Industrial park elevator system optimization control system and method thereof

Abstract

The invention relates to the field of air conditioning unit control, in particular to an industrial park elevator system optimization control system and a method thereof.The hardware part comprises a sensor unit, an executing mechanism, a field control unit, communication equipment and a server unit; the software part comprises a real-time operating system, a sensor data acquisition and fusion preprocessing module, a digital twinborn simulation platform, a model prediction control module and a cloud deep reinforcement learning module; the method comprises the steps of S1, data acquisition and primary processing, S2, data transmission and caching, S3, digital twin modeling and system simulation, S4, optimization control strategy generation and issuing, S5, online scheduling and deep reinforcement learning optimization, S6, safety monitoring and exception handling, and S7, closed-loop feedback and continuous optimization. Efficient integration of elevator modules is achieved through digital twinning and model prediction control cooperation; intelligent data processing is realized through deep reinforcement learning and adaptive data fusion; and safe and energy-saving operation is realized through fault tolerance and safety monitoring.

Description

Technical Field

[0001] The present invention relates to the field of air conditioner unit control, and particularly to an optimized control system and method for an elevator system in an industrial park. Background Art

[0002] Currently, as a region where modern manufacturing and high-tech are concentrated, the buildings and equipment in industrial parks are dense, posing high requirements for the safety, efficiency, and energy consumption management of elevator systems. The elevator systems in industrial parks are mainly used for the vertical transportation of personnel and materials, and usually adopt standardized equipment and basic automatic control technologies to meet the daily operation requirements.

[0003] However, in practical applications, there are obvious deficiencies in the existing optimized control schemes for elevator systems in industrial parks. Traditional control methods and preliminary artificial intelligence scheduling algorithms have great limitations in system integration, data acquisition and real-time processing, dynamic scheduling, and energy consumption management, and are difficult to cope with the operation challenges in high-traffic, high-frequency, and complex environments in industrial parks, resulting in response delays, insufficient safety guarantees, and low energy utilization efficiency.

[0004] Therefore, we propose an optimized control system and method for an elevator system in an industrial park. Summary of the Invention

[0005] The purpose of the present invention is to solve the deficiencies in the integration, real-time response, energy consumption management, and safety guarantee of existing elevator systems in industrial parks. By constructing an optimization platform based on digital twin simulation and model predictive control, and combining cloud deep reinforcement learning to achieve dynamic scheduling and online optimization, the operation efficiency of the elevator system is improved, energy consumption is reduced, and operation safety is ensured, meeting the requirements of modern industrial parks for efficient, intelligent, safe, and energy-saving operation.

[0006] On the one hand, an embodiment of the present invention provides an optimized control system for an elevator system in an industrial park. The hardware part includes a sensor unit, an actuator, a field control unit, communication equipment, and a server unit. The software part includes a real-time operating system, a sensor data acquisition and fusion preprocessing module, a digital twin simulation platform, a model predictive control module, and a cloud deep reinforcement learning module; The sensor unit includes traditional sensors and advanced environmental sensors. The traditional sensors include position sensors, speed sensors, load sensors, door status sensors, and temperature and humidity sensors, which are used to collect basic data on elevator operation. The advanced environmental sensors include an energy consumption monitoring module and vibration temperature sensors, which are used to collect auxiliary data on the environment and equipment operation. The data collected by the sensor unit is transmitted to the field control unit through the communication equipment for preliminary processing, and further transmitted to the server unit to support the calculation, analysis, and optimized scheduling of software modules; The actuator includes a driving device, a door machine, and an environmental control device. The driving device includes a driving motor and a speed regulation module for driving the elevator car. The door machine is used to open and close the elevator door. The environmental control device is used to regulate environmental parameters such as temperature and humidity inside the elevator and related areas. The operating parameters of the actuator are determined by the control strategy calculated by the model predictive control module. This strategy is calculated by the cloud server module and sent to the on-site control unit via the communication device, and then forwarded by the on-site control unit to the actuator for final execution; The on-site control unit includes an industrial controller installed on each elevator control terminal. The industrial controller integrates an edge computing module for real-time data acquisition, preprocessing, local control, and safety monitoring. The on-site control unit collects and preliminarily processes the data of the sensor unit and transmits it to the server group via the communication device for analysis and optimization calculation of advanced algorithms. At the same time, it is responsible for receiving the control instructions sent by the server group and performing corresponding local adjustments; The communication device includes an industrial Ethernet switch and a wireless data terminal for data transmission and control instruction issuance between the on-site control unit, the server group, and the digital twin simulation platform. The communication device ensures the efficient transmission of data between the sensor unit, the on-site control unit, and the server group, ensuring the reliability and real-time nature of data interaction and control signals between the system modules; The server group includes a local server module and a cloud server module. The local server module is responsible for short-cycle data caching and local optimization calculation for on-site data storage and fast processing. The cloud server module undertakes complex computing tasks, including running deep reinforcement learning and optimization scheduling models, and generating control strategies to be sent to the on-site control unit for big data processing, global optimization scheduling, and running advanced intelligent algorithms; The real-time operating system is deployed on the industrial controller in the on-site control unit. The real-time operating system manages and schedules various computing tasks of the on-site control unit, ensuring real-time processing of sensor data and efficient response of the actuator. At the same time, it conducts data interaction with the server group to ensure the stable operation of the system, ensuring timely response to local control and safety monitoring tasks; The sensor data acquisition and fusion preprocessing module is deployed on the industrial controller in the on-site control unit for filtering, verification, formatting, and data fusion processing of elevator operation data and environmental data from the sensor unit to ensure data quality. Special processing algorithms can be used according to different data characteristics. The output data of the sensor data acquisition and fusion preprocessing module is not only used for local control but also transmitted to the cloud server module via the communication device for digital twin simulation and optimization calculation; The digital twin simulation platform is deployed on the cloud server module, receives data from the sensor data acquisition and fusion preprocessing module, and constructs a virtual model of elevator operation and environmental control by combining historical operation data and real-time data, which is used to predict the system state and provide a basis for optimized scheduling, realize real-time simulation of elevator operation status, energy flow and environmental control equipment status, and conduct data interaction with the field control unit through communication equipment; The model predictive control module is deployed on the cloud server module and works in coordination with the digital twin simulation platform. It performs optimized calculations based on the data provided by the digital twin simulation platform, is used to predict the elevator operation status within a certain period of time in the future, online solves the optimal control strategy, and sends the optimized control strategy to the field control unit through communication equipment. The field control unit forwards it to the actuator, thereby dynamically adjusting the working status of the driving device, door machine and environmental control equipment to achieve precise control; The cloud deep reinforcement learning module is deployed on the cloud server module, which is used to train elevator scheduling, energy consumption management and safety decision-making models using big data and historical operation records, and periodically send the optimized models to the industrial controller in the field control unit through communication equipment to achieve global optimal scheduling and local real-time control. At the same time, the cloud deep reinforcement learning module continuously learns historical data and real-time data to improve the accuracy of the scheduling algorithm, and updates the optimized model parameters to the cloud server module to support the iterative upgrade of optimized calculations and scheduling decisions; The real-time operating system works in coordination with the sensor data acquisition and fusion preprocessing module to achieve real-time acquisition, preprocessing and safety monitoring of on-site data; the digital twin simulation platform and the model predictive control module work together to achieve real-time simulation and prediction of elevator operation status, energy flow and environmental control equipment status, providing a basis for dynamic control strategies; the cloud deep reinforcement learning module coordinates with the industrial controller in the field control unit through the cloud server module to achieve effective integration of global optimized scheduling and local real-time control, thereby ensuring that the system achieves comprehensive optimization in terms of energy conservation, intelligent scheduling, safety protection and personalized services.

