monitors that the target neural network is changed, the target neural network loaded in the system computing terminal is replaced based on the changed target neural network.

In a second aspect, embodiments of the present application further provide an intelligent production monitoring system, where the intelligent production monitoring system includes a processor and a machine-readable storage medium, where the machine-readable storage medium stores a computer program, and the computer program is loaded and executed in conjunction with the processor to implement the anomaly analysis method applied to the environmental protection glue production control system in the first aspect.

Based on the technical scheme in any aspect, firstly, a production scheduling event sequence of the environmental protection glue production control system is acquired, a target neural network for making abnormal resource occupation decisions is acquired, the target neural network is configured according to a preset network parameter knowledge base, target network parameters are generated, knowledge learning is conducted on the target neural network based on the production scheduling event sequence and the target network parameters, and the target neural network for completing the knowledge learning is generated, so that the target neural network for completing the knowledge learning is loaded into a system computing terminal of the environmental protection glue production control system, and a resource occupation optimization scheme is determined by utilizing the target neural network, so that operation of corresponding production control nodes is optimized. Therefore, the resource occupation in the production process of the environment-friendly adhesive can be effectively optimized, the production efficiency is improved, and the cost is reduced.

Namely, a neural network with knowledge learning completed is generated by acquiring a production scheduling event sequence of the environmental protection glue production control system and configuring and learning the target neural network by using a preset network parameter knowledge base. This enables the neural network to efficiently make resource occupancy decisions based on the data characteristics in the production scheduling event sequence. The neural network with knowledge learning completed is loaded into a system computing terminal of the environment-friendly glue production control system, and the resource occupation optimization scheme of the system is determined by utilizing the neural network, so that the resource occupation in the environment-friendly glue production process can be optimized, the production efficiency is improved, and the production cost is reduced. Therefore, intelligent management of the environment-friendly glue production control system is realized, the production efficiency is improved, the self-adaptability and stability of the system are also improved, the necessity of manual intervention is reduced, and the degree of automation of the production process is greatly improved.

Drawings

For a clearer description of the technical solutions of the embodiments of the present application, reference will be made to the accompanying drawings, which are needed to be activated, for the sake of simplicity, and it should be understood that the following drawings only illustrate some embodiments of the present application and should therefore not be considered as limiting the scope, and that other related drawings can be obtained by those skilled in the art without the inventive effort.

Fig. 1 is a schematic flow chart of an anomaly analysis method applied to an environmental protection glue production control system according to an embodiment of the present application;

fig. 2 is a schematic functional block diagram of an intelligent production monitoring system for implementing the anomaly analysis method applied to the environmental protection glue production control system according to the embodiment of the present application.

Detailed Description

The following description is presented to enable one of ordinary skill in the art to make and use the application and is provided in the context of a particular application and its requirements. It will be apparent to those having ordinary skill in the art that various changes can be made to the disclosed embodiments and that the general principles defined herein may be applied to other embodiments and applications without departing from the principles and scope of the present application. Thus, the present application is not limited to the embodiments described, but is to be accorded the widest scope consistent with the claims.

Referring to fig. 1, the application provides an anomaly analysis method applied to an environmental protection glue production control system, which comprises the following steps.

Step S110, a production scheduling event sequence of the environmental protection glue production control system is obtained.

In this embodiment, the environmental protection glue production control system includes a plurality of production control nodes, each production control node defines at least one production control node field correspondingly, the production scheduling event sequence includes a node state value corresponding to a node field defined correspondingly by each production control node, and the environmental protection glue production control system includes at least one production control node which includes an adjustable node field in a node field defined correspondingly by the production control node.

For example, in an environmental glue production control system, there may be multiple production control nodes, such as an order processing node, a production scheduling node, and an inventory management node. Each production control node has its specific operating parameters or fields such as processing time, response speed, memory usage, etc. When the environment-friendly glue production control system is in operation, the states of all production control nodes and corresponding operation parameter values are recorded, and a series of production scheduling events are formed.

By way of example, if the environmental glue production control system is a complex software system for managing and optimizing the production process, it may contain the following production control nodes (i.e., software modules):

an order processing module: is responsible for receiving and processing production orders.

A production scheduling module: and is responsible for production scheduling according to the order and the current resource status.

Inventory management module: is responsible for tracking and managing the inventory of raw materials and products.

Each module has some node fields (i.e., operating parameters) that can be defined, such as:

the maximum concurrency processing order quantity of the order processing module: it is determined how many orders the order processing module can process simultaneously.

Priority rules of the production scheduling module: the scheduling order of the production tasks is determined.

Minimum inventory threshold for inventory management module: when inventory falls below this value, the system will trigger a new procurement process.

The node state value is then the specific value of these parameters at a certain point in time. For example, the maximum concurrency processing order amount of the order processing module may be 10 at 8 a.m. and adjusted to 15 at 12 a.m.

The adjustable node field is a parameter that can be dynamically changed according to the need. For example, the maximum amount of concurrently processed orders by the order processing module may be adjusted based on the real-time number of orders to more efficiently utilize system resources.

