Intelligent energy-saving optimization control system for production and processing based on data analysis

Abstract

The application discloses a production and processing intelligent energy-saving optimization control system based on data analysis, and relates to the technical field of industrial intelligent manufacturing and energy-saving control.The application integrates multi-modal data acquisition, space-time feature extraction, distributed reinforcement learning and digital twin verification technology.The system adopts a three-level architecture of edge-fog-cloud, fuses multi-source data such as temperature and pressure by dynamically adjusting the sampling frequency (10ms-5s), extracts the space-time joint features of the equipment state by using a 3D convolutional neural network and a bidirectional LSTM, and realizes local real-time decision and global energy efficiency optimization collaboration by using an improved TD3 algorithm.The digital twin platform ensures the safety of the control strategy through a three-level verification mechanism (sensor level, subsystem level and whole line level), realizes dynamic load prediction and abnormal state matching in combination with an energy consumption knowledge graph database.The system supports LoRaWAN low-latency transmission, has adaptive noise reduction capability and a fuzzy PID abnormal switching mechanism, and forms a complete closed-loop optimization system.

Description

Technical Field

[0001] This invention belongs to the field of industrial intelligent manufacturing and energy-saving control technology, and in particular relates to an intelligent energy-saving optimization control system for production and processing based on data analysis. Background Technology

[0002] In today's manufacturing industry, energy consumption in production processes is becoming increasingly prominent. Traditional production equipment control methods often lack comprehensive perception and in-depth analysis of equipment operating status, environmental variables, and energy flow data, leading to serious energy waste.

[0003] Specifically, energy consumption optimization in traditional production and processing systems suffers from the following problems: Limited data acquisition: Relying on a single sensor (such as a current sensor) cannot comprehensively reflect equipment status and ignores the impact of multi-dimensional parameters such as temperature and vibration on energy consumption; Rigid control strategies: PID control based on fixed thresholds is difficult to adapt to dynamic operating conditions (such as sudden load changes or equipment aging), leading to low energy efficiency; Lack of verification system: Control commands are directly issued to physical equipment, lacking a virtual verification process, posing safety risks; Insufficient global optimization: Independent optimization of subsystems leads to suboptimal overall energy efficiency, lacking a cross-device collaborative mechanism. With the development of industrial IoT, big data analytics, and artificial intelligence technologies, data-driven intelligent optimization has become key to solving these problems. Summary of the Invention

[0004] The purpose of this invention is to provide a data analysis-based intelligent energy-saving optimization control system for production and processing, which can comprehensively collect multimodal data from production equipment, analyze and process the data through intelligent algorithms, generate optimized control commands, and execute them after rigorous verification, thereby achieving energy-saving optimization of the production and processing process.

[0005] To solve the above-mentioned technical problems, the present invention is achieved through the following technical solution: This invention is a data analysis-based intelligent energy-saving optimization control system for production and processing, comprising: The multimodal data acquisition and fusion module integrates temperature sensors, pressure sensors, flow sensors, vibration sensors, and power quality sensors to collect and integrate physical parameters, environmental variables, and energy flow data of production equipment in real time. A spatiotemporal feature extraction engine, connected to the multimodal data acquisition and fusion module, is used to extract spatiotemporal joint features of the device status; The distributed reinforcement learning controller adopts a dual-delay deep deterministic policy gradient algorithm framework, which includes a local decision-maker and a central coordinator, and generates optimized control commands based on the spatiotemporal joint features. The digital twin verification platform integrates a computational fluid dynamics simulation module and a discrete event simulation module, and outputs the optimized control commands to the actuator after performing safety verification. The system adopts a three-level computing architecture of edge-fog-cloud, in which the edge layer performs data preprocessing, the fog computing layer performs feature extraction, and the cloud computing layer completes strategy optimization.

[0006] As a preferred embodiment of the present invention, the multimodal data acquisition and fusion module includes: The dynamic sampling frequency adjustment unit adaptively adjusts the sampling period within the range of 10ms-5s according to the equipment load rate. The data cleaning pipeline uses an improved median filtering algorithm to eliminate impulse noise. The feature extraction unit uses wavelet packet decomposition technology to generate a 128-dimensional feature vector, which integrates the frequency domain features of vibration signals, the time domain features of power parameters, and the spatial distribution features of environmental parameters.