[0007] Preferably, the digital twin simulation platform adopts a hybrid modeling method combining deep neural network and physical modeling to achieve high-precision prediction of elevator operation status, energy flow and environmental control equipment status; This hybrid modeling method includes an adaptive parameter correction unit, which is used to dynamically adjust model parameters according to real-time data, thereby further improving prediction accuracy and system robustness. Its specific algorithms include: Using the convolutional neural network CNN to extract the spatial features in the elevator operation time series data; Combining physical modeling constraints to construct a hybrid model, and its mathematical expression is: L = α·MSE(Y_pred, Y_actual)+β·‖F_phy(Y_pred)-C_phy‖² Where: α: Positive weight coefficient, used to adjust the proportion of the mean square error term in the overall loss function; MSE(Y_pred, Y_actual): Mean square error, measuring the average error between the predicted output Y_pred and the actual output Y_actual; Y_pred: The output value obtained by neural network prediction; Y_actual: The target data actually collected; β: Positive weight coefficient, used to adjust the proportion of the physical model error term in the overall loss function; F_phy(Y_pred): The calculation result of the predicted output Y_pred based on the physical modeling prediction function; C_phy: Preset physical constraint constant, used to define the ideal state of the physical model; ‖·‖²: Square Euclidean norm, used to quantify the magnitude of the physical model prediction error; The weights of the neural network are updated online through the adaptive parameter correction unit to ensure that the model can dynamically adapt to real-time data changes, thereby improving the prediction accuracy and system robustness.

[0008] Preferably, the model predictive control module further includes a control strategy optimization unit based on adaptive weight allocation. This unit dynamically adjusts the weights of each control parameter using the real-time data and historical operation data provided by the industrial controller in the field control unit to achieve precise regulation of the drive device, door machine, and environmental control equipment. Its core uses the following formula for parameter optimization: u* = arg min_u{∑_(i=1)^Nw_i·J_i(u)} Where: u*: Optimal control input, that is, the control variable that makes the total cost function reach the minimum; u: Control input variable to be optimized; N: Number of control objectives, that is, the number of control objectives for the drive device, door machine, and environmental control equipment respectively; w_i: Adaptive weight of the i-th control objective, dynamically adjusted by the online algorithm according to real-time data; J_i(u): Cost (cost) function for the i-th control objective, reflecting the impact of the control input u on this objective; ∑_(i=1)^N: Sum of the cost functions for all control objectives; This algorithm realizes precise regulation of different controlled objects by optimizing the weight distribution of each sub-goal in real time, improving the comprehensive optimization effect of the control strategy.

[0009] Preferably, the cloud deep reinforcement learning module adopts a deep Q-network algorithm based on a multi-objective reward function. While comprehensively considering energy consumption optimization, scheduling efficiency, and user comfort, it continuously optimizes the elevator scheduling, energy consumption management, and safety decision-making strategies through online training, and updates the control strategy in real time according to the feedback data. The reward function R of the deep Q-network DQN algorithm based on the multi-objective reward function is defined as follows: R = γ1·(ΔE) + γ2·(ΔT) + γ3·(ΔS) Where: R: Reward value, used to evaluate the comprehensive effect of the current scheduling decision and reflect the achievement of the optimization goal; γ1, γ2, γ3: Positive weight parameters, respectively adjusting the proportions of energy consumption optimization, waiting time reduction, and safety index improvement in the overall reward; ΔE: Energy consumption reduction amount, indicating the degree of improvement in energy consumption achieved through optimized control; ΔT: Reduction amount of passenger waiting time, indicating the reduction range of passenger waiting time and reflecting the improvement of service efficiency; ΔS: Improvement value of the safety index, indicating the degree of improvement in safety performance, and quantifying safety through monitoring indicators.

[0010] Preferably, the communication device further includes a fault-tolerant network protocol that supports redundant backup, ensuring higher reliability and real-time performance of data transmission between system modules in a harsh environment, thereby guaranteeing the stable operation of the entire system. The core algorithms of the fault-tolerant network protocol based on the redundant backup mechanism include: ① Implementing redundant storage and backup of data packets based on the distributed hash table DHT; ② Adopting fast retransmission and adaptive rate adjustment algorithms to ensure the real-time performance and integrity of data transmission even in case of communication anomalies; This solution effectively improves the reliability and data transmission efficiency of the system in a complex environment.

