Industrial big data driven intelligent production decision support system and method

Abstract

The invention relates to an intelligent production decision support system and method driven by industrial big data, and the method comprises the steps: constructing a multi-source heterogeneous industrial big data collection system to collect data in real time, carrying out the data cleaning and preprocessing of the collected original data, and carrying out the deep learning-based multi-modal feature fusion module, and constructing a dynamic decision-making model based on reinforcement learning, inputting the fused high-dimensional features into the dynamic decision-making model, and outputting an optimal production decision-making scheme. Through the multi-modal feature fusion module based on deep learning, effective fusion of different types of data such as images, time sequences and structuralization is realized, and the problem of limitation of single data type analysis in the prior art is solved; by constructing the dynamic decision model based on reinforcement learning, adaptive decision making for a complex production environment is realized, and the problems of decision lagging and local optimization in the prior art are solved.

Description

Technical Field

[0001] This invention relates to the field of industrial big data technology, specifically to an intelligent production decision support system and method driven by industrial big data. Background Technology

[0002] With the development of industry and intelligent manufacturing, modern industrial production processes are becoming increasingly complex and dynamic. Factories are equipped with a large number of sensors and devices, generating massive amounts of industrial data, including equipment operation data, environmental parameters, and product quality data. This data is characterized by its multi-source heterogeneity, high dimensionality, and strong temporal sequence, containing rich information on production status and optimization potential.

[0003] Patent CN118568582B relates to an intelligent decision-making method and system based on industrial big data, belonging to the field of intelligent decision-making system technology. Specifically, it includes: collecting energy consumption data in the aluminum smelting process in real time through sensors and the Internet of Things (IoT), and using edge computing and machine learning methods to perform preliminary processing and real-time enhancement of the data; using statistical analysis and data mining algorithms, combined with time series analysis and graph neural networks, to analyze high-energy-consuming links and steps, as well as the correlation between energy consumption and production process parameters, equipment status, and ambient temperature; based on the analysis results, constructing a multi-objective, multi-constraint energy consumption optimization model, and solving it using a hybrid intelligent optimization algorithm to obtain the optimal energy consumption reduction scheme; translating the scheme into practical operation, and evaluating and adjusting the effect through real-time data monitoring and feedback mechanisms, thus realizing intelligent decision-making and optimization of the aluminum smelting process. However, the following problems still exist: The patent proposed methods that can only process single types of data and lack effective methods for fusion of multi-source heterogeneous data. The data standards between different systems are not unified, making it difficult to form a global perspective for production decision-making. Moreover, most of them are based on static rules or offline analysis, which makes it difficult to respond to dynamic changes in the production process in a timely manner. When equipment malfunctions or raw material fluctuations occur, decision adjustments are often significantly delayed. There is a lack of effective verification methods before the implementation of the decision-making scheme, which may lead to production interruption or quality risks. The simulation verification process is not perfect. Summary of the Invention

[0004] The purpose of this invention is to provide an intelligent production decision support system and method driven by industrial big data, so as to solve the problems mentioned in the background art.

[0005] To achieve the above objectives, the present invention provides the following technical solution: An intelligent production decision support system driven by industrial big data is provided, the system comprising: A multi-source heterogeneous industrial big data acquisition system is used to collect real-time data on production equipment operation, environmental monitoring, product quality testing, and supply chain logistics. The processing module is used to clean and preprocess the collected raw data, including outlier detection and handling, missing value imputation, data standardization and time-series alignment. The generation module is used by the deep learning-based multimodal feature fusion module to extract and fuse features from preprocessed multi-source heterogeneous data to generate a unified high-dimensional feature representation. The output module is used to build a dynamic decision-making model based on reinforcement learning, which includes a state awareness module, a policy generation module, and a reward evaluation module. The output module is used to input the fused high-dimensional features into the dynamic decision-making model and output the optimal production decision scheme. The verification module is used to simulate and verify the decision-making scheme through a digital twin system, and to evaluate the feasibility and expected effects of the decision-making scheme. The execution module is used to push the validated decision-making solutions to the production execution system and monitor the execution results in real time. The adjustment module is used to update the parameters of the dynamic decision-making model based on execution feedback data through an online learning mechanism.

