Coke oven header pressure control system and method based on artificial intelligence

Abstract

The application discloses a coke oven gas collecting pipe pressure control system and method based on artificial intelligence, belongs to the technical field of coke oven coking, and solves the problems of slow response, low control precision and large pressure fluctuation of the existing control system. The control system comprises a pressure sensor, a data acquisition module, an AI processing module and an executing mechanism which are sequentially signal-connected, and further comprises a cloud platform which is in communication connection with the AI processing module. The control method comprises the following steps: pressure data acquisition, data preprocessing, AI model analysis and trend prediction, control signal generation, pressure closed-loop regulation, data storage and model iterative optimization. The application has the advantages of high control precision, fast response speed, strong working condition adaptability, self-learning ability, closed-loop precise regulation and control, improved production safety, reduced energy consumption, improved intelligent management level, high system reliability and fault self-diagnosis capability.

Description

Technical Field

[0001] This invention belongs to the field of coke oven coking technology, specifically relating to a coke oven gas collecting pipe pressure control system and method based on artificial intelligence. Background Technology

[0002] In the coke oven production process, stable control of the gas collecting pipe pressure is a core element in ensuring production safety, improving coke quality, and reducing environmental pollution. If the gas collecting pipe pressure is too high, it can easily cause gas leakage, leading to safety accidents and wasting energy; if the pressure is too low, it can cause air to be drawn into the gas collecting pipe, causing gas combustion or even an explosion, while also affecting the quality of the refined coke product.

[0003] Currently, the mainstream methods for controlling the pressure of gas collecting pipes in the industry are PID control and fuzzy control. However, they have significant technical defects in practical applications: the parameters of PID control are fixed and cannot be adaptively adjusted according to dynamic operating conditions such as coke oven coal charging and coke pushing, resulting in slow response and large overshoot; fuzzy control is based on a rule base formulated by human experience for regulation, which has poor flexibility and cannot explore the temporal characteristics and potential change patterns of pressure data. It is not adaptable to complex operating conditions and is prone to low control accuracy and large pressure fluctuations.

[0004] With the development of artificial intelligence technology in industrial process control, algorithms such as machine learning and neural networks have provided new solutions for the precise control of industrial parameters. However, existing technologies lack specific solutions for combining improved deep learning algorithms with closed-loop control in coke oven gas collecting pipe pressure control, and have not yet achieved remote iterative optimization of algorithm models or full-condition self-learning of the system. Therefore, developing an AI-based coke oven gas collecting pipe pressure control system and method with high response speed, high control accuracy, and strong adaptability to operating conditions has become an urgent technical problem to be solved in the industry. Summary of the Invention

[0005] The purpose of this invention is to provide an artificial intelligence-based coke oven gas collecting pipe pressure control system to solve the problems of slow response, low control accuracy, and large pressure fluctuation in existing control systems.

[0006] Another objective of this invention is to provide a coke oven gas collecting pipe pressure control method based on artificial intelligence.

[0007] The technical solution of this invention is: (one) An artificial intelligence-based coke oven gas collecting pipe pressure control system includes a pressure sensor, a data acquisition module, an AI processing module, and an actuator connected in sequence, and also includes a cloud platform that is communicatively connected to the AI ​​processing module. Pressure sensors are installed at key monitoring points in the coke oven gas collecting pipe to collect simulated pressure signals in the gas collecting pipe in real time. The AI ​​processing module has a built-in algorithm model with an improved LSTM neural network and an attention mechanism. It is configured to receive standardized digital pressure data, and after data cleaning and normalization preprocessing, it uses the algorithm model to mine the temporal characteristics of the pressure data, predict the pressure change trend, and adaptively generate precise control signals that match the working conditions based on the prediction results. The actuator is an electric regulating valve or a variable frequency fan with closed-loop feedback, used to receive the control signal and adjust the action parameters to achieve dynamic control of the gas collecting pipe pressure; the electric regulating valve is installed on the gas output pipeline of the coke oven gas collecting pipe, and the variable frequency fan is installed in the coke oven gas blower unit or the negative pressure regulating pipeline of the coke oven gas collecting pipe.

