# US 2022/0413455 A1 · Adaptive-learning intelligent scheduling unified computing frame and system for industrial personalized customized production

> Agents: this data is available as JSON from the patentgrep API and MCP server — see https://patentgrep.com/llms.txt

- Number: US 2022/0413455 A1
- Document ID: US20220413455A1
- Status: Published application
- Kind: A1
- Application number: 17/781,370
- Filed: 2020-11-13
- Published: 2022-12-29
- Earliest claimed: 2020-11-13
- Assignees: Zhejiang University
- Applicants: Zhejiang University
- Inventors: ZAIXING HE; YUE CHAO; DONGSHENG YANG; ZHILE YANG
- Web page: https://patentgrep.com/patent/US20220413455A1

## CPC

- G05B 19/41865: {characterised by job scheduling, process planning, material flow}
- G05B 13/027: {using neural networks only}
- G06F 9/50: Allocation of resources, e.g. of the central processing unit [CPU]
- G06F 18/2155: {characterised by the incorporation of unlabelled data, e.g. multiple instance learning [MIL], semi-supervised techniques using expectation-maximisation [EM] or naïve labelling}
- G06F 18/217: {Validation; Performance evaluation; Active pattern learning techniques}
- G05B 2219/32247: Real time scheduler
- G05B 2219/32252: Scheduling production, machining, job shop
- G05B 2219/32329: Real time learning scheduler, uses ANN, fuzzy
- G05B 2219/32334: Use of reinforcement learning, agent acts, receives reward
- G05B: CONTROL OR REGULATING SYSTEMS IN GENERAL; FUNCTIONAL ELEMENTS OF SUCH SYSTEMS; MONITORING OR TESTING ARRANGEMENTS FOR SUCH SYSTEMS OR ELEMENTS (systems for controlling or regulating non-electric variables G05D; systems for regulating electric or magnetic variables G05F; control devices or systems insofar as characterised by mechanical features only G05G)
- G06F: ELECTRIC DIGITAL DATA PROCESSING (computer systems based on specific computational models G06N)

## Abstract

The present invention discloses an adaptive-learning intelligent scheduling unified computing frame and system for industrial personalized customized production. Based on a deep neural network and reinforcement learning, a most suitable optimization algorithm is selected by automatic decision-making for a global customized production task with an industrial big data module at the bottom as an information basis, and a global optimal static scheduling plane is generated; a current dynamic event in a factory are monitored in real time; if no dynamic event requiring dynamic scheduling optimization is monitored, the global optimal static plan is executed sequentially; when a dynamic event impact requiring dynamic scheduling optimization is monitored, information of the current dynamic event is interpreted and classified, and corresponding optimization algorithms are automatically selected for dynamic scheduling optimization according to different types of dynamic events; and a dynamic scheduling scheme is evaluated by a subsequent module, an optimization scheme is regenerated or a most suitable optimization algorithm is automatically decided based on the scheme according to an evaluation result, and an equipment deployment sequence is generated for an automatic deployment. Considering the features of complicated procedures, a large amount of customization information and the high frequency of diversified dynamic events in personalized customized production, the present invention provides the adaptive-learning intelligent scheduling unified computing frame and system for industrial personalized customized production, that adopt two steps in the three aspects of static scheduling planning, dynamic scheduling planning and equipment deployment based on deep learning, that is, targeted optimization is performed after classification, thus improving the optimization efficiency and effect; and the system better fits the features of personalized customized production, and can effectively improve the efficiency of personalized customized production and minimize manual decision-making costs.

## Claims

### Claim 1

An adaptive-learning intelligent scheduling unified computing frame and system for industrial personalized customized production, comprising the following steps:
S1: the system reading information in a real-time updated industrial big data module; prior to production, an optimal static scheduling planning module in the system generating an optimal static scheduling plan according to customization task information, equipment information and material information in the real-time updated industrial big data module; and when production is started, the system invoking a dynamic event interpreting module to dynamically monitor an event queue until a processable dynamic event is read;
S2: automatically classifying dynamic events by the dynamic event interpreting module based on a deep reinforcement learning technique according to data information in the industrial big data module, generating classification labels, and deciding specific contents of a corresponding optimization objective and constraint according to the classification labels;
S3: automatically receiving an optimization objective decision scheme generated in the step S2, automatically selecting different optimization algorithms based on the deep reinforcement learning technique according to the classification labels and the optimization objective and constraint, selecting a suitable method to optimize parameters of the algorithms until the corresponding optimization objective and constraint are met, and forming a final dynamic scheduling optimization scheme, by a dynamic scheduling optimization module;
S4: automatically receiving the dynamic scheduling optimization scheme generated in the step S3, evaluating the optimization scheme in the aspects of efficiency, energy consumption, deployment complexity and stability by means of a deep neural network to obtain a score of the optimization scheme, optimizing an equipment scheduling deployment in a spatial scale if the score is higher than a score standard, automatically selecting a corresponding deployment optimization algorithm based on the optimization scheme generated in the step S3 according to a current industrial production condition and the equipment distribution condition of a production line, and generating an equipment scheduling sequence, by a scheduling deployment and evaluation module; and
S5: performing an automatic deployment by the system according to the deployment sequence generated in the step S4, receiving equipment scheduling change information in real time by the industrial big data module, and updating the current industrial production environment, such that a multi-module collaborative dynamic decision chain is formed by four modules of the frame, and decision-making in the whole process is completed by a computer without participation of workers; and after one cycle is completed, the system avoids the influence of a current dynamic event, continuously adopts the optimized optimal static plan and invokes the dynamic event interpreting module to dynamically monitor the event queue to repeat the step S2 through the step S5 until all personalized customized production task are completed.