Step S120, a target neural network for making an abnormal resource occupation decision on the environmental protection glue production control system is obtained, network parameter configuration is conducted on the target neural network according to a preset network parameter knowledge base, and target network parameters called by the target neural network when making the resource occupation decision are generated.

For example, embodiments of the present application require a neural network model to learn how to optimize system resource occupancy based on a sequence of production scheduling events and adjustable parameters. For example, a pre-set network parameter knowledge base may be pre-configured, which contains network parameter settings that perform well on similar problems. Thus, the target neural network may be configured based on this knowledge base of network parameters.

And step S130, carrying out knowledge learning on the target neural network based on the production scheduling event sequence and the target network parameters, and generating a target neural network for completing knowledge learning.

For example, a target neural network may be trained using a production scheduling event sequence and target network parameters. Thus, the target neural network will learn how to predict and optimize the system's resource occupancy based on the operating states and adjustable parameters of the various production control nodes. For example, memory usage may be reduced by reducing the processing time of the order processing node when memory shortage may be learned.

Illustratively, in previous steps, operational state history data has been collected for individual production control nodes (e.g., modules of production control nodes such as order processing, production scheduling, and inventory management) in the environmental glue production control system. These operational state history data constitute a sequence of production scheduling events. For example, each production scheduling event may include the following information: the maximum concurrency processing order number of the order processing module is 10, the processing time is 200 milliseconds, and the memory occupation is 200MB; the priority rule of the production scheduling module is first in first out, the processing time is 300 milliseconds, and the memory occupation is 250MB; etc.

The target network parameters are preset network parameters and guide the configuration and training process of the target neural network. These target network parameters may include network structure (e.g., how many layers there are, how many production control nodes per layer, etc.), learning rate, activation function type, etc.

In this step, a sequence of production scheduling events is input into the target neural network, letting it learn how to predict the optimal resource occupancy scheme based on the operational status of the individual production control nodes. In particular, the target neural network may attempt to find a relationship between the operating parameters (e.g., maximum concurrent processing orders, priority rules, etc.) of the individual production control nodes and the overall system resource occupancy.

For example, the target neural network may learn such knowledge: when the maximum concurrent processing order number of the order processing module is increased, the memory occupation of the system is correspondingly increased; when a more efficient priority rule is employed, more orders can be processed without increasing memory usage. In this way, the target neural network completes knowledge learning. After a period of learning, the target neural network is trained into a model capable of predicting an optimal resource occupation scheme according to the current production scheduling event, namely the target neural network for completing knowledge learning.

That is, this step is to train the neural network with historical data so that it can make optimal resource occupancy decisions based on the real-time status of the production control system.

Step S140, loading the knowledge learning-completed target neural network into a system computing terminal of the environmental protection glue production control system, so that the system computing terminal determines a resource occupation optimization scheme of the environmental protection glue production control system by using the knowledge learning-completed target neural network, wherein the resource occupation optimization scheme comprises node state values corresponding to adjustable node fields of the at least one production control node, and optimizing operation of corresponding production control nodes in the environmental protection glue production control system according to node state values corresponding to each adjustable node field in the resource occupation optimization scheme.

For example, the trained target neural network may be loaded into a system computing terminal of an environmental protection glue production control system. The target neural network can be utilized in real time to determine an optimal resource occupancy optimization scheme as the system operates.

By way of example, the subject neural network may have learned how to predict an optimized maximum concurrency process order based on information such as the number of orders entered, priority rules, minimum inventory thresholds, etc., by learning a production scheduling event sequence, i.e., operational parameters and status history data for each production control node.

Thus, a trained target neural network is deployed into the actual operating environment to make decisions in real time. When a new production event occurs (e.g., a new order needs to be processed), the system computing terminal predicts an optimal resource usage scenario using the target neural network. This scheme includes a predicted value of at least one adjustable node field (i.e., an adjustable parameter). For example, the target neural network may suggest increasing the maximum amount of concurrent processing orders from 10 to 15.

Finally, the operation parameters of each production control node are adjusted in real time according to the advice of the target neural network. For example, in this example, the maximum amount of concurrent processing orders for the order processing module would be increased from 10 to 15.

In this way, the system's resource usage can be dynamically optimized to cope with various changes and challenges in the production process.

Based on the steps, firstly, a production scheduling event sequence of the environmental protection glue production control system is obtained, a target neural network for making abnormal resource occupation decisions is obtained, the target neural network is configured according to a preset network parameter knowledge base, target network parameters are generated, knowledge learning is conducted on the target neural network based on the production scheduling event sequence and the target network parameters, and the target neural network for completing the knowledge learning is generated, so that the target neural network for completing the knowledge learning is loaded into a system computing terminal of the environmental protection glue production control system, a resource occupation optimization scheme is determined by utilizing the target neural network, and operation of corresponding production control nodes is optimized. Therefore, the resource occupation in the production process of the environment-friendly adhesive can be effectively optimized, the production efficiency is improved, and the cost is reduced.