[0007] As a preferred embodiment of the present invention, the dynamic sampling adjustment unit uses a fuzzy PID algorithm to control the sampling frequency, thereby achieving the following functions: When the device load rate is less than 60%, switch to 1Hz low-frequency sampling mode; When abnormal vibration is detected, the 50kHz high-frequency sampling mode is triggered; The sampled data is transmitted via the LoRaWAN protocol with a latency of <20ms.

[0008] As a preferred embodiment of the present invention, the spatiotemporal feature extraction engine includes: Spatial feature extraction layer: a 3D convolutional neural network layer that extracts the spatial distribution features of the device status; Temporal feature extraction layer: a bidirectional long short-term memory network layer that captures the temporal evolution of feature parameters; Feature selection layer: The weight coefficients of each feature channel are dynamically calculated using the Pearson correlation coefficient.

[0009] As a preferred embodiment of the present invention, the distributed reinforcement learning controller employs an improved dual-delay deep deterministic policy gradient algorithm framework, comprising: Local decision-makers, deployed in each processing unit, execute real-time control decisions based on the current equipment status; The central coordinator coordinates the decision-makers of each unit through the Nash equilibrium algorithm to achieve optimal global energy efficiency.

[0010] As a preferred embodiment of the present invention, the digital twin verification platform includes: The computational fluid dynamics simulation module simulates the thermodynamic processes of equipment; The discrete event simulation module constructs a mapping relationship between production cycle time and energy consumption; The security verification sandbox simulates the effects of control strategies in a virtual environment. The platform is equipped with a three-level verification mechanism and switches to a fuzzy PID control mode in case of abnormalities.

[0011] As a preferred embodiment of the present invention, the three-level verification mechanism of the digital twin verification platform includes: Sensor-level verification: Inject ±5% noise data into the digital twin to verify the robustness of the control strategy; Subsystem-level verification: Simulate equipment failure and test the response time of safety interlocks; Full-line verification: Run 1000 hours of simulation on a virtual production line to verify the compliance rate of energy consumption indicators.

[0012] As a preferred embodiment of the present invention, it also includes an energy consumption knowledge graph database: Construct a spectral feature library for 23 typical process modes; Establish a cosine similarity matching matrix for 78 types of abnormal states; To achieve dynamic load forecasting.

[0013] As a preferred embodiment of the present invention, the method of using the system includes the following steps: S1. Dynamic Sampling Control: Dynamically adjusts the sampling frequency to acquire multi-source data, and performs wavelet packet decomposition and outlier detection; S2. Data preprocessing: Median filtering is performed at the edge nodes to generate 128-dimensional standardized feature vectors; S3. Feature Engineering: Extract spatiotemporal joint features using hybrid neural networks and calculate the correlation coefficient matrix between features; S4. Strategy Generation: The TD3 algorithm is used to generate the control strategy, with an initial exploration rate ε=0.3 that linearly decays to 0.05; S5. Verification and Execution: Verify the security of the policy on the digital twin platform, and issue encrypted control commands after passing the verification.

[0014] The present invention has the following beneficial effects: This invention, through a multimodal data acquisition and fusion module, can comprehensively and in real time collect various types of data from production equipment, providing a rich data foundation for intelligent energy-saving optimization; The combination of a spatiotemporal feature extraction engine and a distributed reinforcement learning controller can fully mine the spatiotemporal features in the data, generate efficient optimization control strategies, and achieve optimal global energy efficiency. The three-level verification mechanism of the digital twin verification platform and the fuzzy PID control switching under abnormal conditions ensure the safety and effectiveness of the control strategy and improve the reliability of the system. The establishment of an energy consumption knowledge graph database helps to achieve dynamic load forecasting and abnormal state identification, further improving the system's intelligent energy-saving level.