[0011] The industrial controller in the field control unit integrates a safety monitoring sub-module based on machine learning. The safety monitoring sub-module adopts a safety monitoring algorithm that combines support vector machine (SVM) and random forest (RF) to analyze elevator operation data in real time, automatically identify potential anomalies, and send early warning signals in advance, thereby realizing active prediction and prevention of system failures, specifically as follows: ① The safety monitoring module first conducts preliminary anomaly detection through SVM, using the formula: f(x) = sign(∑_(i=1)^l α_i y_i K(x_i, x) + b) Wherein: sign(·): Sign function, with the output of a positive value corresponding to +1 and a negative value corresponding to -1; l: The number of support vectors, i.e., the number of key samples selected during the training process; α_i: The weight coefficient corresponding to the i-th support vector, determined by the SVM training process; y_i: The class label of the i-th support vector, usually taking values of +1 or -1; K(x_i, x): Kernel function, used to calculate the similarity between the support vector x_i and the input sample x; b: Bias term, used to adjust the classification boundary; ② For the detected abnormal data, use RF for multi-level classification to ensure the accuracy and timeliness of fault warning; This combined algorithm can significantly improve the prediction and prevention capabilities of system faults.

[0012] On the other hand, another embodiment of the present invention provides an optimization control method for an elevator system in an industrial park, comprising the following steps: S1. Data acquisition and preliminary processing: Collect elevator operation and environmental data through traditional sensors and advanced environmental sensors in the sensor unit, including position, speed, load, door status, temperature and humidity, energy consumption, and vibration temperature information, and use the sensor data acquisition and fusion preprocessing module on the industrial controller in the on-site control unit to filter, verify, format, and fuse the collected multi-source data, and output high-quality fused data; S2. Data transmission and caching: Transmit the preprocessed data to the server unit through communication devices, where the local server module realizes short-cycle data caching and fast response, and at the same time uses a redundant communication mechanism to transmit the data to the cloud server module to provide a basis for subsequent big data analysis; S3. Digital twin modeling and system simulation: Use the digital twin simulation platform on the cloud server module to construct a virtual model of elevator operation and environmental control by combining historical data and real-time data, and use a hybrid modeling algorithm to perform real-time simulation and prediction on elevator status, energy flow, and environmental control equipment status; S4. Generation and distribution of optimization control strategies: Based on the data output by the digital twin simulation platform, use the model predictive control module on the cloud server module to online solve the optimal control strategy, and transmit this control strategy to the on-site control unit through communication devices, and then forward it to the actuator by the on-site control unit to realize the dynamic adjustment of the drive device, door machine, and environmental control equipment; S5. Online Scheduling and Optimization with Deep Reinforcement Learning: On the cloud server module, use the cloud deep reinforcement learning module to conduct online training on the elevator scheduling, energy consumption management, and safety decision-making models. Adopt a multi-objective reward function to continuously optimize the control strategy, and periodically update the model parameters and control strategy. Transmit the optimization results to the on-site control unit; S6. Safety Monitoring and Exception Handling: The industrial controller in the on-site control unit integrates a safety monitoring module to continuously monitor the elevator operation status and automatically detect abnormalities. Once an abnormality is detected, the system immediately triggers an early warning mechanism, transmits the abnormality information to the server unit through communication devices, and at the same time, the on-site control unit executes preset safety emergency control measures to ensure the safe operation of the system; S7. Closed-loop Feedback and Continuous Optimization: The system feeds back the execution results and operation data of the on-site control unit to the cloud server module to form a closed-loop feedback mechanism, supporting the continuous iterative optimization of digital twin simulation, model predictive control, and deep reinforcement learning modules, and achieving continuous improvement of the overall system performance.

[0013] Preferably, in the step S3, the digital twin modeling and system simulation adopt a hybrid modeling algorithm, where the hybrid modeling algorithm combines a deep neural network and a physical model, and dynamically adjusts the model parameters according to online feedback.

[0014] Preferably, in the step S4, the model predictive control module adopts an adaptive weight allocation algorithm, and realizes precise regulation of each sub-control target by optimizing the objective function \(J(u)=\sum_{i = 1}^{N}w_{i}\cdot J_{i}(u)\); Where: \(J(u)\): The overall optimization objective function. By weighted summation of each sub-objective, a comprehensive cost value is obtained, which takes the minimum value under the optimal control input, thereby realizing precise regulation; \(N\): Represents the number of control objectives, covering the control objectives of each subsystem such as the drive device, door machine, and environmental control equipment; \(w_{i}\): The adaptive weight of the \(i\)-th control objective, dynamically adjusted by an online algorithm according to real-time collected data, used to reflect the importance of each control objective; \(J_{i}(u)\): The cost function for the \(i\)-th control objective, quantifying the error or deviation degree generated by the control input \(u\) for this objective.

[0015] Preferably, in the step S5, the cloud deep reinforcement learning module adopts a deep Q-network (DQN) algorithm based on a multi-objective reward function, and its reward function is \(R = \gamma_{1}\cdot(\Delta E)+\gamma_{2}\cdot(\Delta T)+\gamma_{3}\cdot(\Delta S)\) to achieve energy consumption optimization, waiting time reduction, and safety index improvement; Where: R: The reward value, which is used to evaluate the comprehensive effect of the current scheduling decision and reflects the achievement of the optimization goal. γ1, γ2, γ3: Positive weight parameters that respectively adjust the proportions of energy consumption optimization, waiting time reduction, and safety index improvement in the overall reward. ΔE: The amount of energy consumption reduction, which represents the degree of improvement in energy consumption achieved through optimal control. ΔT: The amount of reduction in passengers' waiting time, which represents the reduction range of passengers' waiting time and reflects the improvement in service efficiency. ΔS: The improvement value of the safety index, which represents the degree of improvement in safety performance and quantifies the safety through monitoring indicators.

[0016] Advantages of the present invention: 1. Integrated optimization and rapid response: Through the collaborative action of the digital twin simulation platform and the model predictive control module, the present invention realizes the efficient integration and precise regulation of each sub-module of the elevator system, significantly reduces the system complexity, improves the response speed, and is convenient for deployment and maintenance. 2. Intelligent data processing and enhanced robustness: By introducing the adaptive data acquisition and fusion preprocessing algorithm, as well as the model prediction based on deep learning and the deep reinforcement learning optimization technology, the problems of strong data dependence and response delay are significantly improved, and the adaptability of the system to environmental changes is enhanced. 3. Remarkable safety and energy-saving effects: Through the safety monitoring module, the fault-tolerant communication mechanism, and the optimization of the multi-objective reward function, the comprehensive improvement of the elevator energy consumption, scheduling efficiency, and safety index is realized, which not only ensures the operation safety but also achieves energy conservation and consumption reduction, meeting the operation requirements of high efficiency, intelligence, safety, and energy conservation in industrial parks. Description of the drawings

[0017] Figure 1 It is a structural block diagram of an optimization control system for an elevator system in an industrial park provided by an embodiment of the present invention.