In one possible embodiment, the target neural network includes a plurality of neural network branches responsive to each production control node in the environmental protection glue production control system, the target network parameters including: and each neural network branch in the target neural network calls a resource occupation decision function. Each neural network branch in the target neural network is respectively based on a respective resource occupation decision function to jointly decide the resource occupation state of the environment-friendly glue production control system.

For example, in an environmental glue production control system, it is assumed that there are three main production control nodes: order processing, production scheduling, and inventory management. To optimize the operation of each production control node, a neural network branch may be defined for each production control node. The task of this neural network branch is to learn how to predict an optimal resource occupancy scheme based on historical data.

Step S130 may include:

and step S131, performing knowledge learning on the target neural network based on the production scheduling event sequence and the target network parameters to generate a temporary neural network.

And step S132, performing application index verification on the temporary neural network by using a production control simulation process, and generating an application index verification result.

And step S133, if the application index verification result reflects that the temporary neural network passes the application index verification, the temporary neural network is used as a target neural network for completing knowledge learning.

And step S134, if the application index verification result reflects that the temporary neural network fails the application index verification, carrying out knowledge learning on the temporary neural network, and generating a target neural network for completing the knowledge learning.

The target network parameters include a resource occupancy decision function invoked by each neural network branch. These resource occupancy decision functions may take into account factors such as current memory occupancy, CPU occupancy, etc., and determine how to adjust the parameters of the various production control nodes to optimize resource occupancy.

Thus, the target neural network may be trained using the production scheduling event sequence (i.e., historical state data for each production control node) and the target network parameters. For example, the target neural network may learn how to determine the maximum amount of concurrent orders for the order processing node according to the number of orders, the current resource status, and other factors. In this way, a temporary neural network is obtained.

A simulated production control flow is then used to test the performance of the temporary neural network. Some application metrics, such as overall resource occupancy of the system, average delay of order processing, etc., may be set and the performance of the temporary neural network is evaluated based on these metrics.

If the application index verification result shows that the performance of the temporary neural network meets the requirement, the temporary neural network is used as the neural network for completing knowledge learning. If the verification result is not ideal, the network parameters are required to be continuously adjusted or a learning algorithm is required to be improved, then a new temporary neural network is generated, and verification is performed again. This process may need to be repeated multiple times until a satisfactory neural network is obtained.

Finally, the obtained target neural network for completing knowledge learning can be used in an actual environment-friendly glue production control system. When a new production scheduling event occurs, the target neural network provides corresponding optimization suggestions, such as adjusting the maximum concurrent processing order quantity of the order processing nodes, so as to realize the optimal utilization of system resources.

In one possible implementation, the temporary neural network includes temporary neural network branches obtained by knowledge learning of each neural network branch.

In step S132, for an i-th temporary neural network branch of the temporary neural networks, verification output data of the i-th temporary neural network branch is generated using the production control simulation process. i belongs to [1, M ], M is the number of temporary neural network branches included in the temporary neural network. And then, carrying out decision accuracy verification on the ith temporary neural network branch based on the verification output data, and generating an application index verification result.

For example, in an environmental glue production control system, it is assumed that there are three main production control nodes: order processing, production scheduling, and inventory management. For each production control node, a temporary neural network branch is trained to predict the optimal resource occupancy scheme.

A simulated production control flow is then used to test the performance of each temporary neural network branch. For example, for a temporary neural network branch of an order processing node, a series of simulated order quantities may be entered, and then it is observed how the neural network adjusts the maximum number of concurrent processing orders to optimize resource occupancy.

After the validation output data is generated, the decision accuracy of the temporary neural network branch can be evaluated. For example, the difference between the predicted optimal concurrency processing order amount and the actual optimal value of the neural network may be compared to evaluate the decision accuracy of the neural network.

And finally, taking the evaluation result of the decision accuracy as an application index verification result. This application index validation result will reflect to the temporary neural network branch whether an accurate resource occupancy decision can already be made.

In one possible implementation, the green glue production control system is located within a global production line control system.

In step S132, the following sub-steps may be included:

step S1321, running each production control node in the production control simulation process.

For example, in a global production line control system, a simulated production control flow may be run, including nodes for order processing, production scheduling, and inventory management.

And S1322, determining a resource occupation optimization scheme of the production control simulation process by using the temporary neural network at intervals of set time intervals in a target time domain range.

For example, the temporary neural network is used to predict the optimal resource occupancy scheme at every set time (set period) within a set time range (referred to as a target time domain range).

Step S1323, if a group of resource occupation optimization schemes are determined, controlling the operation of the corresponding production control nodes in the production control simulation process according to the node state values corresponding to the adjustable node fields in the resource occupation optimization schemes determined in the round.

For example, each time a new resource occupancy optimization scheme is obtained, the corresponding production control node in the simulation process is adjusted according to the scheme. For example, if the optimization proposal suggests an adjustment of 15 to the maximum number of concurrent processing orders for the order processing node, it will operate according to the proposal.

Step S1324, after the target time domain range is terminated, determining a target resource occupation state of the global production line control system in the target time domain range, where the global production line control system generates a plurality of production control scenario data in the target time domain range, where the target resource occupation state includes: the global production line control system characterizes a value of a resource occupancy under each of the plurality of production control scenario data.