[0015] Of course, any product implementing this invention does not necessarily need to achieve all of the advantages described above at the same time. Attached Figure Description

[0016] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0017] Figure 1 This is a schematic diagram of the system framework in this invention; Figure 2 This is a schematic diagram of the multimodal data acquisition and fusion process in this invention; Figure 3 This is a diagram of the spatiotemporal feature extraction engine network structure in this invention; Figure 4 This is a flowchart of the digital twin verification platform in this invention; Figure 5 The graph shows a comparison of energy-saving effects when applied in specific embodiments of the present invention (traditional PID vs. 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] like Figures 1-4 As shown: One embodiment of the present invention is as follows: The intelligent energy-saving optimization control system for production and processing based on data analysis of this invention adopts a three-level computing architecture of edge-fog-cloud. The edge layer performs data preprocessing, the fog computing layer performs feature extraction, and the cloud computing layer completes strategy optimization. This architecture can fully utilize the advantages of computing resources at different levels, improving the system's operating efficiency and response speed. The edge layer, being close to the data source, can quickly perform preliminary screening and cleaning of raw data, reducing data transmission volume; the fog computing layer has certain computing capabilities and can extract features from the data, preparing for subsequent in-depth analysis; the cloud computing layer possesses powerful computing resources for performing complex strategy optimization calculations.

[0020] The specific architecture of this system is as follows: Multimodal data acquisition and fusion module This module integrates temperature sensors, pressure sensors, flow sensors, vibration sensors, and power quality sensors to collect physical parameters, environmental variables, and energy flow data of production equipment in real time.

[0021] Dynamic sampling frequency adjustment unit: Adaptively adjusts the sampling period within the range of 10ms-5s based on the equipment load rate. Specifically, this unit uses a fuzzy PID algorithm to control the sampling frequency. The steps of the fuzzy PID algorithm are as follows: Define the input and output variables: the inputs are the equipment load rate deviation e (the difference between the actual load rate and the set load rate) and the deviation change rate ec (the rate of change of the load rate deviation), and the output is the sampling frequency adjustment amount Δf; Fuzzification: Mapping e and ec to fuzzy linguistic variables through membership functions, such as "negative large", "negative small", "zero", "positive small", "positive large", etc. For example, if the load rate is set to 80%, and the actual load rate is 50%, then e = 50% - 80% = -30%. Determine its affiliation in the fuzzy linguistic variables based on the membership function. Fuzzy rule formulation: Fuzzy control rules are formulated based on experience, such as when e is "negatively large" and ec is "negatively large", Δf is "significantly reduced". These rules form a fuzzy control rule table; Fuzzy inference: Based on the fuzzy control rule table, the fuzzy quantity of the output is calculated by the fuzzy inference synthesis algorithm; Defuzzification: The fuzzy output is converted into a precise sampling frequency adjustment Δf using methods such as the centroid method.

[0022] Functionality implementation: When the device load rate is less than 60%, switch to 1Hz low-frequency sampling mode to reduce data transmission volume and processing burden; When abnormal vibration is detected, a 50kHz high-frequency sampling mode is triggered to more accurately capture abnormal equipment conditions. The sampled data is transmitted via the LoRaWAN protocol with a latency of less than 20ms, ensuring timely data transmission.

[0023] Data cleaning pipeline: An improved median filtering algorithm is used to eliminate impulse noise. Traditional median filtering sorts the data within a window by size and takes the median value as the filtered output. The improved median filtering incorporates the time-series characteristics of the data; the algorithm steps are as follows: Let the data sequence be {x1,x2,...,xn}, and the window size be m (usually an odd number). For each data point xi, take a window centered on xi [x(i-(m-1) / 2),...,xi,...,x(i+(m-1) / 2)]; Calculate the mean μ and standard deviation σ of the data within the window; If a data point xi satisfies |xi-μ|>kσ (k is an empirical constant, typically 2-3), then xi is determined to be an impulse noise point, and xi is replaced by the median value within the window; otherwise, xi remains unchanged. This method can more effectively remove noise interference and improve data quality.