[0018] Figure 2 It is a flowchart of an optimization control method for an elevator system in an industrial park provided by another embodiment of the present invention. Detailed implementation manners

[0019] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present invention.

[0020] Please refer to Figure 1, which presents an optimized control system for an industrial park elevator system provided by an embodiment of the present invention. The hardware part includes a sensor unit, an actuator, a field control unit, communication equipment, and a server unit. The software part includes a real-time operating system, a sensor data acquisition and fusion preprocessing module, a digital twin simulation platform, a model predictive control module, and a cloud deep reinforcement learning module; The sensor unit includes traditional sensors and advanced environmental sensors. The traditional sensors include a position sensor, a speed sensor, a load sensor, a door status sensor, and a temperature and humidity sensor, which are used to collect basic data on elevator operation. The advanced environmental sensors include an energy consumption monitoring module and a vibration temperature sensor, which are used to collect auxiliary data on the environment and equipment operation. The data collected by the sensor unit is transmitted to the field control unit through the communication equipment for preliminary processing, and further transmitted to the server unit to support the calculation, analysis, and optimized scheduling of the software module; The actuator includes a driving device, a door machine, and environmental control equipment. The driving device includes a driving motor and a speed regulation module, which are used to drive the elevator car. The door machine is used to open and close the elevator door. The environmental control equipment is used to regulate environmental parameters such as temperature and humidity inside the elevator and related areas. The operating parameters of the actuator are determined by the control strategy calculated by the model predictive control module. This strategy is calculated by the cloud server module and sent to the field control unit through the communication equipment, and then forwarded by the field control unit to the actuator for final execution; The field control unit includes an industrial controller, which is installed on each elevator control terminal. The industrial controller integrates an edge computing module, which is used to achieve real-time data acquisition, preprocessing, local control, and safety monitoring. The field control unit collects and preliminarily processes the data of the sensor unit, and transmits it to the server unit through the communication equipment for analysis and optimized calculation of advanced algorithms. At the same time, it is responsible for receiving the control instructions sent by the server unit and performing corresponding local adjustments; The communication equipment includes an industrial Ethernet switch and a wireless data terminal, which are used to achieve data transmission and control instruction issuance between the field control unit, the server unit, and the digital twin simulation platform. The communication equipment ensures the efficient transmission of data between the sensor unit, the field control unit, and the server unit, and ensures the reliability and real-time nature of data interaction and control signals between the system modules; The server unit includes a local server module and a cloud server module. The local server module is responsible for short-cycle data caching and local optimized calculation, which is used to achieve data storage and rapid processing near the site. The cloud server module undertakes complex calculation tasks, including running deep reinforcement learning and optimized scheduling models, and generating control strategies to be sent to the field control unit for big data processing, global optimized scheduling, and running advanced intelligent algorithms; The real-time operating system is deployed on the industrial controller in the field control unit. The real-time operating system manages and schedules various computing tasks of the field control unit, ensuring real-time processing of sensor data and efficient response of the actuator. At the same time, it conducts data interaction with the server unit to ensure the stable operation of the system, so as to ensure the timely response of local control and safety monitoring tasks; The sensor data acquisition and fusion preprocessing module is deployed on the industrial controller in the field control unit, and is used to filter, verify, format and perform data fusion processing on the elevator operation data and environmental data from the sensor unit to ensure data quality. Special processing algorithms can be adopted according to different data characteristics. The output data of the sensor data acquisition and fusion preprocessing module is not only used for local control, but also transmitted to the cloud server module through communication equipment for digital twin simulation and optimization calculation; The digital twin simulation platform is deployed on the cloud server module, receives data from the sensor data acquisition and fusion preprocessing module, and constructs a virtual model of elevator operation and environmental control by combining historical operation data and real-time data, which is used to predict the system state and provide a basis for optimized scheduling, realizing real-time simulation of elevator operation state, energy flow and environmental control equipment state, and conducting data interaction with the field control unit through communication equipment; The model predictive control module is deployed on the cloud server module, works in coordination with the digital twin simulation platform, conducts optimized calculation based on the data provided by the digital twin simulation platform, is used to predict the elevator operation state in a future period of time, online solve the optimal control strategy, and send the optimized control strategy to the field control unit through communication equipment. The field control unit forwards it to the actuator, thereby dynamically adjusting the working states of the drive device, door machine and environmental control equipment to achieve precise control; The cloud deep reinforcement learning module is deployed on the cloud server module, and is used to train elevator scheduling, energy consumption management and safety decision-making models using big data and historical operation records, and periodically send the optimized models to the industrial controller in the field control unit through communication equipment to achieve global optimal scheduling and local real-time control. At the same time, the cloud deep reinforcement learning module continuously learns historical data and real-time data, improves the accuracy of the scheduling algorithm, and updates the optimized model parameters to the cloud server module to support the iterative upgrade of optimized calculation and scheduling decision-making; The real-time operating system works in collaboration with the sensor data acquisition and fusion preprocessing module to achieve real-time acquisition, preprocessing, and safety monitoring of on-site data; the digital twin simulation platform and the model predictive control module work together to achieve real-time simulation and prediction of the elevator operation status, energy flow, and environmental control equipment status, providing a basis for dynamic control strategies; the cloud deep reinforcement learning module collaborates with the industrial controller in the on-site control unit through the cloud server module to achieve effective integration of global optimization scheduling and local real-time control, thereby ensuring that the system achieves comprehensive optimization in terms of energy conservation, intelligent scheduling, safety protection, and personalized services.