Step S1325, obtaining a control resource occupancy state of the global production line control system, where the control resource occupancy state includes: when the temporary neural network is not invoked, the global production line control system characterizes a characteristic value of resource occupancy under each of the plurality of production control scenario data.

For example, after the target time domain range is over, the resource occupancy state of the global production line control system is recorded. At the same time, the resource occupation state when the temporary neural network is not used is recorded as a comparison.

Step S1326, performing resource occupation influence verification on the temporary neural network based on the target resource occupation state and the comparison resource occupation state, and generating a target resource occupation influence weight of the temporary neural network.

For example, the target resource occupancy state and the reference resource occupancy state may be compared to evaluate the impact of the temporary neural network on system resource occupancy. For example, if the memory footprint of the system is significantly reduced when a temporary neural network is used, then this temporary neural network can be considered to have a positive impact on the resource footprint.

Step S1327, verifying the resource occupation evaluation index of the temporary neural network according to the target resource occupation influence weight of the temporary neural network and the set resource occupation requirement, and generating an application index verification result.

For example, the performance of the temporary neural network can be evaluated according to the influence weight and the set resource occupation requirement of the temporary neural network, and an application index verification result can be generated. For example, if the goal is to minimize memory usage and the temporary neural network successfully helps achieve this goal, then the application index validation results will be positive.

In one possible implementation, the resource occupancy characterization feature value is a ratio between a quantized value of a computational resource used by the occupied resource of the global line control system and a quantized value of all computational resources used by the global line control system.

For example, if the global line control system has a total of 100GB of memory and the currently occupied memory is 20GB, then the resource occupancy characterization feature value is 0.2.

Step S1326 may include:

1. and the global production line control system walks a plurality of production control scene data generated in the target time domain range, the production control scene data walked by the round is output as the production control scene data of the round, the maximum resource occupation value and the minimum resource occupation value corresponding to the occupation resource range in the production control scene data of the round are determined, the average value calculation is carried out on the maximum resource occupation value and the minimum resource occupation value, and the average quantized value of occupation resources corresponding to the production control scene data of the round is generated.

For example, within a target time domain, a plurality of production control scenario data may be generated, each reflecting the state of the system at a particular point in time. For example, if the maximum resource occupation value is 500 units and the minimum resource occupation value is 100 units in all production control scenarios of the present round, max=500, min=100.

And then, calculating an average value of the maximum resource occupation value and the minimum resource occupation value, and generating an average quantized value (Ave) of occupied resources corresponding to the production control scene data of the round. The calculation formula can be: ave= (max+min)/2. In the above example, the average quantization value is (500+100)/2=300 units.

Thus, the average quantized value of the resource occupation corresponding to the production control scene data of the round is obtained, and the average quantized value reflects the average level of the resource occupation condition of the system in the production control of the round.

2. Calculating a difference value between a resource occupation representation characteristic value corresponding to the production control scene data of the round in the target resource occupation state and a resource occupation representation characteristic value corresponding to the production control scene data of the round in the comparison resource occupation state, multiplying the corresponding difference value and an average quantized value of occupied resources corresponding to the production control scene data of the round, and generating the resource occupation optimization completion degree of the temporary neural network under the production control scene data of the round.

And then, calculating a difference value of the characteristic value of the occupied resources when the temporary neural network is used and the temporary neural network is not used, and multiplying the difference value by an average quantized value of the occupied resources to obtain the resource occupancy optimization completion degree of the temporary neural network in the current scene. This resource occupancy optimization completion may help to understand how optimal the temporary neural network is for the resource occupancy in this scenario.

Exemplary, specific calculation steps and formulas may be as follows:

and calculating a difference value (Diff) between a resource occupation representation characteristic value (Target) corresponding to the production Control scene data of the round in the Target resource occupation state and a resource occupation representation characteristic value (Control) corresponding to the production Control scene data of the round in the contrast resource occupation state. The calculation formula can be: diff=target-Control.

And multiplying the calculated difference value (Diff) and an average quantized value (Ave) of the resource occupation corresponding to the production control scene data of the round to generate the resource occupation optimization completion degree (optimizationdevice) of the temporary neural network under the production control scene data of the round. The calculation formula can be: optimizationdeviee=diff Ave.

Thus, the resource occupation optimization completion degree of the temporary neural network under the production control scene data of the round is obtained, and the effect of optimizing the resource occupation of the neural network is reflected. If the value of the optimizationdevice is larger, the neural network has better effect of optimizing the resource occupation in the production control of the round; otherwise, the optimization effect is poor.

3. And calculating a resource optimization effect value of the temporary neural network under the current round of production control scene data based on the determined resource occupation optimization completion degree and the existence time of the current round of production control scene data in the target time domain range.

Considering that different scenes can exist for different time periods, the resource occupation optimization completion degree is multiplied by the scene existence time to obtain the resource optimization effect value of the temporary neural network in the current scene. This resource optimization effect value reflects the overall optimization effect of the temporary neural network on the resource occupancy over the entire lifetime.