[0024] Feature extraction unit: A 128-dimensional feature vector is generated using wavelet packet decomposition technology, fusing frequency domain features of vibration signals, time domain features of electrical parameters, and spatial distribution features of environmental parameters. The wavelet packet decomposition algorithm steps are as follows: Perform wavelet transform on the signal x(t) to obtain the approximation coefficients A and detail coefficients D; The approximation coefficients A and detail coefficients D are subjected to wavelet transforms again to further subdivide the frequency bands and form a wavelet packet tree. By selecting appropriate wavelet packet nodes, their energy, variance, and other characteristic values ​​can be calculated. For example, for vibration signals, the energy of wavelet packet coefficients in a specific frequency band can be selected as a frequency domain feature; for electrical parameters, the mean and variance of current and voltage can be calculated as time domain features; for environmental parameters, such as the spatial distribution of temperature, temperature values ​​at different locations can be used as spatial distribution features. By fusing features from different types of data, more comprehensive information can be provided for subsequent analysis and processing.

[0025] Spatiotemporal feature extraction engine This engine connects to a multimodal data acquisition and fusion module to extract spatiotemporal joint features of device status.

[0026] Spatial Feature Extraction Layer: This layer employs a 3D convolutional neural network to extract the spatial distribution features of the equipment's status through convolution operations on the equipment's status data in the spatial dimension. Examples include spatial relationships such as temperature and pressure distributions among the equipment's components. The convolution operation formula for the 3D convolutional neural network is as follows: in, Is the output feature map in The pixel value of the location, Is it the convolution kernel in Position weight, Is the input data in The pixel values ​​at the location, a, b, and c are the sizes of the convolution kernel, and b is the bias. By continuously adjusting the weights of the convolution kernel, the spatial distribution pattern of the device state data is learned.

[0027] Temporal Feature Extraction Layer: Utilizing a bidirectional long short-term memory (LSTM) network layer, this layer captures the evolution of feature parameters over time, such as temperature trends and energy consumption fluctuations during equipment operation. The bidirectional LSTM network consists of forward LSTM and backward LSTM layers. The internal structure of each LSTM unit includes input gates. Forgotten Gate Output gate and memory unit The forward LSTM processes time series from the past to the future, while the backward LSTM processes time series from the future to the past. The outputs of the two are concatenated as the final output, thus comprehensively capturing the evolutionary patterns of time.

[0028] Feature selection layer: The weight coefficients of each feature channel are dynamically calculated using the Pearson correlation coefficient. The formula for calculating the Pearson correlation coefficient is:

[0029] in, and It is a data sequence with two feature channels. and These are their means, and n is the number of data points. The Pearson correlation coefficient between each feature channel and the equipment status or energy-saving optimization target is calculated. Weighting coefficients are determined based on the correlation coefficient values ​​to screen out features that have a significant impact on equipment status and energy-saving optimization, thereby improving the efficiency and accuracy of subsequent analysis and decision-making.

[0030] Distributed reinforcement learning controller The controller employs a dual-delay deep deterministic policy gradient algorithm (TD3) framework, which includes a local decision maker and a central coordinator.

[0031] Local Decision Maker: Deployed in each processing unit, it executes real-time control decisions based on the current equipment status. The local decision maker generates control commands suitable for its specific processing unit based on equipment status features provided by the spatiotemporal feature extraction engine, combined with local production needs and constraints. Its decision-making process is based on the TD3 algorithm, the steps of which are as follows: Initialize policy network Target-Policy Network Q network , and its target network , Set the learning rate and discount factor. Parameters; For each training cycle: The local decision-maker determines the status based on the current device state. Through policy network Generate Actions ; Execute action Receive a reward and the next state ; Will( , , , Stored in the experience replay pool; A batch of data was randomly sampled from the experience replay pool. , , , ); Calculate the target Q value; Update the Q-network: Update it separately by minimizing the mean square error. and ; Regularly update the policy network: by maximizing renew At the same time, update the target network parameters.