### Claim 2 (depends on 1)

The adaptive-learning intelligent scheduling unified computing frame and system for industrial personalized customized production according to claim 1, wherein the optimal static scheduling planning module is used for acquiring information provided by the industrial big data module, extracting features of an industrial production environment by semi-supervised learning based on the deep neural network, performing training in terms of information labeled in advance and other non-labeled information to enable a semi-supervising model to regress to learn how to classify information according to features to allow the system to decide corresponding static objective weight parameters and constraints in different production environments, and selecting a most suitable static optimization algorithm according to decision contents, such that the system is adaptable to different complicated conditions during personalized customized production, and an advanced offline generation method is adopted in the whole process, such that an optimal static plan of a global production task is obtained before operation of the production line; the system comprises a static optimization algorithm library containing multiple classic static optimization algorithms including a quality prediction method, an optimization algorithm, a heuristic method, a maximum inherit algorithm, a genetic algorithm and derivative algorithms thereof, users are able to update the algorithm library according to personal requirements and features of an actual production environment, and after the algorithm library is updated, the system takes personal algorithms of the users into account in the decision-making process;
static objectives include a shortest maximum completion time, lowest energy consumption and a lowest production cost, and decision of the static objectives mainly comprises analyzing and optimizing a comprehensive weight of multiple static objectives and generating a static multi-objective comprehensive weight parameter;
the optimal static scheduling plan is a scheduling planning scheme, in conformity with the shortest maximum completion time, the lowest energy consumption and the lowest production cost from the macroscopic aspect, generated based on a global personalized customized production task; an optimization algorithm of the optimal static scheduling plan is independently decided and selected according to the current environment and the personalized customized production task and is an important constituent part of intelligent production line macro strategy modeling.

### Claim 3 (depends on 1)

The adaptive-learning intelligent scheduling unified computing frame and system for industrial personalized customized production according to claim 1, wherein the dynamic event interpreting module is used for monitoring production data information provided by the industrial big data module, monitoring dynamic events requiring dynamic scheduling optimization, classifying the dynamic events according to contents of the dynamic events based on a deep reinforcement learning technique, enabling an intelligent agent to observe features of the dynamic events in the current environment with current industrial production information as a reinforcement learning environment of the intelligent agent, and setting a corresponding stimulus to enhance dynamic event classification; the classification labels include time deviation of the production line, machine fault, material problem and contract event; a multi-objective dynamic weight parameter and a corresponding constraint of dynamic scheduling optimization under the current event are automatically decided according to current industrial production environment data provided by the industrial big data module, and an event classification result;
dynamic objectives include a shortest maximum completion time, lowest energy consumption, a lowest production cost, a highest adjustment flexibility of production plans, and a highest scheduling stability (that is, a lowest variation with respect to the optimal static scheduling plan), and decision of the dynamic objectives mainly comprises analyzing and optimizing a comprehensive weight of multiple dynamic objectives and generating a final dynamic multi-objective comprehensive weight;
the form of the final dynamic multi-objective comprehensive weight is expressed by the following formula:
minZ=α·fnorm(T)+β·fnorm(E)+γ·fnorm(C)+(1−δ·fnorm(P))+(1−ε·fnorm(Q))
wherein, α represents a time weight parameter; β represents an energy consumption weight parameter; γ represents a cost weight parameter; δ represents a flexibility weight parameter; ε represents a stability weight parameter; T represents a workshop production time, E represents workshop production energy consumption, C presents a workshop production cost, P represents the flexibility of a scheduling scheme, Q represents the stability of the scheduling scheme, fnorm(χ) represents a normalization function for unifying the dimensions of five indicators including time, energy consumption, cost, flexibility and stability, the value of the function value is between 0 and 1, and values of the weight parameters are also between 0 and 1.