[0021] Furthermore, the digital twin simulation platform adopts a hybrid modeling method that combines deep neural networks and physical modeling to achieve high-precision prediction of the elevator operation status, energy flow, and environmental control equipment status; This hybrid modeling method includes an adaptive parameter correction unit for dynamically adjusting model parameters according to real-time data, thereby further improving prediction accuracy and system robustness. Its specific algorithms include: Using the convolutional neural network CNN to extract the spatial features in the elevator operation time series data; Combining physical modeling constraints to construct a hybrid model, whose mathematical expression is: L = α·MSE(Y_pred, Y_actual)+β·‖F_phy(Y_pred)-C_phy‖² Where: α: Positive weight coefficient, used to adjust the proportion of the mean square error term in the overall loss function; MSE(Y_pred, Y_actual): Mean square error, measuring the average error between the predicted output Y_pred and the actual output Y_actual; Y_pred: The output value obtained by neural network prediction; Y_actual: The target data actually collected; β: Positive weight coefficient, used to adjust the proportion of the physical model error term in the overall loss function; F_phy(Y_pred): The calculation result of the predicted output Y_pred based on the physical modeling prediction function; C_phy: Preset physical constraint constant, used to define the ideal state of the physical model; ‖·‖²: Square Euclidean norm, used to quantify the magnitude of the physical model prediction error; Online update the neural network weights through the adaptive parameter correction unit to ensure that the model can dynamically adapt to real-time data changes, thereby improving prediction accuracy and system robustness.

[0022] Furthermore, the model predictive control module further includes a control strategy optimization unit based on adaptive weight allocation. This unit dynamically adjusts the weights of each control parameter using the real-time data and historical operation data provided by the industrial controller in the field control unit to achieve precise regulation of the drive device, door machine, and environmental control equipment. Its core uses the following formula for parameter optimization: u* = arg min_u{∑_(i=1)^Nw_i·J_i(u)} Where: u*: Optimal control input, i.e., the control variable that minimizes the total cost function; u: Control input variable to be optimized; N: Number of control objectives, i.e., the number of control objectives for the drive device, door machine, and environmental control equipment respectively; w_i: Adaptive weight of the i-th control objective, dynamically adjusted by the online algorithm according to real-time data; J_i(u): Cost function for the i-th control objective, reflecting the impact of the control input u on this objective; ∑_(i=1)^N: Sum of the cost functions for all control objectives; This algorithm realizes precise regulation of different control objects by optimizing the weight allocation of each sub-objective in real time, improving the comprehensive optimization effect of the control strategy.

[0023] Furthermore, the cloud deep reinforcement learning module adopts a deep Q-network algorithm based on a multi-objective reward function. It comprehensively considers energy consumption optimization, scheduling efficiency, and user comfort, continuously optimizes the elevator scheduling, energy consumption management, and safety decision-making strategies through online training, and updates the control strategy in real time according to the feedback data. The reward function R of the deep Q-network DQN algorithm based on the multi-objective reward function is defined as follows: R=γ1·(ΔE)+γ2·(ΔT)+γ3·(ΔS) Where: R: Reward value, used to evaluate the comprehensive effect of the current scheduling decision and reflect the achievement of the optimization objective; γ1, γ2, γ3: Positive weight parameters, respectively adjusting the proportions of energy consumption optimization, waiting time reduction, and safety index improvement in the overall reward; ΔE: Energy consumption reduction amount, indicating the degree of improvement in energy consumption achieved through optimized control; ΔT: Reduction amount of passenger waiting time, indicating the reduction amplitude of passenger waiting time and reflecting the improvement in service efficiency; ΔS: Improvement value of the safety index, indicating the degree of improvement in safety performance, and quantifying safety through monitoring indicators.

[0024] Further, the communication device further includes a fault-tolerant network protocol that supports redundant backup, ensuring higher reliability and real-time performance in data transmission between system modules in a harsh environment, thereby guaranteeing the stable operation of the entire system. The core algorithms of the fault-tolerant network protocol based on the redundant backup mechanism include: ① Implementing redundant storage and backup of data packets based on the distributed hash table DHT; ② Adopting fast retransmission and adaptive rate adjustment algorithms to ensure the real-time performance and integrity of data transmission even in case of communication anomalies; this solution effectively improves the reliability and data transmission efficiency of the system in a complex environment.

[0025] Further, the industrial controller in the field control unit integrates a security monitoring sub-module based on machine learning. The security monitoring sub-module adopts a security monitoring algorithm that combines support vector machine (SVM) and random forest (RF) to analyze elevator operation data in real time, automatically identify potential anomalies, and send early warning signals in advance, thereby realizing active prediction and prevention of system failures. Specifically as follows: ① The security monitoring module first conducts preliminary anomaly detection through SVM, using the formula: f(x) = sign(∑_(i=1)^l α_i y_i K(x_i, x) + b) Where: sign(·): Sign function, with positive output corresponding to +1 and negative output corresponding to -1; l: The number of support vectors, that is, the number of key samples selected during the training process; α_i: The weight coefficient corresponding to the i-th support vector, determined by the SVM training process; y_i: The class label of the i-th support vector, usually taking values of +1 or -1; K(x_i, x): Kernel function, used to calculate the similarity between the support vector x_i and the input sample x; b: Bias term, used to adjust the classification boundary; ② For the detected abnormal data, use RF for multi-level classification to ensure the accuracy and timeliness of fault warning; This combined algorithm can significantly improve the prediction and prevention ability of system failures.

[0026] Please refer to Figure 2 , which shows an optimized control method for an industrial park elevator system provided by another embodiment of the present invention, including the following steps: S1. Data acquisition and preliminary processing: Collect elevator operation and environmental data through traditional sensors and advanced environmental sensors in the sensor unit, including position, speed, load, door status, temperature and humidity, energy consumption, and vibration temperature information. Then, use the sensor data acquisition and fusion preprocessing module on the industrial controller in the field control unit to filter, verify, format, and fuse the collected multi-source data, and output high-quality fused data. S2. Data transmission and caching: Transmit the preprocessed data to the server unit through the communication device. The local server module realizes short-cycle data caching and fast response. At the same time, use the redundant communication mechanism to transmit the data to the cloud server module to provide a basis for subsequent big data analysis. S3. Digital twin modeling and system simulation: Use the digital twin simulation platform on the cloud server module to construct a virtual model of elevator operation and environmental control by combining historical data and real-time data. Then, use the hybrid modeling algorithm to perform real-time simulation and prediction on the elevator state, energy flow, and environmental control device state. S4. Generation and distribution of optimized control strategies: Based on the data output by the digital twin simulation platform, use the model predictive control module on the cloud server module to solve the optimal control strategy online, and send this control strategy to the field control unit through the communication device. Then, the field control unit forwards it to the actuator to achieve dynamic adjustment of the drive device, door machine, and environmental control device. S5. Online scheduling and deep reinforcement learning optimization: Use the cloud deep reinforcement learning module on the cloud server module to perform online training on the elevator scheduling, energy consumption management, and safety decision-making models. Adopt a multi-objective reward function to continuously optimize the control strategy, and periodically update the model parameters and control strategy, and transmit the optimization results to the field control unit. S6. Safety monitoring and exception handling: The industrial controller in the field control unit integrates a safety monitoring module to monitor the elevator operation status in real time and automatically detect exceptions. Once an exception is detected, the system immediately triggers an early warning mechanism, transmits the exception information to the server unit through the communication device, and at the same time, the field control unit executes the preset safety emergency control measures to ensure the safe operation of the system. S7. Closed-loop feedback and continuous optimization: The system feeds back the execution results and operation data of the field control unit to the cloud server module to form a closed-loop feedback mechanism, which supports the continuous iterative optimization of the digital twin simulation, model predictive control, and deep reinforcement learning modules, and realizes the continuous improvement of the overall performance of the system.