[0032] Central Coordinator: This coordinator uses the Nash equilibrium algorithm to coordinate the decision-makers of each unit to achieve optimal global energy efficiency. The application process of the Nash equilibrium algorithm is as follows: Let the decision of each processing unit be... The energy efficiency function of the entire production system is ; The central coordinator initializes the decisions of each unit; For each iteration step: Fix the decisions of other units, and let each unit i find the one that makes The largest ; Repeat the above process until the decisions of each unit no longer change, at which point Nash equilibrium is reached, meaning that the decisions of each unit achieve optimal global energy efficiency. The central coordinator comprehensively considers the decisions of each processing unit and the energy consumption, production efficiency, and other indicators of the entire production system. Through iterative calculations, it finds the optimal balance point between the decisions of each unit, so that the energy efficiency of the entire production system is optimized.

[0033] Digital twin verification platform The platform integrates computational fluid dynamics simulation module and discrete event simulation module, and outputs optimized control commands to the actuator after safety verification.

[0034] The computational fluid dynamics simulation module simulates the thermodynamic processes of equipment, such as fluid flow and heat transfer within the equipment. By accurately simulating the physical phenomena during equipment operation, it evaluates the impact of control commands on equipment performance and energy consumption. Its simulation process is based on fundamental fluid dynamics equations such as the Navier-Stokes equations. These equations are solved using numerical methods (such as the finite element method and the finite volume method) to simulate the fluid flow and heat transfer inside the equipment.

[0035] Discrete Event Simulation Module: This module establishes a mapping relationship between production cycle time and energy consumption. By modeling and simulating discrete events in the production process (such as equipment start-up, shutdown, and task switching), it analyzes energy consumption under different production cycle times, providing a reference for optimizing control commands. The modeling process is as follows: Define discrete events, such as device startup events. Equipment stop event Processing task switching event wait.

[0036] Define the triggering conditions and occurrence time for each event. For example, the triggering condition for a device start event might be receiving a production task instruction, and the occurrence time might be the moment the instruction is issued.

[0037] Establish an energy consumption model to calculate the energy consumption at the time of each event and the time interval between events, based on the energy consumption characteristics of the equipment in different states (start-up, operation, shutdown, etc.). For example, the energy consumption of the equipment during operation is related to the operating time and power. ,in It refers to the power of the equipment. This refers to runtime. This method establishes a mapping between production cycle time and energy consumption.

[0038] Security Validation Sandbox: Pre-tests the effects of control strategies in a virtual environment, verifying the security and effectiveness of control commands by simulating various possible operating conditions and abnormal situations in a virtual environment.

[0039] Three-level verification mechanism: Sensor-level verification: Inject ±5% noise data into the digital twin to verify the robustness of the control strategy and ensure that the control strategy can still work normally when there is some noise interference in the sensor data.

[0040] Subsystem-level verification: Simulate equipment failure, test the safety interlock response time, and verify the system's safety protection capabilities when equipment failure occurs. For example, simulate a temperature sensor failure in a heating furnace, observe whether the system can promptly trigger the safety interlock mechanism to shut down the heating furnace or take other protective measures, and record the response time.

[0041] Full-line verification: Run 1000 hours of simulation on a virtual production line to verify the energy consumption target compliance rate and evaluate the overall optimization effect of the control strategy on the energy consumption of the production system. The energy consumption target compliance rate is calculated by comparing the actual energy consumption with the target energy consumption during the virtual production line operation.

[0042] Switching to Fuzzy PID Control Mode in Abnormal Situations: When the system detects an abnormal situation, such as equipment failure or verification results not meeting safety requirements, it automatically switches to fuzzy PID control mode to ensure safe equipment operation and production continuity. The working principle of fuzzy PID control mode is similar to the fuzzy PID algorithm in the dynamic sampling frequency adjustment unit, except that the controlled object becomes the equipment's operating parameters (such as temperature, pressure, etc.). By adjusting the control parameters, the equipment can still maintain relatively stable operation under abnormal conditions.

[0043] Energy Consumption Knowledge Graph A spectral feature library for 23 typical process modes was constructed to collect and organize the energy consumption spectral characteristics under different process modes, providing a reference for subsequent energy consumption analysis and prediction. A cosine similarity matching matrix for 78 types of abnormal states was established. By calculating the cosine similarity between the current state of the equipment and known abnormal states, it is possible to quickly identify whether the equipment is in an abnormal state and the type of abnormality. Dynamic load prediction was implemented based on the historical operating data of the equipment.