### Claim 4 (depends on 1)

The adaptive-learning intelligent scheduling unified computing frame and system for industrial personalized customized production according to claim 1, wherein the dynamic scheduling optimization module comprises a dynamic optimization algorithm selection strategy model and an optimization algorithm parameter adaptive learning method; considering that there are a variety of dynamic events, the rule of which is hard to determine, the system generates the dynamic optimization algorithm selection strategy model by deep reinforcement learning and adaptively matches optimization algorithms and parameter optimization methods by supervised learning;
the dynamic optimization algorithm selection strategy model enables the intelligent agent to observe the difference between optimization results of different optimization algorithms with various optimization algorithms, a current event classification label and a current industrial production environment as a reinforcement learning environment of the intelligent agent, sets a corresponding stimulus to enhance optimization algorithm selection, and decides dynamic scheduling optimization algorithms corresponding to different dynamic events according to different conditions; and the system comprises a dynamic optimization algorithm library containing multiple classic dynamic optimization algorithms including a multi-agent algorithm, a simulated annealing algorithm, a hybrid particle swarm algorithm, a tabu search algorithm, a genetic algorithm and variants thereof, users are able to update the algorithm library according to personal requirements and features of the actual production environment, and after the algorithm library is updated, the system takes personal algorithms of the users into account in the decision-making process;
features of different optimization algorithms are extracted by deep learning, specifically by supervised learning based on the deep neural network, to automatically classify the optimization algorithms, and a suitable parameter optimization method is selected from the parameter optimization algorithm library; and the users are able to update the parameter optimization algorithm library, and the system is able to automatically extract personal algorithm features for classification.

### Claim 5 (depends on 1)

The adaptive-learning intelligent scheduling unified computing frame and system for industrial personalized customized production according to claim 1, wherein the scheduling deployment and planning module comprises an adaptive scheme evaluation strategy based on deep reinforcement learning, calculates the score of a received optimization scheme, and performs a next scheduling deployment if the score is higher a specified score threshold; in the deployment process, features of the production environment and the scheduling scheme are extracted by deep learning, with the current industrial production environment and scheduling scheme as an environment, with reference to the static scheduling scheme or dynamic scheduling scheme generated in the previous step in terms of the current industrial production environment information, and weight parameters and relevant constraints of a current deployment optimization objective are independently decided by reinforcement learning, a suitable optimization algorithm is selected, and a deployment sequence is generated finally; the system comprises a deployment optimization algorithm library containing multiple classic equipment deployment optimization algorithms including an adaptive genetic algorithm, an ant colony optimization algorithm, a multi-objective genetic algorithm, a particle swarm algorithm and derivative algorithms thereof, users are able to update the algorithm library according to personal requirements and features of the actual production environment, and after the algorithm library is updated, the system takes personal algorithms of the users into account in the decision-making process; and optimization objectives taken into account in the deployment optimization process mainly include transport expenditure, equipment delay expenditure, production capacity, and deployment complexity, and the system mainly makes decision on weight proportions of multiple optimization objectives.

### Claim 6 (depends on 1)

The adaptive-learning intelligent scheduling unified computing frame and system for industrial personalized customized production according to claim 1, wherein the real-time updated industrial big data module comprises all data information generated and monitored in the actual industrial production process, including dynamic event information, material information, production equipment information, personalized customized production task information, and current factory operating information.

### Claim 7 (depends on 2)

The adaptive-learning intelligent scheduling unified computing frame and system for industrial personalized customized production according to claim 2, wherein the multi-module collaborative dynamic decision chain is a closed-loop decision route formed by the industrial big data module capable of receiving external data in real time, the dynamic event interpreting module, the dynamic scheduling optimization module and the scheduling deployment and evaluation module of the unified computing frame, and all decision-making tasks are automatically completed by the computer in the whole process if no advanced decision instruction is manually sent; and by means of the closed-loop decision chain, the unified computing frame is able to complete automatic planning and deployment of equipment and procedures on an intelligent production line only by receiving data.

### Claim 8 (depends on 2)

The adaptive-learning intelligent scheduling unified computing frame and system for industrial personalized customized production according to claim 2, wherein the micro strategy modeling refers to considering the operation and optimization of an intelligent production line from the production environment of a whole factory rather than only focusing on the current procedure or event, and still paying attention to the whole production environment and the personalized customized production task in the process of handing a random dynamic event, and includes micro strategy modeling in the aspect of production tasks and micro strategy modeling in the aspects of production factors;
the micro strategy modeling in the aspect of production tasks means that the whole frame automatically selects a static optimization algorithm according to a global production task provided by users after task initialization, and generates an optimal static plan based on the global production task, and the optimal static plan will be referred in subsequent local dynamic scheduling optimization and adjustment to minimize a change to the global optimal static plan to ensure that the whole frame is able to take the global production task into account in the operating process;
the micro strategy modeling in the aspect of production factors means that information in the industrial big data module is updated in real time in the operating process of the computing frame, and other modules uses current information at any time, wherein the information includes all information of the production environment of the whole factory; and the frame provided is not confined to information of devices or procedures involved in the current event, but also refers to information of the whole current production environment, thus taking the influence on other production procedures and devices into account.