[0027] Furthermore, in step S3, the digital twin modeling and system simulation adopt a hybrid modeling algorithm, where the hybrid modeling algorithm combines a deep neural network and a physical model, and dynamically adjusts the model parameters according to the online feedback.

[0028] Further, in step S4, the model predictive control module adopts an adaptive weight allocation algorithm to precisely regulate each sub-control objective by optimizing the objective function \(J(u)=\sum_{i = 1}^{N}w_{i}\cdot J_{i}(u)\); Where: \(J(u)\): The overall optimization objective function. By weighted summation of each sub-objective, a comprehensive cost value is obtained, which takes the minimum value under the optimal control input, thereby achieving precise regulation; \(N\): Represents the number of control objectives, covering the control objectives of each subsystem such as the drive device, door machine, and environmental control equipment; \(w_{i}\): The adaptive weight of the \(i\)-th control objective, dynamically adjusted by an online algorithm according to real-time collected data to reflect the importance of each control objective; \(J_{i}(u)\): The cost function for the \(i\)-th control objective, quantifying the error or deviation degree generated by the control input \(u\) for this objective.

[0029] Further, in step S5, the cloud deep reinforcement learning module adopts a deep Q-network (DQN) algorithm based on a multi-objective reward function, and its reward function is \(R=\gamma_{1}\cdot(\Delta E)+\gamma_{2}\cdot(\Delta T)+\gamma_{3}\cdot(\Delta S)\) to achieve energy consumption optimization, waiting time reduction, and safety index improvement; Where: \(R\): The reward value, used to evaluate the comprehensive effect of the current scheduling decision and reflect the achievement of the optimization objective; \(\gamma_{1},\gamma_{2},\gamma_{3}\): Positive weight parameters, respectively adjusting the proportions of energy consumption optimization, waiting time reduction, and safety index improvement in the overall reward; \(\Delta E\): The amount of energy consumption reduction, indicating the degree of improvement in energy consumption achieved through optimal control; \(\Delta T\): The amount of reduction in passenger waiting time, indicating the reduction amplitude of passenger waiting time and reflecting the improvement in service efficiency; \(\Delta S\): The improvement value of the safety index, indicating the degree of improvement in safety performance, and quantifying safety through monitoring indicators.

[0030] The above content is only an example and explanation of the structure of the present invention. Those skilled in the art of this technology make various modifications or supplements to the described specific embodiments or use similar methods for substitution. As long as they do not deviate from the structure of the invention or exceed the scope defined by this claim book, they should fall within the protection scope of the present invention.

Claims

1. An optimized control system for an elevator system in an industrial park, characterized in that, The hardware part includes a sensor unit, an actuator, a field control unit, communication devices, and a server unit, and the software part includes a real-time operating system, a sensor data acquisition and fusion preprocessing module, a digital twin simulation platform, a model predictive control module, and a cloud deep reinforcement learning module; The sensor unit includes traditional sensors and advanced environmental sensors. The traditional sensors include position sensors, speed sensors, load sensors, door state sensors, and temperature and humidity sensors, which are used to collect basic data on elevator operation. The advanced environmental sensors include an energy consumption monitoring module and a vibration temperature sensor, which are used to collect auxiliary data on the environment and equipment operation. The data collected by the sensor unit is transmitted to the field control unit through the communication device for preliminary processing, and further transmitted to the server unit to support the calculation, analysis, and optimal scheduling of the software module; The actuator includes a driving device, a door machine, and environmental control equipment. The driving device includes a driving motor and a speed regulation module, which are used to drive the elevator car. The door machine is used to open and close the elevator door. The environmental control equipment is used to regulate environmental parameters such as temperature and humidity inside the elevator and related areas. The operating parameters of the actuator are determined by the control strategy calculated by the model predictive control module. This strategy is calculated by the cloud server module and sent to the field control unit through the communication device, and then forwarded by the field control unit to the actuator for final execution; The field control unit includes an industrial controller, which is installed on each elevator control terminal. The industrial controller integrates an edge computing module, which is used to achieve real-time data acquisition, preprocessing, local control, and safety monitoring. The field control unit collects and preprocesses the data of the sensor unit in real time, and transmits it to the server unit through the communication device for analysis and optimization calculation of advanced algorithms. At the same time, it is responsible for receiving the control instructions sent by the server unit and performing corresponding local adjustments; The communication device includes an industrial Ethernet switch and a wireless data terminal, which are used to achieve data transmission and control instruction issuance between the field control unit, the server unit, and the digital twin simulation platform. The communication device ensures the efficient transmission of data between the sensor unit, the field control unit, and the server unit, and ensures the reliability and real-time nature of data interaction and control signals between the system modules; The server unit includes a local server module and a cloud server module. The local server module is responsible for short-cycle data caching and local optimization calculation, and is used to achieve data storage and rapid processing near the site. The cloud server module undertakes complex calculation tasks, including running deep reinforcement learning and optimization scheduling models, and generating control strategies to be sent to the field control unit for big data processing, global optimization scheduling, and running advanced intelligent algorithms; The real-time operating system is deployed on the industrial controller in the field control unit. The real-time operating system manages and schedules various computing tasks of the field control unit, ensures the real-time processing of sensor data and the efficient response of the actuator, and at the same time conducts data interaction with the server unit to ensure the stable operation of the system, so as to ensure the timely response of local control and safety monitoring tasks; The sensor data acquisition and fusion preprocessing module is deployed on the industrial controller in the field control unit, and is used to filter, verify, format and perform data fusion processing on the elevator operation data and environmental data from the sensor unit to ensure data quality, and can adopt special processing algorithms according to different data characteristics. The output data of the sensor data acquisition and fusion preprocessing module is not only used for local control, but also transmitted to the cloud server module through communication equipment for digital twin simulation and optimization calculation; The digital twin simulation platform is deployed on the cloud server module, receives data from the sensor data acquisition and fusion preprocessing module, and constructs a virtual model of elevator operation and environmental control in combination with historical operation data and real-time data, which is used to predict the system state and provide a basis for optimized scheduling, realize the real-time simulation of elevator operation state, energy flow and environmental control equipment state, and conduct data interaction with the field control unit through communication equipment; The model predictive control module is deployed on the cloud server module and works in cooperation with the digital twin simulation platform. Based on the data provided by the digital twin simulation platform, it performs optimized calculation, is used to predict the elevator operation state in a future period of time, online solves the optimal control strategy, and sends the optimized control strategy to the field control unit through communication equipment. The field control unit forwards it to the actuator, so as to dynamically adjust the working states of the driving device, door machine and environmental control equipment to achieve precise control; The cloud deep reinforcement learning module is deployed on the cloud server module, and is used to train elevator scheduling, energy consumption management and safety decision-making models by using big data and historical operation records, and periodically send the optimized models to the industrial controller in the field control unit through communication equipment to achieve global optimal scheduling and local real-time control. At the same time, the cloud deep reinforcement learning module continuously learns historical data and real-time data, improves the accuracy of the scheduling algorithm, and updates the optimized model parameters to the cloud server module to support the iterative upgrade of optimized calculation and scheduling decision-making; The real-time operating system and the sensor data acquisition and fusion preprocessing module work together to achieve real-time acquisition, preprocessing and safety monitoring of field data; the digital twin simulation platform and the model predictive control module work together to achieve real-time simulation and prediction of elevator operation state, energy flow and environmental control equipment state, providing a basis for dynamic control strategies; the cloud deep reinforcement learning module collaborates with the industrial controller in the field control unit through the cloud server module to achieve the effective integration of global optimized scheduling and local real-time control, so as to ensure that the system achieves comprehensive optimization effects in terms of energy conservation, intelligent scheduling, safety protection and personalized services.