[0008] As a further improvement of the present invention, the actuator has a built-in pressure feedback sensor, which is connected to the AI ​​processing module. The pressure feedback sensor can transmit the actual adjusted air collection pipe pressure data back to the AI ​​processing module in real time, forming a closed-loop control loop of pressure acquisition - trend prediction - command output - execution feedback - model correction.

[0009] As a further improvement of the present invention, the data acquisition module includes a signal amplifier, a filter and an analog-to-digital converter connected in sequence, which are used to amplify, filter and reduce noise of the pressure analog signal and convert it into standardized digital pressure data.

[0010] As a further improvement of the present invention, the AI ​​processing module is embedded in an industrial-grade PLC or an edge computing industrial control computer.

[0011] As a further improvement of the present invention, the cloud platform is used to store historical stress data, algorithm model training data and system operating parameters, and can remotely iteratively train and optimize the algorithm model of the AI ​​processing module based on massive historical data. The AI ​​processing module establishes bidirectional communication with the cloud platform to realize data interaction and model synchronization. (two) A coke oven gas collecting pipe pressure control method based on artificial intelligence, employing the aforementioned coke oven gas collecting pipe pressure control system based on artificial intelligence, includes the following steps: S1. Pressure data acquisition: Pressure sensors installed at key points in the gas collection pipe acquire simulated pressure signals in real time and transmit them to the data acquisition module. S2. Data preprocessing: The data acquisition module converts the amplified and noise-reduced analog pressure signal into standardized digital pressure data. After data verification and removal of invalid data, the data is transmitted to the AI ​​processing module. S3, AI Model Analysis and Trend Prediction: The AI ​​processing module cleans and normalizes the received digital stress data, removes outliers and missing values, and inputs the preprocessed data into an algorithm model that integrates an improved LSTM neural network with an attention mechanism. This algorithm model assigns high weights to the key features of the stress time series data through the attention mechanism. The LSTM neural network then completes the prediction of the stress change trend at different time gradients and outputs the stress deviation prediction value and the trend change rate. S4. Control signal generation: The AI ​​processing module adaptively generates a control signal that matches the current operating conditions based on the predicted pressure deviation value, the trend change rate, and the threshold set by the coke oven gas collecting pipe pressure. S5. Pressure Closed-Loop Regulation: After receiving the control signal, the actuator adjusts the opening of the electric regulating valve or the speed of the variable frequency fan. At the same time, it collects the actual pressure data after adjustment through the built-in pressure feedback sensor and transmits it back to the AI ​​processing module in real time. The AI ​​processing module compares the actual pressure data with the set threshold. If there is a deviation, it corrects the control signal in time to realize the closed-loop regulation of pressure. S6. Data storage and model iteration optimization: The AI ​​processing module uploads real-time pressure data, control signals, and execution feedback data to the cloud platform for encrypted storage; the cloud platform performs remote iterative training and fine-tuning of the algorithm model based on massive historical data and newly added operating data, and distributes the optimized algorithm model to the AI ​​processing module to realize the model's self-learning and adaptive updates, adapting to different production conditions of the coke oven.

[0013] Furthermore, in step S3, the improved LSTM neural network algorithm model, which integrates the attention mechanism, strengthens the weights of key features in the stress time series data through the attention mechanism, suppresses interference from invalid and redundant data, and the LSTM neural network completes multi-dimensional prediction of the short-term and medium-term trends of stress based on the weighted feature data. The prediction time dimension includes three gradients: 10s, 30s, and 60s.

[0014] Furthermore, in step S3, the normalization preprocessing adopts the max-min normalization method to map the digital pressure data to the [0,1] interval, eliminating the influence of dimensions.

[0015] Furthermore, in step S3, outlier removal is performed using the 3σ criterion, identifying and removing abnormal pressure data that deviate from the data mean by three times the standard deviation.