[0006] A smart manufacturing decision support method driven by industrial big data is provided based on the above system. The method includes: S1. Construct a multi-source heterogeneous industrial big data acquisition system to collect real-time production equipment operation data, environmental monitoring data, product quality testing data, and supply chain logistics data; S2. Perform data cleaning and preprocessing on the collected raw data, including outlier detection and handling, missing value imputation, data standardization, and time-series alignment. S3. A deep learning-based multimodal feature fusion module extracts and fuses features from preprocessed multi-source heterogeneous data to generate a unified high-dimensional feature representation. S4. Construct a dynamic decision-making model based on reinforcement learning, which includes a state awareness module, a policy generation module, and a reward evaluation module. S5. Input the fused high-dimensional features into the dynamic decision-making model and output the optimal production decision scheme; S6. Simulate and verify the decision-making scheme through a digital twin system to evaluate the feasibility and expected effects of the decision-making scheme; S7. Push the verified decision-making solutions to the production execution system and monitor the execution results in real time; S8. Update the parameters of the dynamic decision-making model based on execution feedback data through an online learning mechanism.

[0007] Preferably, the multi-source heterogeneous industrial big data acquisition system in step S1 includes: Device-level sensor networks are used to collect device operating status data; Environmental monitoring terminals are used to collect environmental parameters such as temperature, humidity, and vibration. Visual inspection systems are used to collect image data on product quality. ERP / MES system interface, used to obtain production plan and material data; Each data source undergoes protocol conversion and data aggregation through an industrial IoT gateway.

[0008] Preferably, the data cleaning in step S2 includes: An outlier detection method based on the isolated forest algorithm is adopted; Use time series forecasting models to impute missing values.

[0009] Preferably, the preprocessing in step S2 includes: Min-max standardization was performed on data of different dimensions. Perform dynamic time warping on asynchronously acquired data.

[0010] Preferably, the multimodal feature fusion module in step S3 adopts a three-level fusion architecture, specifically including: The first stage is single-modal feature extraction, using CNN to process image data and LSTM to process time-series data; The second level is cross-modal feature interaction, which establishes the association between features of different modalities through an attention mechanism; The third level is global feature fusion, which uses a gating mechanism to dynamically adjust the contribution weights of each modality feature.

[0011] Preferably, the reinforcement learning-based dynamic decision-making model in step S4 includes: The state-aware module uses graph neural networks to model the topological relationships between devices; The strategy generation module employs a near-end strategy optimization algorithm. The reward evaluation module comprehensively considers multi-objective optimization of production efficiency, energy consumption, and quality.

[0012] Preferably, the digital twin system in step S6 includes: The system includes a physical entity digital mapping module, a virtual simulation engine, and a virtual-real interactive verification interface. The simulation verification includes discrete event simulation and physics-based numerical simulation.

[0013] Preferably, the decision-making scheme push in step S7 adopts a hierarchical execution strategy, including: Conventional optimization decisions are executed automatically. Major adjustments require manual confirmation before implementation. Emergency and abnormal handling decisions have the highest priority.

[0014] Preferably, the online learning mechanism in step S8 includes: an incremental model update algorithm, a sliding window sample selection strategy, and a model performance drift detection method.

[0015] Beneficial effects By constructing a multi-source heterogeneous industrial big data acquisition system, and through the collaborative work of equipment-level sensor networks, environmental monitoring terminals, visual inspection systems, and ERP / MES system interfaces, a complete data acquisition system covering equipment, environment, quality, and supply chain has been formed. This system enables real-time acquisition and aggregation of all production element data, solving the data silo problem in existing technologies. Through data preprocessing steps such as outlier detection and processing, missing value imputation, data standardization, and time series alignment, the system achieves quality improvement and consistency processing of multi-source heterogeneous data, solving the problem of inconsistent data standards in existing technologies. Through a deep learning-based multimodal feature fusion module, the system effectively integrates different types of data such as images, time series, and structured data, overcoming the limitations of single data type analysis in existing technologies. By constructing a reinforcement learning-based dynamic decision-making model, the system achieves adaptive decision-making in complex production environments, solving the problems of decision lag and local optimization in existing technologies. This invention constructs a complete, intelligent, and adaptive production decision support system through multi-source data fusion, dynamic decision modeling, digital twin verification, and continuous online learning. It effectively solves the problems of data silos, decision lag, local optimization, insufficient verification, and insufficient adaptability in existing technologies, and significantly improves the level of intelligence and decision quality in industrial production. Attached Figure Description

[0016] Figure 1 This is a schematic diagram of the structure of an intelligent production decision support system driven by industrial big data according to the present invention.