2. The optimized control system for an elevator system in an industrial park according to claim 1, wherein, The digital twin simulation platform adopts a hybrid modeling method combining deep neural network and physical modeling to achieve high-precision prediction of elevator operation status, energy flow, and environmental control equipment status; This hybrid modeling method includes an adaptive parameter correction unit, which is used to dynamically adjust model parameters according to real-time data, thereby further improving prediction accuracy and system robustness. Its specific algorithms include: Using a convolutional neural network (CNN) to extract spatial features from elevator operation time-series data; Combining physical modeling constraints to construct a hybrid model, whose mathematical expression is: L = α·MSE(Y_pred, Y_actual)+β·‖F_phy(Y_pred)-C_phy‖² Where: α: Positive weight coefficient, used to adjust the proportion of the mean square error term in the overall loss function; MSE(Y_pred, Y_actual): Mean square error, measuring the average error between the predicted output Y_pred and the actual output Y_actual; Y_pred: The output value obtained by neural network prediction; Y_actual: The target data actually collected; β: Positive weight coefficient, used to adjust the proportion of the physical model error term in the overall loss function; F_phy(Y_pred): The calculation result of the predicted output Y_pred based on the physical modeling prediction function; C_phy: Preset physical constraint constant, used to define the ideal state of the physical model; ‖·‖²: Square Euclidean norm, used to quantify the magnitude of the physical model prediction error; Online update the neural network weights through the adaptive parameter correction unit to ensure that the model can dynamically adapt to real-time data changes, thereby improving prediction accuracy and system robustness.

3. The optimized control system for an elevator system in an industrial park according to claim 1, characterized in that, The model predictive control module further includes a control strategy optimization unit based on adaptive weight allocation. This unit dynamically adjusts the weights of each control parameter using the real-time data and historical operation data provided by the industrial controller in the field control unit to achieve precise regulation of the drive device, door machine, and environmental control equipment. Its core uses the following formula for parameter optimization: u* = arg min_u{∑_(i=1)^Nw_i·J_i(u)} Where: u*: Optimal control input, that is, the control variable that makes the total cost function reach the minimum value; u: Control input variable to be optimized; N: Number of control objectives, that is, the number of control objectives for the drive device, door machine, and environmental control equipment respectively; w_i: Adaptive weight of the i-th control objective, dynamically adjusted by the online algorithm according to real-time data; J_i(u): Cost function for the i-th control objective, reflecting the impact of the control input u on this objective; ∑_(i=1)^N: Sum the cost functions of all control objectives; This algorithm realizes precise regulation of different control objects by optimizing the weight allocation of each sub-objective in real time, improving the comprehensive optimization effect of the control strategy.

4. The optimized control system for an elevator system in an industrial park according to claim 1, wherein, The cloud deep reinforcement learning module adopts the deep Q-network (DQN) algorithm based on a multi-objective reward function, comprehensively considering energy consumption optimization, scheduling efficiency, and user comfort. Through online training, it continuously optimizes elevator scheduling, energy consumption management, and safety decision-making strategies, and updates the control strategy in real-time according to the feedback data. The reward function R of the deep Q-network (DQN) algorithm based on the multi-objective reward function is defined as follows: R = γ1·(ΔE) + γ2·(ΔT) + γ3·(ΔS) Where: R: Reward value, used to evaluate the comprehensive effect of the current scheduling decision and reflect the achievement of optimization goals; γ1, γ2, γ3: Positive weight parameters, respectively adjusting the proportions of energy consumption optimization, waiting time reduction, and safety index improvement in the overall reward; ΔE: Reduction in energy consumption, indicating the degree of improvement in energy consumption achieved through optimized control; ΔT: Reduction in passenger waiting time, indicating the reduction amplitude of passenger waiting time and reflecting the improvement in service efficiency; ΔS: Improvement value of the safety index, indicating the degree of improvement in safety performance, and quantifying safety through monitoring indicators.