[0016] This invention achieves precise, dynamic, and closed-loop control of the gas collecting pipe pressure, while possessing self-learning and adaptive capabilities, adapting to different production conditions in coke ovens, and solving the problems of slow response, poor adaptability, and low control accuracy of traditional control methods. Compared with existing technologies, this invention has the following advantages: 1. High control precision and fast response speed: This invention adopts an algorithm model with an improved LSTM neural network and an attention mechanism, which can mine the temporal characteristics and potential change patterns of pressure data. By strengthening key features through the attention mechanism, it can achieve accurate multi-gradient prediction of pressure change trends. The data processing latency of the AI ​​processing module is ≤50ms. Compared with traditional PID control and fuzzy control, the control precision is improved by ≥30% and the response speed is improved by ≥50%, effectively avoiding pressure overshoot and fluctuation.

[0017] 2. Strong adaptability to operating conditions and self-learning ability: The system realizes remote iterative optimization of the algorithm model through the cloud platform. Based on the new data of dynamic operating conditions such as coke oven coal charging and coke pushing, the model is fine-tuned, so that the model can continuously learn new pressure change patterns without manual parameter adjustment. This solves the problems of fixed parameters and rigid rule base in traditional control methods and adapts to the operating condition changes of the entire coke oven production cycle.

[0018] 3. Achieve closed-loop precise control and improve production safety: The system constructs a fully closed-loop control loop of pressure acquisition, trend prediction, command output, execution feedback and model correction. The feedback data from the actuator can correct the control signal in real time, effectively avoiding problems such as gas leakage and air intake, reducing the risk of safety accidents, and ensuring the quality of coke products.

[0019] 4. Reduce energy consumption and improve intelligent management: This invention avoids the ineffective operation of fans and regulating valves through precise pressure control, thereby reducing energy consumption; at the same time, the cloud platform realizes remote data storage, remote system monitoring and remote model optimization. Staff can use the cloud platform to monitor the pressure status of the gas collecting pipe and the system operation in real time, realizing intelligent and remote management of the coke oven gas collecting pipe pressure.

[0020] 5. High system reliability and self-diagnostic capability: All modules of the system use industrial-grade equipment, which is suitable for the harsh production environment of coke ovens; at the same time, a self-diagnostic procedure is set up to monitor the equipment status in real time, and to immediately alarm and switch emergency control strategies when a fault occurs, so as to ensure that the pressure of the gas collecting pipe is always within a safe range, thereby improving the system's operational reliability and production continuity. Attached Figure Description

[0021] Figure 1 This is a schematic diagram of the structure of a coke oven gas collecting pipe pressure control system based on artificial intelligence according to the present invention; Figure 2 This is a process flow diagram of a coke oven gas collecting pipe pressure control method based on artificial intelligence according to the present invention; Figure 3 This is a flowchart of the internal algorithm processing of the AI ​​processing module in this invention.

[0022] In the diagram, 1-pressure sensor; 2-data acquisition module; 3-AI processing module; 4-actuator; 6-cloud platform. Detailed Implementation

[0023] The present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0024] Example 1 like Figure 1 As shown, a coke oven gas collecting pipe pressure control system based on artificial intelligence includes a pressure sensor 1, a data acquisition module 2, an AI processing module 3 and an actuator 4 connected in sequence, and also includes a cloud platform 6 that is communicatively connected to the AI ​​processing module 3. Pressure sensor 1 is a high-precision diffused silicon pressure transmitter, installed at key monitoring points in the coke oven gas collecting pipe. It has a measurement accuracy of ≤±0.1%FS and a protection level of ≥IP65. It can collect the pressure simulation signal in the gas collecting pipe in real time at a sampling frequency of 50Hz. It has the characteristics of high temperature resistance and dust resistance, and is suitable for the harsh production environment of coke oven.

[0025] The data acquisition module 2 consists of a signal amplifier, a filter, and an analog-to-digital converter. It amplifies and filters the pressure analog signal in sequence to eliminate signal errors caused by environmental interference. Then, the analog signal is converted into 16-bit standardized digital pressure data by the analog-to-digital converter. After data verification, the data is transmitted to the AI ​​processing module to ensure the accuracy and standardization of the data.