[0017] Figure 2 This is a flowchart illustrating an intelligent production decision support method driven by industrial big data according to the present invention. Detailed Implementation

[0018] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0019] Please see Figure 1 As shown, Figure 1-2This is a schematic diagram of the structure of an intelligent production decision support system driven by industrial big data according to the present invention. The present invention also proposes an intelligent production decision support system driven by industrial big data, the system comprising: The data acquisition module 201 is used to build a multi-source heterogeneous industrial big data acquisition system to collect real-time production equipment operation data, environmental monitoring data, product quality testing data, and supply chain logistics data. The processing module 202 is used to clean and preprocess the collected raw data, including outlier detection and processing, missing value imputation, data standardization and time-series alignment; The generation module 203 is used as a deep learning-based multimodal feature fusion module to extract and fuse features from preprocessed multi-source heterogeneous data to generate a unified high-dimensional feature representation. Module 204 is used to build a dynamic decision-making model based on reinforcement learning. The model includes a state awareness module, a policy generation module, and a reward evaluation module. Output module 205 is used to input the fused high-dimensional features into the dynamic decision model and output the optimal production decision scheme; The verification module 206 is used to simulate and verify the decision-making scheme through the digital twin system, and to evaluate the feasibility and expected effects of the decision-making scheme. The execution module 207 is used to push the verified decision-making plan to the production execution system and monitor the execution effect in real time. Adjustment module 208 is used to update the parameters of the dynamic decision model based on execution feedback data through an online learning mechanism.

[0020] This invention proposes an intelligent production decision support method driven by industrial big data, the method comprising: S1. Construct a multi-source heterogeneous industrial big data acquisition system to collect real-time production equipment operation data, environmental monitoring data, product quality testing data, and supply chain logistics data; S2. Perform data cleaning and preprocessing on the collected raw data, including outlier detection and handling, missing value imputation, data standardization, and time-series alignment. S3. A deep learning-based multimodal feature fusion module extracts and fuses features from preprocessed multi-source heterogeneous data to generate a unified high-dimensional feature representation. S4. Construct a dynamic decision-making model based on reinforcement learning, which includes a state awareness module, a policy generation module, and a reward evaluation module. S5. Input the fused high-dimensional features into the dynamic decision-making model and output the optimal production decision scheme; S6. Simulate and verify the decision-making scheme through a digital twin system to evaluate the feasibility and expected effects of the decision-making scheme; S7. Push the verified decision-making solutions to the production execution system and monitor the execution results in real time; S8. Update the parameters of the dynamic decision-making model based on execution feedback data through an online learning mechanism.

[0021] Optionally, the multi-source heterogeneous industrial big data acquisition system in step S1 includes: Device-level sensor networks are used to collect device operating status data; Environmental monitoring terminals are used to collect environmental parameters such as temperature, humidity, and vibration. Visual inspection systems are used to collect image data on product quality. ERP / MES system interface, used to obtain production plan and material data; Each data source undergoes protocol conversion and data aggregation through an industrial IoT gateway.

[0022] It should be noted that the multi-source heterogeneous industrial big data acquisition system includes an equipment-level sensor network, environmental monitoring terminals, a vision inspection system, and an ERP / MES system interface. The equipment-level sensor network is deployed on various production equipment to collect equipment operating parameters such as vibration, temperature, and current, with the sampling frequency set according to the criticality of the equipment. The environmental monitoring terminals are distributed throughout the workshop to collect environmental parameters such as temperature, humidity, light, and noise, with a sampling interval of 1 minute. The vision inspection system is installed at key quality inspection stations, using industrial cameras to collect images of product surfaces with a resolution of no less than 5 megapixels. The ERP / MES system interface obtains information such as production plans and material inventory through standard APIs. All data is converted through an industrial IoT gateway and then transmitted to the data center via the MQTT protocol.

[0023] Optionally, the data cleaning and preprocessing in step S2 specifically includes: An outlier detection method based on the isolated forest algorithm is adopted; Use time series forecasting models to impute missing values; Min-max standardization was performed on data of different dimensions. Perform dynamic time warping on asynchronously acquired data.

[0024] It should be noted that outlier detection employs an improved isolated forest algorithm, which identifies outliers by constructing multiple isolated trees. The branching strategy and sampling method have been adjusted to suit the characteristics of industrial data. Missing value imputation uses an LSTM time series prediction model, which uses data from preceding and following time periods to predict missing values. Data standardization uses the min-max method to map data of different dimensions to the [0,1] interval. Time series alignment uses a dynamic time warping algorithm to address the issue of asynchronous data acquisition times from different devices. The preprocessed data is stored in a time series database for subsequent analysis.