5. An optimized control system for an elevator system in an industrial park according to claim 1, characterized in that, The communication device further includes a fault-tolerant network protocol that supports redundant backup, ensuring higher reliability and real-time performance of data transmission between system modules in a harsh environment, thereby guaranteeing the stable operation of the entire system. The core algorithms of the fault-tolerant network protocol based on the redundant backup mechanism include: ① Implementing redundant storage and backup of data packets based on the distributed hash table (DHT); ② Adopting a fast retransmission and adaptive rate adjustment algorithm to ensure the real-time performance and integrity of data transmission in case of communication anomalies; this solution effectively improves the reliability and data transmission efficiency of the system in a complex environment.

6. The optimized control system of an elevator system in an industrial park according to claim 1, wherein The industrial controller in the field control unit integrates a safety monitoring sub-module based on machine learning. The safety monitoring sub-module adopts a safety monitoring algorithm that combines support vector machine (SVM) and random forest (RF) to analyze elevator operation data in real-time, automatically identify potential anomalies, and send early warning signals in advance, thereby realizing active prediction and prevention of system failures. Specifically as follows: ① The safety monitoring module first performs preliminary anomaly detection through SVM, using the formula: f(x) = sign(∑_(i = 1)^l α_i y_i K(x_i, x) + b) Where: sign(·): Sign function, with a positive output corresponding to +1 and a negative output corresponding to -1; l: Number of support vectors, that is, the number of key samples selected during the training process; α_i: Weight coefficient corresponding to the i-th support vector, determined by the SVM training process; y_i: Class label of the i-th support vector, usually taking values of +1 or -1; K(x_i, x): Kernel function, used to calculate the similarity between the support vector x_i and the input sample x; b: Bias term, used to adjust the classification boundary; ② For the detected abnormal data, use RF for multi-level classification to ensure the accuracy and timeliness of fault warnings; This combined algorithm can significantly improve the prediction and prevention ability of system failures.

7. An optimized control method for an elevator system in an industrial park, characterized in that, Including the following steps: S1. Data collection and preliminary processing: Collect elevator operation and environmental data, including position, speed, load, door status, temperature and humidity, energy consumption, and vibration temperature information, through traditional sensors and advanced environmental sensors in the sensor unit. Then, use the sensor data acquisition and fusion preprocessing module on the industrial controller in the field control unit to filter, verify, format, and fuse the collected multi-source data, and output high-quality fused data; S2. Data transmission and caching: Transmit the preprocessed data to the server unit through communication devices. The local server module realizes short-cycle data caching and fast response. At the same time, use a redundant communication mechanism to transmit the data to the cloud server module, providing a basis for subsequent big data analysis; S3. Digital twin modeling and system simulation: Use the digital twin simulation platform on the cloud server module to construct a virtual model of elevator operation and environmental control by combining historical data and real-time data. Then, use a hybrid modeling algorithm to perform real-time simulation and prediction on elevator status, energy flow, and environmental control device status; S4. Generation and distribution of optimized control strategies: Based on the data output by the digital twin simulation platform, use the model predictive control module on the cloud server module to online solve the optimal control strategy, and transmit this control strategy to the field control unit through communication devices. Then, the field control unit forwards it to the actuator to achieve dynamic adjustment of the drive device, door machine, and environmental control devices; S5. Online scheduling and deep reinforcement learning optimization: Use the cloud deep reinforcement learning module on the cloud server module to online train the elevator scheduling, energy consumption management, and safety decision-making models. Adopt a multi-objective reward function to continuously optimize the control strategy, and periodically update the model parameters and control strategy, and transmit the optimization results to the field control unit; S6. Safety monitoring and exception handling: The industrial controller in the field control unit integrates a safety monitoring module to monitor the elevator operation status in real time and automatically detect exceptions. Once an exception is detected, the system immediately triggers an early warning mechanism, transmits the exception information to the server unit through communication devices, and at the same time, the field control unit executes preset safety emergency control measures to ensure the safe operation of the system; S7. Closed-loop feedback and continuous optimization: The system feeds back the execution results and operation data of the field control unit to the cloud server module to form a closed-loop feedback mechanism, supporting the continuous iterative optimization of the digital twin simulation, model predictive control, and deep reinforcement learning modules, and realizing the continuous improvement of the overall performance of the system.

8. The optimized control method for an elevator system in an industrial park according to claim 7, wherein, In step S3, the digital twin modeling and system simulation adopt a hybrid modeling algorithm, where the hybrid modeling algorithm combines a deep neural network and a physical model, and dynamically adjusts the model parameters according to online feedback.

9. The optimization control method for an elevator system in an industrial park according to claim 7, wherein, In step S4, the model predictive control module adopts an adaptive weight allocation algorithm to achieve precise regulation of each sub-control objective by optimizing the objective function \(J(u)=\sum_{i = 1}^{N}w_i\cdot J_i(u)\); Where: J(u): The overall optimization objective function. By weighted summation of each sub-objective, a comprehensive cost value is obtained, which takes the minimum value under the optimal control input, thereby achieving precise regulation. N: Represents the number of control objectives, covering the control objectives of each subsystem such as the drive device, door machine, and environmental control equipment. w_i: The adaptive weight of the i-th control objective, dynamically adjusted by an online algorithm according to real-time collected data, used to reflect the importance of each control objective. J_i(u): The cost function for the i-th control objective, quantifying the error or deviation degree generated by the control input u for this objective.

10. A method for optimizing the control of an elevator system in an industrial park according to claim 7, characterized in that, In the step S5, the cloud deep reinforcement learning module adopts the deep Q-network DQN algorithm based on a multi-objective reward function, and its reward function is R = γ1·(ΔE) + γ2·(ΔT)+γ3·(ΔS), to achieve energy consumption optimization, waiting time reduction, and safety index improvement. Where: R: The reward value, used to evaluate the comprehensive effect of the current scheduling decision, reflecting the achievement of the optimization objective. γ1, γ2, γ3: Positive weight parameters, respectively adjusting the proportions of energy consumption optimization, waiting time reduction, and safety index improvement in the overall reward. ΔE: The amount of energy consumption reduction, indicating the degree of improvement in energy consumption achieved through optimal control. ΔT: The reduction in passenger waiting time, indicating the reduction amplitude of passenger waiting time, reflecting the improvement in service efficiency. ΔS: The improvement value of the safety index, indicating the degree of improvement in safety performance, and quantifying safety through monitoring indicators.

Images