[0026] AI processing module 3 is embedded in an industrial-grade PLC or edge computing industrial control computer, equipped with a Linux real-time operating system, with a data processing latency of ≤50ms. It has a built-in improved LSTM neural network with an attention mechanism algorithm model, which can clean and normalize the received digital pressure data. Through the algorithm model, it can mine the temporal characteristics of the pressure data, complete the prediction of pressure change trends at three time gradients of 10s, 30s, and 60s, and adaptively generate control signals based on the prediction results. It also supports multi-protocol communication such as Modbus-RTU and Profinet to achieve efficient data interaction with various modules.

[0027] Actuator 4 is an electric regulating valve or variable frequency fan with closed-loop feedback. It has a built-in pressure feedback sensor that can receive control signals from the AI ​​processing module, adjust the valve opening or fan speed, and simultaneously transmit the actual adjusted pressure data back to the AI ​​processing module in real time to form closed-loop control. It adopts a graded control strategy to achieve three adjustment modes: micro-amplitude gradual, normal gradient, and emergency rapid, based on the pressure deviation, to avoid overshoot.

[0028] The cloud platform 6 adopts a distributed cloud storage architecture, supports encrypted data transmission and breakpoint resume, and is used to store historical stress data, algorithm model training data and system operating parameters. It has a built-in model optimization engine that can remotely iterate and fine-tune the algorithm model based on massive historical data and new working condition data using the small-batch gradient descent method. The optimized model can be remotely sent to the AI ​​processing module to realize the model's self-learning and adaptive update.

[0029] like Figure 2 As shown, an artificial intelligence-based coke oven gas collecting pipe pressure control method is adopted, which uses the aforementioned artificial intelligence-based coke oven gas collecting pipe pressure control system. The method includes the following steps: S1. Pressure data acquisition: Pressure sensor 1 installed at key points in the gas collection pipe acquires the simulated pressure signal in the gas collection pipe in real time at a sampling frequency of 50Hz, and transmits the signal to the data acquisition module 2 in real time. S2. Data preprocessing: The data acquisition module 2 amplifies and filters the pressure analog signal in sequence to eliminate signal errors caused by environmental interference such as electromagnetic fields and dust in the coke oven production environment. Then, the amplified and noise-reduced pressure analog signal is converted into 16-bit standardized digital pressure data through an analog-to-digital converter. After invalid data is removed by data verification, the data is transmitted to the AI ​​processing module 3 to ensure the accuracy and standardization of the data. S3. AI Model Analysis and Trend Prediction: AI processing module 3 preprocesses the received digital stress data, using the 3σ criterion to remove outliers and fill in missing values. Then, it uses the max-min standardization method to map the stress data to the [0,1] interval to eliminate the influence of dimensions. The preprocessed stress data is then input into an algorithm model that integrates an improved LSTM neural network with an attention mechanism. This model uses the attention mechanism to strengthen the weights of key features (such as stress mutation points and trend inflection points) in the stress time series data, suppressing the interference of invalid and redundant data. The LSTM neural network then uses the weighted feature data to predict the stress change trend at three time gradients of 10s, 30s, and 60s, and outputs the predicted stress deviation value and the trend change rate. S4. Control signal generation: The AI ​​processing module 3 pre-stores the set threshold of the coke oven gas collecting pipe pressure. Based on the pressure deviation prediction value and trend change rate output in step S3, and combined with the current coke oven production conditions (coal charging, coke pushing, normal coking), it adaptively generates control signals. The control signals include the action type of the actuator (valve opening adjustment / fan speed adjustment), action amplitude and action duration. S5. Pressure Closed-Loop Regulation: After receiving the control signal, the actuator 4 adjusts the opening of the electric regulating valve or the speed of the variable frequency fan according to the graded regulation strategy. At the same time, the pressure feedback sensor built into the actuator 4 collects the actual pressure data after regulation at a sampling frequency of 50Hz and transmits it back to the AI ​​processing module 3 in real time. The AI ​​processing module 3 compares the actual pressure data with the set threshold in real time. If there is a deviation, it immediately corrects the control signal and sends it to the actuator for secondary regulation, forming a closed-loop control loop of pressure acquisition-trend prediction-command output-execution feedback-model correction, realizing dynamic and precise pressure regulation. The actuator's action parameter adjustment adopts a graded regulation strategy: when the predicted pressure deviation is ≤5Pa, a micro-gradual adjustment is performed; when 5Pa < the predicted pressure deviation is ≤20Pa, a conventional gradient adjustment is performed; when the predicted pressure deviation is >20Pa, an emergency rapid adjustment is performed, and the action amplitude is limited in real time during the adjustment process to avoid overshoot. S6. Data Storage and Model Iteration Optimization: AI processing module 3 integrates real-time pressure data, control signals, execution feedback data, and operating condition data, and uploads them to cloud platform 6 for distributed storage via 5G / Ethernet encryption. The model optimization engine of cloud platform 6, based on massive historical data and newly added operating data, uses the small-batch gradient descent method to remotely iteratively train and fine-tune the algorithm model of AI processing module 3, explore the pressure change patterns under new operating conditions, and optimize the model's prediction accuracy and control strategy. The optimized algorithm model is sent to AI processing module via encrypted communication to achieve self-learning and adaptive updates of the model, enabling the system to continuously adapt to different production conditions of the coke oven. S7. Fault Self-Diagnosis: The AI ​​processing module 3 monitors the operating status and data transmission status of the pressure sensor 1, data acquisition module 2, and actuator 4 in real time. If a fault is detected (such as sensor failure or valve jamming) or abnormal data transmission (such as signal interruption or data loss), an audible and visual fault alarm signal is immediately issued, and the system automatically switches to the preset emergency control strategy to maintain the pressure in the gas collection pipe within a safe range. At the same time, the fault type, fault location, and fault time are uploaded to the cloud platform to facilitate remote troubleshooting and maintenance by staff.