[0044] A specific application of this embodiment is as follows: Control scenario for carbon material calcining furnace (for effect comparison) Figure 5 ) Data Acquisition and Preprocessing: In the carbon material calcining furnace control scenario, temperature sensors are installed on the furnace body to monitor the temperature distribution in different areas of the furnace; pressure sensors are installed to detect the gas pressure inside the furnace; flow sensors are installed to measure the fuel gas flow rate; vibration sensors are also equipped to monitor the vibration of the furnace body; and power quality sensors are used to monitor power supply parameters in real time. The dynamic sampling frequency adjustment unit adjusts the sampling frequency in real time within the range of 10ms-5s according to the calcining furnace load rate. For example, when the calcining furnace is operating stably and the load rate is 75%, a moderate sampling frequency of once every 150ms is maintained using a fuzzy PID algorithm. Once abnormal vibration is detected in a certain part of the furnace body, a high-frequency sampling mode of 50kHz is immediately triggered. The acquired data is processed through a data cleaning pipeline, using an improved median filtering algorithm to remove impulse noise. Then, the feature extraction unit uses wavelet packet decomposition technology to generate a 128-dimensional feature vector, fusing the frequency domain features of the vibration signal, the time domain features of the power parameters, and the spatial distribution features of the furnace environmental parameters to complete the data preprocessing.

[0045] Feature Extraction and Strategy Generation: The spatiotemporal feature extraction engine begins operation, with the 3D convolutional neural network of the spatial feature extraction layer performing convolution operations on the multidimensional data within the furnace. The input data includes information such as temperature, pressure, and fuel gas flow rate at different locations within the furnace, presented as a 3D matrix. The convolutional kernel slides in 3D space, extracting the distribution patterns of temperature at different heights and radial positions within the furnace, the spatial trends of pressure variation, and the correlation features between fuel gas flow rate and various regions within the furnace through weighted summation of data from different locations. For example, it identifies the distribution range of high-temperature regions within the furnace and their spatial correspondence with regions of significant pressure variation. These spatial features are crucial for understanding the reaction state inside the calcining furnace and optimizing the calcination process. The bidirectional long short-term memory network of the temporal feature extraction layer processes the data in a time-series manner. It not only captures parameter changes over time in the forward direction but also analyzes them in the backward direction, thus comprehensively grasping the temporal evolution of the data. For the operation of the calcining furnace, the time-series data includes the values ​​of temperature, pressure, and flow rate at each sampling moment.

[0046] Bidirectional Long Short-Term Memory (LSTM) networks, through a gating mechanism, can remember key information over long periods, such as how temperature gradually increases over time during a complete calcination cycle, pressure fluctuations at different stages, and the dynamic adjustment of fuel gas flow according to production process requirements. By learning these temporal characteristics, the network can predict parameter trends in the near future, providing a temporal basis for control strategy formulation. The feature selection layer dynamically calculates the weight coefficients of each feature channel using the Pearson correlation coefficient. During the calcination of carbon materials, different features have varying degrees of impact on the furnace's operating status and the final energy-saving optimization goal. For example, regarding the calcination efficiency and energy consumption of the furnace, the correlation between temperature and fuel gas flow in key areas of the furnace may be high, while vibration characteristics at certain locations may have a weaker correlation with current production efficiency. By calculating the Pearson correlation coefficient, the correlation between each feature channel and production goals (such as improving carbon material quality, reducing energy consumption, and increasing production efficiency) is determined, thereby assigning weights to different features. Features with higher weights are retained and used primarily for subsequent analysis and decision-making, while features with lower weights are weakened or discarded, thereby improving the efficiency and accuracy of analysis and decision-making.

[0047] The local decision-maker in the distributed reinforcement learning controller is deployed in the calcining furnace control unit. Based on the feature information extracted by the spatiotemporal feature extraction engine, and combined with the current production task of the calcining furnace, it generates preliminary control decisions. For example, when excessively high local temperatures are detected in the furnace, and feature analysis indicates that this is due to uneven fuel gas flow distribution, the local decision-maker will calculate specific values ​​for adjusting the opening of different nozzles in the fuel gas supply equipment according to the pre-trained strategy model. This optimizes the fuel gas flow distribution, improves combustion conditions in the furnace, and increases calcination efficiency.