[0025] Optionally, the multimodal feature fusion module in step S3 adopts a three-level fusion architecture, specifically including: The first stage is single-modal feature extraction, using CNN to process image data and LSTM to process time-series data; The second level is cross-modal feature interaction, which establishes the association between features of different modalities through an attention mechanism; The third level is global feature fusion, which uses a gating mechanism to dynamically adjust the contribution weights of each modality feature.

[0026] It should be noted that in the first-level feature extraction, visual features of image data are extracted through a ResNet-18 network, temporal features of time-series data are extracted through a bidirectional LSTM network, and statistical features of structured data are extracted through a fully connected network. The second-level cross-modal interaction uses a multi-head attention mechanism to calculate the correlation weights between features of different modalities. The third-level global fusion uses a gated fusion unit to dynamically adjust the contribution of each modal feature, and finally outputs a unified feature vector of 512 dimensions. This module adopts an end-to-end approach in the training phase, and the loss function is designed as a weighted sum of the downstream task loss and the feature reconstruction loss.

[0027] Optionally, the reinforcement learning-based dynamic decision-making model in step S4 includes: The state-aware module uses graph neural networks to model the topological relationships between devices; The strategy generation module employs a near-end strategy optimization algorithm. The reward evaluation module comprehensively considers multiple objectives, including production efficiency, energy consumption, and quality optimization.

[0028] It should be noted that the state awareness module constructs the device feature vectors into a graph structure, where nodes represent devices and edges represent material flow or information flow relationships, and uses a graph attention network for embedding learning; the policy generation module adopts the PPO algorithm, which includes two sub-networks: Actor and Critic. The Actor network outputs the action probability distribution, and the Critic network evaluates the state value; the reward function of the reward evaluation module is: R = α P+β Q-γ E, where P is the production efficiency index, Q is the product quality index, E is the energy consumption index, and α, β, and γ are adjustable weight coefficients; in the multi-objective reward function of the reward evaluation module, the production efficiency index P is calculated as the output of good products per unit time, the quality index Q is the product qualification rate, the energy consumption index E is the ratio of process energy consumption to the benchmark value, and the weight coefficients α, β, and γ are determined by the analytic hierarchy process and can be adjusted according to the production strategy.

[0029] Optionally, the digital twin system in step S6 includes: The system includes a physical entity digital mapping module, a virtual simulation engine, and a virtual-real interactive verification interface. The simulation verification includes discrete event simulation and physics-based numerical simulation.

[0030] It should be noted that the physical entity digitization mapping module constructs a virtual model corresponding to the real equipment in a 1:1 ratio through 3D modeling and parametric configuration; the virtual simulation engine supports discrete event simulation and physics-based numerical simulation, the former simulating production logistics and scheduling processes, and the latter simulating physical phenomena in the processing process based on the finite element method; the virtual-real interaction verification interface converts decision scheme parameters into simulation inputs and collects simulation result data; the simulation verification process adopts a stepwise approximation strategy, first performing simplified simulation for rapid evaluation, and then performing high-precision simulation for key decisions.

[0031] Optionally, the decision-making scheme push in step S7 adopts a hierarchical execution strategy, including: Conventional optimization decisions are executed automatically. Major adjustments require manual confirmation before implementation. Emergency and abnormal handling decisions have the highest priority.

[0032] It should be noted that the decision execution adopts a three-level strategy: routine optimization decisions (such as fine-tuning of process parameters) are automatically issued by the system to the equipment controller for execution; major adjustment decisions (such as changing the processing route) require confirmation by the workshop director on the visual interface; emergency exception handling decisions (such as emergency equipment shutdown) have the highest priority and can directly interrupt the current operation; in the online learning mechanism, incremental updates use an elastic weight consolidation algorithm to prevent new data from overwriting old knowledge; the sliding window retains valid samples from the most recent 3 months; performance drift detection is achieved by monitoring the accuracy changes of the model on the validation set, and model retraining is triggered when the accuracy exceeds a threshold.