[0030] The internal processing flow of AI processing module 3 is as follows: Figure 3 As shown, the process is as follows: Start → Data Input (receiving digital pressure data from the data acquisition module) → Data Preprocessing (cleaning, normalization, outlier removal) → Attention Mechanism Feature Weighting (strengthening key features and suppressing redundant data) → LSTM Neural Network Trend Prediction (10s / 30s / 60s gradient) → Control Decision (generating control signals) → Signal Output (sent to the actuator) → Feedback Data Reception (actual pressure data from the actuator) → Real-time Model Correction → End.

[0031] In this embodiment, pressure sensor 1 is connected to data acquisition module 2 by cable, data acquisition module 2 is connected to AI processing module 3 via Profinet bus, AI processing module 3 is connected to actuator 4 by control cable and actuator 4 establishes feedback signal connection with AI processing module 3, and AI processing module 3 communicates bidirectionally with cloud platform 6 via 5G network; actuator 4 is an electric regulating valve or variable frequency fan, and pressure feedback sensor is embedded inside electric regulating valve or variable frequency fan.

[0032] In this embodiment, the pressure setting threshold of the coke oven gas collecting pipe is 80±5Pa. Pressure sensor 1 is a diffused silicon pressure transmitter with a measurement range of 0-200Pa, an accuracy of ±0.1%FS, and an IP67 protection rating. Data acquisition module 2 is an industrial-grade multi-channel acquisition card that supports signal amplification, filtering, and 16-bit AD conversion. AI processing module 3 is embedded in an edge computing industrial control computer, equipped with a Linux real-time operating system, and has a built-in algorithm model with an improved LSTM neural network fusion attention mechanism, with a data processing latency of ≤40ms. Actuator 4 is a combination of an electric regulating valve and a variable frequency fan. The regulating accuracy of the electric regulating valve is 0.5%, and the speed range of the variable frequency fan is 0-50Hz. Cloud platform 6 adopts Alibaba Cloud's distributed cloud storage architecture, is equipped with a self-developed model optimization engine, and supports 5G wireless communication.