Exemplary, specific calculation steps and formulas may be as follows:

the resource occupancy optimization completion level (optimizations device) is determined, which has been calculated in the previous step.

A time of existence (ExistenceTime) of the present round of production control scene data within the target time domain is determined. For example, if the target time domain of interest is one hour and a particular production control scenario persists for 30 minutes during that hour, then ExistenceTime is 0.5 (or 30 minutes).

Based on the two values, a resource optimization effect value (optimizationEffectvalue) of the temporary neural network under the current production control scene data is calculated. The calculation formula can be: optimizationeffect value=optimizationdevice.

Thus, the resource optimization effect value of the temporary neural network under the data of the production control scene of the round is obtained, and the overall effect of optimizing the resource occupation of the neural network under the specific production control scene according to the time of the scene in the target time domain is reflected.

4. And after the plurality of production control scene data are all walked, fusing the resource optimization effect values of the temporary neural network under each production control scene data to generate a target resource optimization effect value of the temporary neural network.

And finally, fusing, such as weighting or adding, the resource optimization effect values of the temporary neural network in all scenes to obtain the target resource optimization effect value of the temporary neural network. The target resource optimization effect value can help to comprehensively evaluate the optimization capacity of the temporary neural network on the resource occupation in the whole target time domain range.

Through the steps, the application index verification result of the temporary neural network, namely the target resource optimization effect value of the temporary neural network, can be obtained. If the target resource optimization effect value meets the preset standard, the temporary neural network can be considered to complete knowledge learning and can be used for actual operation of the environment-friendly glue production control system.

In one possible implementation, the set resource occupancy requirement is used to reflect: and the target resource occupation influence weight of the temporary neural network is not smaller than the preset resource occupation influence weight.

In step S1327, if the target resource occupation influence weight of the temporary neural network is not less than the preset resource occupation influence weight, an application index verification result for reflecting that the temporary neural network passes the application index verification is output. And if the target resource occupation influence weight of the temporary neural network is smaller than the preset resource occupation influence weight, outputting an application index verification result for reflecting that the temporary neural network fails application index verification.

In this embodiment, the setting of the resource occupation requirement reflects the resource optimization effect that the temporary neural network is expected to achieve. For example, a requirement may be set that the temporary neural network should be able to reduce the memory footprint of the system to below 50% of the total memory. This corresponds to setting a preset resource occupancy impact weight.

Then, the target resource occupation influence weight of the temporary neural network (i.e., the resource optimization effect that it actually achieves) and the preset resource occupation influence weight are compared.

If the target resource occupation influence weight of the temporary neural network is not smaller than the preset resource occupation influence weight, the temporary neural network is considered to pass the application index verification. For example, if the temporary neural network successfully reduces the memory footprint of the system to 40% of the total memory, then it meets the resource footprint requirements, and thus the application index validation results are positive.

If the target resource occupation influence weight of the temporary neural network is smaller than the preset resource occupation influence weight, the temporary neural network is considered to be not verified by the application index. For example, if the temporary neural network can only reduce the memory footprint of the system to 60% of the total memory, then it does not meet the resource footprint requirements, and thus the application index validation results are negative.

In one possible implementation, the environmental protection glue production control system correspondingly defines a structural relation model, wherein the structural relation model comprises Q monitoring area members, W characteristic members, R network members and P function members.

One monitoring area member corresponds to one monitoring area, and the monitoring area is used for monitoring the node state value of a production control node in the environment-friendly glue production control system. For a kth feature member, there are H members in the structural relation model, and each member in the H members is mapped to the kth feature member according to a mapping link, where the feature data sequence corresponding to the kth feature member is generated based on the feature data of the member corresponding to the H member, and the H member includes at least one of the following: at least one monitoring area member and at least one feature member other than the kth feature member. k is [1, W ], H is a positive integer.

One network member corresponds to one neural network branch, for the f-th network member, G members exist in the structural relation model, each member in the G members is mapped to the f-th network member according to a mapping link, the neural network branch corresponding to the f-th network member performs network operation based on data corresponding to the G members, and the G members comprise at least one of the following: at least one feature member and at least one network member other than the f-th network member. f is [1, R ], G is a positive integer.

For the t-th function member, u network members exist in the structural relation model, each network member in the u-th network member is mapped to the t-th function member according to a mapping link, the resource occupation decision function corresponding to the t-th function member is configured by the neural network corresponding to each network member in the u-th network member, t belongs to [1, P ], and u belongs to [1, W ].

For example, nodes such as order processing, production scheduling, and inventory management may be considered monitoring area members; the state values (such as the number of orders to be processed, the number of orders to be processed and the like) of all the nodes are regarded as characteristic members; the temporary neural network branch predicting the optimal resource occupation scheme is regarded as a network member; the function that determines the resource occupancy decision is considered a function member.

The monitoring area member corresponds to a monitoring area and is used for monitoring the node state value of the production control node in the environment-friendly glue production control system. For example, the order processing monitoring area may monitor the number of orders currently processed, the amount of orders to be processed, and the like.