[0033] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0034] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. An intelligent production decision support system driven by industrial big data, characterized in that, The system includes: A multi-source heterogeneous industrial big data acquisition system is used to collect real-time data on production equipment operation, environmental monitoring, product quality testing, and supply chain logistics. The processing module is used to clean and preprocess the collected raw data, including outlier detection and handling, missing value imputation, data standardization and time-series alignment. The generation module is used by the deep learning-based multimodal feature fusion module to extract and fuse features from preprocessed multi-source heterogeneous data to generate a unified high-dimensional feature representation. The output module is used to build a dynamic decision-making model based on reinforcement learning, which includes a state awareness module, a policy generation module, and a reward evaluation module. The output module is used to input the fused high-dimensional features into the dynamic decision-making model and output the optimal production decision scheme. The verification module is used to simulate and verify the decision-making scheme through a digital twin system, and to evaluate the feasibility and expected effects of the decision-making scheme. The execution module is used to push the validated decision-making solutions to the production execution system and monitor the execution results in real time. The adjustment module is used to update the parameters of the dynamic decision-making model based on execution feedback data through an online learning mechanism.

2. A method for an industrial big data-driven intelligent production decision support system as described in claim 1, characterized in that, The method includes the following steps: Step S1: Construct a multi-source heterogeneous industrial big data acquisition system to collect real-time production equipment operation data, environmental monitoring data, product quality testing data, and supply chain logistics data; Step S2: Perform data cleaning and preprocessing on the collected raw data, including outlier detection and handling, missing value imputation, data standardization, and time-series alignment; Step S3: The deep learning-based multimodal feature fusion module extracts and fuses features from the preprocessed multi-source heterogeneous data to generate a unified high-dimensional feature representation. Step S4: Construct a dynamic decision-making model based on reinforcement learning, which includes a state awareness module, a policy generation module, and a reward evaluation module. Step S5: Input the fused high-dimensional features into the dynamic decision model and output the optimal production decision scheme; Step S6: Simulate and verify the decision-making scheme through a digital twin system to evaluate its feasibility and expected effects; Step S7: Push the verified decision-making solution to the production execution system and monitor the execution effect in real time; Step S8: Update the parameters of the dynamic decision model based on the execution feedback data through an online learning mechanism.

3. The method of an intelligent production decision support system driven by industrial big data according to claim 2, characterized in that: The multi-source heterogeneous industrial big data acquisition system in step S1 includes: Device-level sensor networks are used to collect device operating status data; Environmental monitoring terminals are used to collect environmental parameters such as temperature, humidity, and vibration. Visual inspection systems are used to collect image data on product quality. ERP / MES system interface, used to obtain production plan and material data; All data are converted and aggregated through an industrial IoT gateway.

4. The method of an intelligent production decision support system driven by industrial big data according to claim 2, characterized in that: The data cleaning in step S2 includes: An outlier detection method based on the isolated forest algorithm is adopted; Use time series forecasting models to impute missing values.

5. The method of an intelligent production decision support system driven by industrial big data according to claim 2, characterized in that: The preprocessing in step S2 includes: Min-max standardization was performed on data of different dimensions. Perform dynamic time warping on asynchronously acquired data.

6. The method of an intelligent production decision support system driven by industrial big data according to claim 2, characterized in that: The multimodal feature fusion module in step S3 adopts a three-level fusion architecture, specifically including: The first stage is single-modal feature extraction, using CNN to process image data and LSTM to process time-series data; The second level is cross-modal feature interaction, which establishes the association between features of different modalities through an attention mechanism; The third level is global feature fusion, which uses a gating mechanism to dynamically adjust the contribution weights of each modality feature.

7. The method of an intelligent production decision support system driven by industrial big data according to claim 2, characterized in that: The reinforcement learning-based dynamic decision-making model mentioned in step S4 includes: The state-aware module uses graph neural networks to model the topological relationships between devices; The strategy generation module employs a near-end strategy optimization algorithm. The reward evaluation module comprehensively considers multi-objective optimization of production efficiency, energy consumption, and quality.

8. The method of an intelligent production decision support system driven by industrial big data according to claim 2, characterized in that: The digital twin system in step S6 includes: The system includes a physical entity digital mapping module, a virtual simulation engine, and a virtual-real interactive verification interface. The simulation verification includes discrete event simulation and physics-based numerical simulation.

9. The method of an intelligent production decision support system driven by industrial big data according to claim 2, characterized in that: The decision-making scheme push in step S7 adopts a hierarchical execution strategy, including: Conventional optimization decisions are executed automatically. Major adjustments require manual confirmation before implementation. Emergency and abnormal handling decisions have the highest priority.

10. The method of an intelligent production decision support system driven by industrial big data according to claim 2, characterized in that: The online learning mechanism in step S8 includes: an incremental model update algorithm, a sliding window sample selection strategy, and a model performance drift detection method.

Images