[0033] During implementation, pressure sensor 1 is installed at three key monitoring points: the connection between the riser pipe and the gas collecting pipe of the coke oven gas collecting pipe, and the main gas collecting pipe pipeline. It collects simulated pressure signals in real time at a sampling frequency of 50Hz. Data acquisition module 2 amplifies, filters, and reduces noise in the simulated signals, converts them into 16-bit digital pressure data, and transmits it to AI processing module 3. AI processing module 3 cleans and normalizes the digital pressure data, then inputs it into an algorithm model that integrates an improved LSTM neural network with an attention mechanism. This model analyzes the temporal characteristics of the pressure data and predicts pressure trends for 10s, 30s, and 60s. When it predicts that the pressure will deviate from the set threshold, it generates a control signal. After receiving the control signal, actuator 4 adjusts the opening of the electric regulating valve or the speed of the variable frequency fan in stages, while simultaneously transmitting the actual pressure data back to the AI ​​processing module to form a closed-loop regulation. AI processing module 3 uploads all operating data to cloud platform 6. Based on the daily added production data, cloud platform 6 uses a small-batch gradient descent method to fine-tune the algorithm model. The optimized model is then sent back to AI processing module 3 every morning to achieve iterative updates.

[0034] This embodiment was applied to a 4.3m coke oven in a steel plant. After actual operation and testing, the pressure fluctuation range of the gas collecting pipe was controlled within 80±2Pa. Compared with traditional fuzzy control, the pressure control accuracy was improved by 35%, the response speed was improved by 60%, the energy consumption of the variable frequency fan was reduced by 18%, and no problems such as gas leakage or air intake occurred throughout the process, effectively ensuring the safety of coke oven production and the quality of coke.

[0035] This invention addresses the core technical deficiencies of traditional coke oven gas collecting pipe pressure control systems (PID control, fuzzy control) in practical applications. It deeply integrates artificial intelligence and machine learning technologies into the entire pressure control process, achieving innovative breakthroughs in control algorithms, adaptability, control accuracy, and system scalability. Compared to traditional systems, it possesses fundamental technical advantages, specifically in the following five aspects: First, the control algorithm has been improved from "fixed rules / parameters" to "intelligent learning and prediction".

[0036] Traditional PID control uses proportional-integral-derivative adjustment with fixed parameters, while fuzzy control is based on a fixed rule base established by human experience. Neither of them has the ability to learn data or predict trends. They can only passively adjust for pressure deviations that have already occurred and cannot predict the trend of pressure changes.

[0037] This invention introduces an AI processing module based on machine learning / neural network / deep learning algorithms, which can perform in-depth analysis of the time series data of gas collection tube pressure, mine the inherent change patterns of the data, and realize the proactive prediction of pressure change trends. It upgrades from "passive correction" to "proactive prediction + precise control", making pressure regulation more forward-looking and solving the core problem of lag in response of traditional systems from the algorithm level.

[0038] Second, the adaptability to operating conditions has been improved from "single operating condition adaptation" to "full operating condition self-adaptation".

[0039] Traditional control systems cannot adjust parameters / rules automatically once they are set. When the operating conditions of a coke oven change dynamically during coal charging, coke pushing, and normal coking, the system cannot adapt, which can easily lead to large pressure fluctuations and control failures. Frequent manual adjustment of parameters / rules is required, which is cumbersome and has low accuracy.

[0040] The AI ​​processing module of this invention has self-learning and self-adaptation capabilities. It can continuously train and optimize the model through historical pressure data and real-time operating condition data, automatically adapt to the pressure change characteristics of different production conditions of coke ovens, and achieve stable control under all operating conditions without manual intervention. This completely solves the technical problem of "poor adaptability to changes in operating conditions" in traditional systems.

[0041] Third, the control precision has been improved from "extensive adjustment" to "precise and refined adjustment".

[0042] Traditional systems lack trend prediction, can only be passively adjusted, and are limited by fixed parameters / rules. As a result, they are prone to problems such as overshoot, undershoot, and lag in adjusting pressure deviations. They have low pressure control accuracy and poor stability, and the pressure in the gas collection pipe is prone to large fluctuations, which can lead to the risk of gas leakage or air intake.

[0043] This invention utilizes the precise trend prediction and optimized control signal generation of the AI ​​processing module, combined with the refined signal preprocessing (filtering, amplification, analog-to-digital conversion) of the data acquisition module, to make the adjustment actions of the actuator more precise and the amplitude more suitable. This effectively avoids overshoot and undershoot, significantly improves the accuracy and stability of pressure control, and ensures the safety of coke oven pressure control at the system level.