[0048] The local decision-maker's strategy model is trained based on extensive historical production data and practical operational experience, enabling it to respond quickly and accurately based on the current equipment status. The central coordinator employs a Nash equilibrium algorithm to comprehensively consider the decisions of various production units within the carbon material production plant (calcining furnaces, molding equipment, roasting furnaces, etc.). Taking calcining furnaces and molding equipment as examples, the quality and quantity of carbon materials produced by the calcining furnace affect the production rhythm and energy consumption of the molding equipment, while the production demands of the molding equipment, in turn, require adjustments to the production parameters of the calcining furnace.

[0049] The central coordinator iteratively calculates to find the optimal balance between the decisions of each unit. In each iteration, the decisions of other units are fixed, allowing each unit to find the decision that maximizes the overall benefits of the entire carbon material production system (including energy consumption, output, quality, and other indicators). For example, while meeting market demand for carbon material output, the energy consumption of the entire production system is minimized by coordinating the production rhythm of the calcining furnace and the operating parameters of the molding equipment, while ensuring that the quality of the carbon materials meets standards. After multiple iterations, the decisions of each unit reach a Nash equilibrium, achieving optimal energy efficiency for the entire carbon material production process.

[0050] Verification and Execution: The digital twin verification platform plays a crucial role. The computational fluid dynamics simulation module simulates gas flow and heat transfer within the calcining furnace, evaluating the impact of control commands on the furnace reaction. The discrete event simulation module constructs a mapping relationship between production cycle time (e.g., discharge cycle) and energy consumption. A safety verification sandbox pre-simulates the effectiveness of the control strategy. After passing verification through a three-level verification mechanism, encrypted control commands are issued to the actuators, such as adjusting the power of fuel gas supply equipment and controlling the furnace cooling water flow rate, achieving intelligent energy-saving optimization of the carbon material calcination process. The energy consumption knowledge graph is continuously updated, providing a basis for subsequent optimization, such as optimizing the production process based on the energy consumption spectrum characteristics under different carbon material yields. The energy consumption knowledge graph collects spectrum characteristics under different calcination processes, establishing an abnormal state matching matrix. When furnace parameters deviate from the normal range, the system quickly locates the abnormal type through cosine similarity matching, such as incomplete fuel gas combustion or localized furnace blockage, and adjusts the control strategy accordingly.

[0051] In the description of this specification, references to terms such as "an embodiment," "example," "specific example," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of the invention. In this specification, illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples.

[0052] The preferred embodiments of the present invention disclosed above are merely illustrative of the invention. These preferred embodiments do not exhaustively describe all details, nor do they limit the invention to the specific implementations described. Clearly, many modifications and variations can be made based on the content of this specification. This specification selects and specifically describes these embodiments to better explain the principles and practical applications of the invention, thereby enabling those skilled in the art to better understand and utilize the invention. The invention is limited only by the claims and their full scope and equivalents.

Claims

1. A data-based intelligent energy-saving optimization control system for production and processing, characterized in that: include: The multimodal data acquisition and fusion module integrates temperature sensors, pressure sensors, flow sensors, vibration sensors, and power quality sensors to collect and integrate physical parameters, environmental variables, and energy flow data of production equipment in real time. A spatiotemporal feature extraction engine, connected to the multimodal data acquisition and fusion module, is used to extract spatiotemporal joint features of the device status; The distributed reinforcement learning controller adopts a dual-delay deep deterministic policy gradient algorithm framework, which includes a local decision-maker and a central coordinator, and generates optimized control commands based on the spatiotemporal joint features. The digital twin verification platform integrates a computational fluid dynamics simulation module and a discrete event simulation module, and outputs the optimized control commands to the actuator after performing safety verification. The system adopts a three-level computing architecture of edge-fog-cloud, in which the edge layer performs data preprocessing, the fog computing layer performs feature extraction, and the cloud computing layer completes strategy optimization.