The characteristic members correspond to a characteristic data sequence, and are generated by a plurality of members (including at least one monitoring area member and other characteristic members) through mapping links. For example, the kth feature member may be a feature representing the sum of the number of orders currently processed by the order processing node and the amount of orders to be processed.

The network members correspond to a neural network branch that performs network operations based on data of the plurality of members (including at least one characteristic member and other network members) through the mapped links. For example, the f-th network member may be a branch of a neural network that predicts an optimal resource occupancy scheme based on the values of the individual feature members.

The function member corresponds to a resource occupation decision function, which is configured by the neural networks corresponding to the plurality of network members. For example, the t-th function member may be a decision function that determines the system's resource occupancy decision based on the outputs of the various neural network branches.

1. and acquiring an input sample of each neural network branch based on the node state value corresponding to each member field of each production control node in the production scheduling event sequence and the model architecture data of the structural relation model of the environmental protection glue production control system.

2. Training each neural network branch based on the input samples.

For example, in an environmental glue production control system, a sequence of production scheduling events may include status values of various nodes (e.g., order processing, production scheduling, inventory management, etc.) at different points in time, which may be the number of orders currently being processed, the number of orders to be processed, the number of inventory, etc.

The structural relationship model describes relationships between elements in the system, for example, a feature member (representing a particular state, such as the number of orders currently processed) may affect a network member (neural network branch, used to predict the optimal resource usage scenario).

Then, for the training step of each neural network branch, it is assumed that there is one neural network branch whose task is to predict an optimal resource occupation scheme according to the number of orders currently being processed and the number of orders to be processed. In this case, the two state values may be extracted from the production schedule event sequence and taken as input samples for the neural network branches.

The neural network branches will then be trained based on these input samples. The goal of the training is to adjust the parameters of the neural network so that it can accurately predict the optimal resource occupancy scheme based on the number of orders entered. The training process may include common neural network training algorithms such as back propagation, gradient descent, etc.

Thus, through the above steps, a trained neural network branch is obtained, which can predict an optimal resource occupancy scheme based on the current production control node state values. In actual operation, the prediction result can be obtained only by inputting the real-time node state value into the branch of the neural network, thereby helping to make decisions.

In one possible implementation, step S120 may include:

step S121, for each neural network branch in the target neural network, determining a network member corresponding to the each neural network branch from the structural relation model.

Step S122, based on the model architecture data of the structural relation model, the resource occupation decision function corresponding to the function member mapped to the determined network member through the mapping link is used as the resource occupation decision function called by each neural network branch according to the preset network parameter knowledge base.

For example, in the structural relationship model, a network member corresponding to each neural network branch in the target neural network may be found. For example, if there is one neural network branch for predicting the optimal resource occupancy scheme of the order processing node, then this neural network branch corresponds to one network member in the structural relationship model.

And then, based on the model architecture data of the structural relation model and a preset network parameter knowledge base, taking the resource occupation decision function corresponding to the function member mapped by the determined network member through the mapping link as the resource occupation decision function called by each neural network branch. For example, if the knowledge base of preset network parameters reflects that the optimal resource occupancy decision function for the order processing node is a predictive model based on the number of historical orders and the current inventory status, then this decision function is configured to the corresponding neural network branch.

Thus, target network parameters called by the target neural network when making resource occupation decisions are generated. These parameters are invoked when the neural network is running to help the system make optimal resource occupancy decisions.

In one possible implementation, step S140 may specifically include:

step S141, generating a plurality of groups of strategy data based on the simulated annealing algorithm, wherein each group of strategy data comprises node state values corresponding to each adjustable node field.

In this embodiment, the simulated annealing algorithm is a probabilistic search algorithm, which is commonly used to find an approximately optimal solution for a complex problem. Here, multiple sets of policy data are generated using a simulated annealing algorithm, each set of data containing node state values corresponding to respective adjustable node fields (i.e., adjustable parameters of respective production control nodes). For example, a set of policy data may suggest that the maximum number of concurrent processing orders for an order processing module be set to 10, that the priority rules for a production scheduling module be set to first in first out, etc.

By way of example, the following is one possible simulated annealing algorithm process:

1. initializing policy data: first, an initial strategy X needs to be randomly selected. For example, X may suggest that the maximum concurrent processing order number of the order processing module be set to 10 and the priority rule of the production scheduling module be set to first in first out.

2. Defining an objective function and an energy function: assuming that the objective function or energy function is E (X), it is used to measure the quality or fitness of policy X. In this example, E (X) may be designed as the total cost or production time of the system when policy X is executed.

3. Setting an initial temperature and a cooling plan: a higher initial temperature T is set and the temperature is gradually reduced according to a predetermined cooling schedule (e.g., an exponential cooling schedule t=αt, where 0< α < 1).

4. Generating a new strategy: at each step, the current policy X is slightly modified to generate a new policy Y. The manner of modification can be achieved simply by randomly changing its state value at a certain node.