[0044] Fourth, the system architecture has been improved from a "single closed loop" to an "intelligent data closed loop + scalable architecture".

[0045] Traditional control systems are mostly simple "acquisition-regulation" hardware closed loops, which only realize pressure data acquisition and basic adjustment of actuators, without professional data processing steps such as data cleaning and normalization. Data interference can easily lead to regulation errors.

[0046] This invention constructs an intelligent data closed loop of "pressure acquisition - data preprocessing - AI analysis and prediction - control output - pressure regulation". The AI ​​processing module adds a professional data preprocessing process (cleaning and normalization) to eliminate the influence of invalid and interfering data, making the generation of control signals more in line with the actual pressure change requirements. At the same time, the system adopts a modular architecture design, with each module having independent functions and close connection, making data transmission and command execution more efficient.

[0047] Compared to the "non-scalable" nature of traditional systems, this invention designs an optional cloud platform module, which breaks through the limitations of local control in traditional systems and provides an expansion foundation for subsequent remote monitoring, data storage, and remote model optimization and updates, thus realizing the combination of local intelligent control and remote data management.

[0048] Fifth, data utilization has been improved from "no data reuse" to "data-driven continuous optimization".

[0049] Traditional control systems only use pressure data as a basis for real-time adjustment, without the ability to store and reuse data. A large amount of data generated during system operation is wasted, and the system has no autonomous optimization capability. During use, the control performance will gradually decline as the equipment ages and the operating conditions change.

[0050] On the one hand, this invention enables remote storage and management of pressure data, control data, and operating condition data through a cloud platform, making industrial data a reusable asset. On the other hand, it can continuously optimize and train AI models based on stored historical and real-time data, enabling autonomous iterative upgrades of the AI ​​processing module's control strategy. This allows the system's control performance to continuously optimize over time rather than gradually decline, achieving continuous evolution of the data-driven system.

[0051] VI. Supporting modules make data collection more aligned with the needs of intelligent algorithms.

[0052] Traditional control systems lack specialized modular design in their signal acquisition stages, often relying on simple signal conversion. This makes them susceptible to interference from electromagnetic fields and dust in the coke oven production environment, resulting in low signal accuracy and consequently affecting the control performance.

[0053] This invention features an innovative design for the data acquisition module, explicitly integrating a signal amplifier and an analog-to-digital converter. It specifically addresses the issues of weak pressure signals and high interference at the coke oven site, achieving amplification, noise reduction, and standardized digital conversion of analog signals. This provides a high-quality, standardized data source for the AI ​​processing module's accurate analysis and prediction, ensuring the effectiveness of intelligent control from the data acquisition source and achieving precise matching between the acquisition module and the AI ​​algorithm.

[0054] In summary, this invention employs an improved deep learning algorithm combined with closed-loop control to achieve precise, dynamic, and intelligent regulation of coke oven gas collecting pipe pressure. This overcomes the technical shortcomings of traditional control methods, and the system possesses self-learning, self-adaptive, and fault self-diagnosis capabilities, making it adaptable to various coke oven production conditions while reducing energy consumption and improving production safety. The system and method of this invention can be directly applied to the technical transformation of existing coke ovens or integrated into the control systems of newly built coke ovens, demonstrating excellent industrial applicability and promotional value. This invention maturely applies artificial intelligence machine learning technology to the field of coke oven gas collecting pipe pressure control, filling the industry gap in the application of AI intelligent control in this scenario, while simultaneously considering production safety, energy economy, and system scalability.

Claims

1. A coke oven gas collecting pipe pressure control system based on artificial intelligence, characterized in that: It includes a pressure sensor (1), a data acquisition module (2), an AI processing module (3) and an actuator (4) connected in sequence, and also includes a cloud platform (6) that is communicatively connected to the AI ​​processing module (3). The pressure sensor (1) is installed at a key monitoring point in the coke oven gas collecting pipe; The AI ​​processing module (3) has a built-in algorithm model of an improved LSTM neural network with an attention mechanism; The actuator (4) is an electric regulating valve or a variable frequency fan with closed-loop feedback. The electric regulating valve is installed on the gas output pipeline of the coke oven gas collecting pipe, and the variable frequency fan is installed in the coke oven gas blower unit or the negative pressure regulating pipeline of the coke oven gas collecting pipe.