2. The intelligent energy-saving optimization control system for production and processing based on data analysis according to claim 1, characterized in that, The multimodal data acquisition and fusion module includes: The dynamic sampling frequency adjustment unit adaptively adjusts the sampling period within the range of 10ms-5s according to the equipment load rate. The data cleaning pipeline uses an improved median filtering algorithm to eliminate impulse noise. The feature extraction unit uses wavelet packet decomposition technology to generate a 128-dimensional feature vector, which integrates the frequency domain features of vibration signals, the time domain features of power parameters, and the spatial distribution features of environmental parameters.

3. The intelligent energy-saving optimization control system for production and processing based on data analysis according to claim 2, characterized in that, The dynamic sampling adjustment unit uses a fuzzy PID algorithm to control the sampling frequency and achieves the following functions: When the device load rate is less than 60%, switch to 1Hz low-frequency sampling mode; When abnormal vibration is detected, the 50kHz high-frequency sampling mode is triggered; The sampled data is transmitted via the LoRaWAN protocol with a latency of <20ms.

4. The intelligent energy-saving optimization control system for production and processing based on data analysis according to claim 1, characterized in that, The spatiotemporal feature extraction engine includes: Spatial feature extraction layer: a 3D convolutional neural network layer that extracts the spatial distribution features of the device status; Temporal feature extraction layer: a bidirectional long short-term memory network layer that captures the temporal evolution of feature parameters; Feature selection layer: The weight coefficients of each feature channel are dynamically calculated using the Pearson correlation coefficient.

5. The intelligent energy-saving optimization control system for production and processing based on data analysis according to claim 1, characterized in that, The distributed reinforcement learning controller employs an improved dual-delay deep deterministic policy gradient algorithm framework, including: Local decision-makers, deployed in each processing unit, execute real-time control decisions based on the current equipment status; The central coordinator coordinates the decision-makers of each unit through the Nash equilibrium algorithm to achieve optimal global energy efficiency.

6. The intelligent energy-saving optimization control system for production and processing based on data analysis according to claim 1, characterized in that, The digital twin verification platform includes: The computational fluid dynamics simulation module simulates the thermodynamic processes of equipment; The discrete event simulation module constructs a mapping relationship between production cycle time and energy consumption; The security verification sandbox simulates the effects of control strategies in a virtual environment. The platform is equipped with a three-level verification mechanism and switches to a fuzzy PID control mode in case of abnormalities.

7. The intelligent energy-saving optimization control system for production and processing based on data analysis according to claim 6, characterized in that, The three-level verification mechanism of the digital twin verification platform includes: Sensor-level verification: Inject ±5% noise data into the digital twin to verify the robustness of the control strategy; Subsystem-level verification: Simulate equipment failure and test the response time of safety interlocks; Full-line verification: Run 1000 hours of simulation on a virtual production line to verify the compliance rate of energy consumption indicators.

8. The intelligent energy-saving optimization control system for production and processing based on data analysis according to claim 1, characterized in that, It also includes an energy consumption knowledge graph database: Construct a spectral feature library for 23 typical process modes; Establish a cosine similarity matching matrix for 78 types of abnormal states; To achieve dynamic load forecasting.

9. The intelligent energy-saving optimization control system for production and processing based on data analysis according to claim 1, characterized in that, The system can be used in the following ways: S1. Dynamic Sampling Control: Dynamically adjusts the sampling frequency to acquire multi-source data, and performs wavelet packet decomposition and outlier detection; S2. Data preprocessing: Median filtering is performed at the edge nodes to generate 128-dimensional standardized feature vectors; S3. Feature Engineering: Spatiotemporal joint features are extracted using a hybrid neural network, and the correlation coefficient matrix between features is calculated. S4. Strategy Generation: The TD3 algorithm is used to generate the control strategy, with an initial exploration rate ε=0.3 that linearly decays to 0.05; S5. Verification and Execution: Verify the security of the policy on the digital twin platform, and issue encrypted control commands after passing the verification.

Images