5. Accept or reject new policies: the energies E (Y) and E (X) of the new and current policies Y and X are calculated and a decision is made as to whether to accept the new policy based on their energy differences Δe=e (Y) -E (X) and the current temperature T. If Δe <0 (i.e. the energy of the new strategy is lower), then it is accepted directly; otherwise, accept with probability of E (- ΔE/T) according to Metropolis criterion.

6. The above procedure was repeated: at each iteration, a new policy is generated and a decision is made as to whether or not to accept. The system temperature is then reduced and the next iteration is continued. The algorithm ends when the system "cools" to a preset minimum temperature, or a preset maximum number of iterations is reached.

7. Outputting policy data: the simulated annealing algorithm generates sets of strategy data that are all strategies accepted during the simulated annealing process.

And step S142, carrying out resource occupation decision on the environmental protection glue production control system based on each group of strategy data by utilizing the target neural network for completing knowledge learning, and generating a resource occupation decision result. The resource occupation decision result comprises: the environment-friendly glue production control system predicts the resource occupation state under each group of strategy data.

Each set of policy data is then evaluated using a target neural network that completes knowledge learning. Specifically, the target neural network predicts the resource occupation state of the environmental protection glue production control system according to each group of strategy data. For example, for the aforementioned policy data, the target neural network may predict that the memory footprint of the system is 200mb, the cpu footprint is 50%, etc.

Step S143, selecting a group of strategy data from the plurality of groups of strategy data based on the resource occupation decision result, and taking the strategy data as a resource occupation optimization scheme of the environment-friendly glue production control system.

Next, optimal policy data is selected based on the predicted outcome of the target neural network. The specific selection criteria may depend on optimization objectives such as minimizing overall resource usage, balancing the usage of various resources, etc. For example, if the goal is to minimize memory usage, then policy data with the lowest predicted memory usage will be selected.

Finally, the selected strategy data becomes a resource occupation optimization scheme of the environment-friendly glue production control system. The operation parameters of each module can be adjusted according to the resource occupation optimization scheme, so that the optimal utilization of system resources is realized. For example, the maximum number of concurrent processing orders for the order processing module may be adjusted to 10, the priority rule for the production scheduling module may be set to first in first out, etc.

For example, in one possible implementation manner, in step S140, during the process of loading the target neural network for completing the knowledge learning into the system computing terminal of the environmental protection glue production control system, the target neural network for completing the knowledge learning may be loaded into a neural network resource library, so that the neural network resource library loads the target neural network for completing the knowledge learning into the system computing terminal of the environmental protection glue production control system.

Illustratively, it is assumed that the environmental glue production control system has completed training and optimization of a target neural network (e.g., a neural network predicting an optimal resource occupancy scheme) that is the target neural network that completes knowledge learning through the foregoing steps. This neural network can then be loaded into a neural network repository, which is a place where various neural network models are stored and managed. The system computing terminal may be a server or workstation running an environmental protection glue production control system. The neural network resource library will send the target neural network to the computing terminals so that they can use this neural network to perform the actual production control tasks.

If the target neural network is updated or optimized, the neural network repository will detect the change and then send the new neural network to the system computing terminal, replacing the original version.

FIG. 2 schematically illustrates an intelligent production monitoring system 100 that may be used to implement various embodiments described herein.

For one embodiment, FIG. 2 illustrates an intelligent production monitoring system 100, the intelligent production monitoring system 100 having a plurality of processors 102, a control module (chipset) 104 coupled to one or more of the processor(s) 102, a memory 106 coupled to the control module 104, a non-volatile memory (NVM)/storage device 108 coupled to the control module 104, a plurality of input/output devices 110 coupled to the control module 104, and a network interface 112 coupled to the control module 104.

Processor 102 may include a plurality of single-core or multi-core processors, and processor 102 may include any combination of general-purpose or special-purpose processors (e.g., graphics processors, application processors, baseband processors, etc.). In some alternative implementations, the intelligent production monitoring system 100 can function as a server device such as a gateway as described in the examples herein.

In some alternative embodiments, the intelligent production monitoring system 100 may include a plurality of computer-readable media (e.g., memory 106 or NVM/storage 108) having instructions 114 and a plurality of processors 102 combined with the plurality of computer-readable media configured to execute the instructions 114 to implement the modules to perform the actions described in this disclosure.

For one embodiment, the control module 104 may include any suitable interface controller to provide any suitable interface to one or more of the processor(s) 102 and/or any suitable device or component in communication with the control module 104.

The control module 104 may include a memory controller module to provide an interface to the memory 106. The memory controller modules may be hardware modules, software modules, and/or firmware modules.

Memory 106 may be used, for example, to load and store data and/or instructions 114 for intelligent production monitoring system 100. For one embodiment, memory 106 may include any suitable volatile memory, such as a suitable DKAM. In some alternative embodiments, memory 106 may comprise a double data rate type four synchronous dynamic random access memory.

For one embodiment, the control module 104 may include a plurality of input/output controllers to provide interfaces to the NVM/storage 108 and the input/output device(s) 110.

For example, NVM/storage 108 may be used to store data and/or instructions 114. NVM/storage 108 may include any suitable non-volatile memory (e.g., flash memory) and/or may include any suitable non-volatile storage(s).