2. The coke oven gas collecting pipe pressure control system based on artificial intelligence according to claim 1, characterized in that: The actuator (4) has a built-in pressure feedback sensor, which is connected to the AI ​​processing module (3).

3. A coke oven gas collecting pipe pressure control system based on artificial intelligence according to claim 1 or 2, characterized in that: The data acquisition module (2) includes a signal amplifier, a filter and an analog-to-digital converter connected in sequence.

4. The coke oven gas collecting pipe pressure control system based on artificial intelligence according to claim 3, characterized in that: The AI ​​processing module (3) is embedded in an industrial-grade PLC or an edge computing industrial control computer.

5. The coke oven gas collecting pipe pressure control system based on artificial intelligence according to claim 4, characterized in that: The AI ​​processing module (3) establishes bidirectional communication with the cloud platform (6).

6. A method for controlling the pressure of coke oven gas collecting pipes based on artificial intelligence, characterized in that: The coke oven gas collecting pipe pressure control system based on artificial intelligence according to any one of claims 2-5 includes the following steps: S1. Pressure data acquisition: The pressure analog signal in the gas collection pipe is collected in real time by the pressure sensor (1) installed at the key point of the gas collection pipe and transmitted to the data acquisition module (2). S2. Data preprocessing: The data acquisition module (2) converts the amplified and noise-reduced pressure analog signal into standardized digital pressure data. After data verification and removal of invalid data, the data is transmitted to the AI ​​processing module (3). S3, AI model analysis and trend prediction: The AI ​​processing module (3) cleans and normalizes the received digital stress data, removes outliers and missing values, and inputs the preprocessed data into the algorithm model of the improved LSTM neural network with attention mechanism. The algorithm model assigns weights to the key features of the stress time series data through the attention mechanism, and then the LSTM neural network completes the stress change trend prediction at different time gradients, and outputs the stress deviation prediction value and trend change rate. S4. Control signal generation: The AI ​​processing module (3) generates a control signal that matches the current working condition based on the predicted pressure deviation value, the trend change rate, and the set threshold of the coke oven gas collecting pipe pressure. S5. Pressure closed-loop regulation: After receiving the control signal, the actuator (4) adjusts the opening of the electric regulating valve or the speed of the variable frequency fan. At the same time, it collects the actual pressure data after adjustment through the built-in pressure feedback sensor and transmits it back to the AI ​​processing module (3) in real time. The AI ​​processing module (3) compares the actual pressure data with the set threshold. If there is a deviation, it corrects the control signal in time to realize the closed-loop regulation of pressure. S6. Data storage and model iteration optimization: The AI ​​processing module (3) uploads real-time pressure data, control signals and execution feedback data to the cloud platform (6) for storage; the cloud platform (6) performs remote iteration training and fine-tuning of the algorithm model based on massive historical data and newly added operating data, and sends the optimized algorithm model to the AI ​​processing module (3) to realize the model's self-learning and adaptive update, and adapt to different coke oven production conditions.

7. The coke oven gas collecting pipe pressure control method based on artificial intelligence according to claim 6, characterized in that: In step S3, the improved LSTM neural network algorithm model, which integrates the attention mechanism, strengthens the key features in the stress time series data through the attention mechanism, suppresses the interference of invalid and redundant data, and the LSTM neural network completes the multi-dimensional prediction of the stress change trend based on the weighted feature data. The prediction time dimension includes three gradients: 10s, 30s, and 60s.

8. The coke oven gas collecting pipe pressure control method based on artificial intelligence according to claim 6, characterized in that: In step S3, the normalization preprocessing uses the max-min normalization method to map the digital pressure data to the [0,1] interval, eliminating the influence of dimensions.

9. The coke oven gas collecting pipe pressure control method based on artificial intelligence according to claim 6, characterized in that: In step S3, outliers are removed using the 3σ criterion, which identifies and eliminates abnormal pressure data that deviates from the data mean by three times the standard deviation.

Images