NVM/storage 108 may include a storage resource that is physically part of the device on which intelligent production monitoring system 100 is installed, or it may be accessible by the device, or it may not be necessary to be part of the device. For example, NVM/storage 108 may be accessed via input/output device(s) 110 in connection with a network.

Input/output device(s) 110 may provide an interface for intelligent production monitoring system 100 to communicate with any other suitable device, and input/output device 110 may include a communication component, a pinyin component, a sensor component, and the like. The network interface 112 may provide an interface for the intelligent production monitoring system 100 to communicate in accordance with a plurality of networks, and the intelligent production monitoring system 100 may communicate wirelessly with a plurality of components of a wireless network in accordance with any of a plurality of wireless network standards and/or protocols, such as accessing a wireless network in accordance with a communication standard, such as WxFx, 2G, 3G, 4G, 5G, etc., or a combination thereof.

For one embodiment, one or more of the processor(s) 102 may be packaged together with logic of a plurality of controllers (e.g., memory controller modules) of the control module 104. For one embodiment, one or more of the processor(s) 102 may be packaged together with logic of multiple controllers of the control module 104 to form a system in package. For one embodiment, one or more of the processor(s) 102 may be integrated on the same die with logic of multiple controllers of the control module 104. For one embodiment, one or more of the processor(s) 102 may be integrated on the same die with logic of multiple controllers of the control module 104 to form a system-on-chip.

In various embodiments, the intelligent production monitoring system 100 may be, but is not limited to: a desktop computing device or a mobile computing device (e.g., a laptop computing device, a handheld computing device, a tablet, a netbook, etc.), and the like. In various embodiments, intelligent production monitoring system 100 may have more or fewer components and/or different architectures. For example, in some alternative embodiments, the intelligent production monitoring system 100 includes multiple cameras, a keyboard, a liquid crystal display screen (including a touch screen display), a non-volatile memory port, multiple antennas, a graphics chip, an application specific integrated circuit, and speakers.

The foregoing has outlined rather broadly the more detailed description of the present application, wherein specific examples have been provided to illustrate the principles and embodiments of the present application, the description of the examples being provided solely to assist in the understanding of the method of the present application and the core concepts thereof; meanwhile, as those skilled in the art will have modifications in the specific embodiments and application scope in accordance with the ideas of the present application, the present description should not be construed as limiting the present application in view of the above.

Claims

1. An anomaly analysis method applied to an environmental protection glue production control system is characterized by being applied to an intelligent production monitoring system, and comprises the following steps:

the method comprises the steps that a production scheduling event sequence of an environmental protection glue production control system is obtained, the environmental protection glue production control system comprises a plurality of production control nodes, at least one production control node field is correspondingly defined by each production control node, the production scheduling event sequence comprises a node state value corresponding to the node field correspondingly defined by each production control node, the environmental protection glue production control system is provided with at least one node field correspondingly defined by the production control node, and the node field correspondingly defined by the at least one production control node comprises an adjustable node field;

2. The anomaly analysis method for an environmental protection glue production control system of claim 1 wherein the target neural network comprises a plurality of neural network branches responsive to each production control node in the environmental protection glue production control system, the target network parameters comprising: each neural network branch in the target neural network invokes a resource occupation decision function;

3. The anomaly analysis method applied to an environmental protection glue production control system according to claim 2, wherein the temporary neural network comprises temporary neural network branches obtained by knowledge learning of each neural network branch;

4. The anomaly analysis method for an environmental protection glue production control system of claim 2, wherein the environmental protection glue production control system is located within a global production line control system;

5. The anomaly analysis method applied to an environmental protection glue production control system according to claim 4, wherein the characteristic value of resource occupation is a ratio between a quantized value of a computing resource used by an occupied resource of the global production line control system and a quantized value of all computing resources used by the global production line control system;

6. The anomaly analysis method for an environmental protection glue production control system according to claim 5, wherein the set resource occupation requirement is used for reflecting: the target resource occupation influence weight of the temporary neural network is not smaller than the preset resource occupation influence weight;

7. The anomaly analysis method applied to an environmental protection glue production control system according to claim 2, wherein the environmental protection glue production control system correspondingly defines a structural relation model, and the structural relation model comprises Q monitoring area members, W characteristic members, R network members and P function members;

Training each neural network branch based on the input samples.

8. The anomaly analysis method applied to an environmental protection glue production control system according to claim 7, wherein the performing network parameter configuration on the target neural network according to a preset network parameter knowledge base, generating target network parameters called by the target neural network when making a resource occupation decision, includes:

9. The anomaly analysis method applied to the environmental protection glue production control system according to claim 1 or 2, wherein the step of determining the resource occupation optimization scheme of the environmental protection glue production control system by the system computing terminal through the target neural network for completing knowledge learning specifically comprises the following steps:

10. An intelligent production monitoring system comprising a processor and a machine-readable storage medium having stored therein machine-executable instructions loaded and executed by the processor to implement the anomaly analysis method of any one of claims 1-9 applied to an environmental protection glue production control system.