Intelligent control method and system for high-compatibility charging machine of all-electric glass melting furnace

Abstract

The invention relates to the technical field of intelligent control, in particular to an intelligent control method and system for a high-compatibility charging machine of a glass all-electric melting furnace. The method comprises the following steps: carrying out data preprocessing on obtained original data; state estimation and feature extraction are carried out on the preprocessed data; carrying out distribution track planning based on the extracted features; performing parameter decision by using the control parameters after the neural network distribution track planning; performing control execution and feedback based on a decision result; and outputting a control instruction and a running state. Through multi-source data acquisition and deep analysis, reinforcement learning optimization and an intelligent decision algorithm are combined, and the system can accurately plan a material distribution track and dynamically adjust control parameters such as the spiral conveying rotating speed and the belt speed.

Description

Technical Field

[0001] The present invention relates to the technical field of intelligent control, and particularly to an intelligent control method and system for a highly compatible feeder of a fully electric glass melting furnace. Background Art

[0002] As an important basic material industry, the glass manufacturing industry is widely used in fields such as construction, electronics, and medicine. With the development of the industry, fully electric glass melting furnaces have gradually become the mainstream production equipment due to advantages such as low energy consumption, low pollution, and high product quality. However, in actual production, it faces many challenges, which is the technical background for the birth of this technical solution.

[0003] The problem of uneven melting caused by raw material and process differences: There are many types of raw materials for glass production. Raw materials such as quartz sand and soda ash from different origins have significant differences in composition and particle size, and changes in the proportion of cullet added will also affect the characteristics of the raw materials. At the same time, there are various types of glass, and the production process requirements for different products such as soda-lime-silica glass and borosilicate glass are different. These differences make it difficult to keep the melting speed of the material surface in the furnace consistent, and it is easy to have local overheating or slow melting, which not only reduces the quality of glass products but also affects the service life of the furnace. For example, uneven melting of the material surface may cause defects such as bubbles and streaks in the glass, making it impossible to meet the production requirements of high-end glass products.

[0004] The limitations of traditional feeding methods: Traditional feeders for fully electric glass melting furnaces mostly use feeding methods with fixed trajectories and parameters, and it is difficult to make dynamic adjustments according to the real-time operating state of the furnace. In the face of raw material changes and different production conditions, this method cannot achieve precise matching of the feeding amount and the melting speed, and it is easy to cause material accumulation or insufficient supply. In addition, traditional control methods lack a comprehensive perception and in-depth analysis of the furnace operating state, and cannot detect potential problems in time and optimize them, resulting in low production efficiency and serious energy waste.

[0005] The urgent need for intelligent production: With the development of Industry 4.0 and intelligent manufacturing, the glass manufacturing industry has put forward higher requirements for the automation and intelligent level of the production process. The traditional production mode has been difficult to meet the market's requirements for product quality, production efficiency, energy conservation, and environmental protection. Realizing intelligent control of the feeding process of a fully electric glass melting furnace can monitor the furnace operating state in real time, precisely adjust the feeding strategy, improve the stability and controllability of the production process, reduce the cost of manual intervention, and enhance the core competitiveness of enterprises. Therefore, developing a highly compatible feeder control technology that can adapt to various raw materials and processes and achieve precise intelligent control has become an urgent problem to be solved in the glass manufacturing industry. Summary of the Invention

[0006] To solve the above-mentioned problems, the present invention provides an intelligent control method and system for a highly compatible feeder of a fully electric glass melting furnace.

[0007] In a first aspect, an intelligent control method for a highly compatible feeder of a fully electric glass melting furnace provided by the present invention adopts the following technical solutions:

[0008] An intelligent control method for a highly compatible feeder of a fully electric glass melting furnace includes:

[0009] Obtain raw data, including the process parameters and operation data of the fully electric glass melting furnace;

[0010] Perform data preprocessing on the obtained raw data;

[0011] Perform state estimation and feature extraction on the preprocessed data;

[0012] Plan the cloth feeding trajectory based on the extracted features;

[0013] Make parameter decisions using the control parameters after neural network cloth feeding trajectory planning;

[0014] Perform control execution and feedback based on the decision result;

[0015] Output control instructions and operation status.

[0016] Further, the data preprocessing of the obtained raw data includes removing high-frequency noise in the raw data using a sliding window filter, fusing multi-source sensor data using a Kalman filter to improve data accuracy, detecting and removing outliers based on the 3σ criterion to ensure the reliability of the input data, and providing high-quality data for subsequent analysis.

[0017] Further, the state estimation and feature extraction of the preprocessed data include calculating the temperature gradient and identifying the hot spot distribution based on the preprocessed data, and then analyzing the temperature field characteristics; using the material level change rate and melting rate estimation model to model the dynamic material surface, calculating the cloth feeding uniformity and heat flux density as feature quantities through the modeling, and integrating the feature quantities with the raw data to construct a state vector reflecting the operation state of the furnace for depicting the operation state of the furnace.

[0018] Further, the cloth feeding trajectory planning based on the extracted features includes dynamically planning the optimal cloth feeding path according to the production scale and real-time state of the furnace. Among them, the D-type trajectory is adopted, and small-range uniform cloth feeding is realized through the sine scanning mode. The AB combined trajectory is selected for 3050 tons, and the double sine superposition enhances the cloth feeding coverage; the C-type Bezier curve trajectory is adopted for 50100 tons to flexibly adapt to large-size furnaces; for more than 100 tons, the E-type dual-machine cooperation is started, and the two feeders cooperate; the trajectory parameters are dynamically adjusted based on the real-time material surface state.

[0019] Furthermore, making parameter decisions using the control parameters after neural network cloth trajectory planning includes constructing an intelligent decision-making center based on a three-layer neural network. Among them, the sensor layer performs non-linear transformation on the original data through the activation function ReLU to extract key features; the time series layer uses the LSTM network to process the historical data sequence to capture the changing trends and patterns of temperature and material level over time; the decision-making layer fuses the outputs of the first two layers and process parameters, and outputs the probability distribution of control parameters through the Softmax function; then through the mapping formula, the neural network output is converted into actual control parameters, including the matching calculation of screw rotation speed and belt speed, fully considering the influence of raw material density, ensuring the coordination of screw conveying volume and belt transmission speed, and achieving uniform cloth feeding.

[0020] Furthermore, performing control execution and feedback based on the decision result includes converting digital decisions into physical actions and forming a closed-loop correction; during the driving of the actuator, the servo motor is used to accurately control the screw conveying speed to achieve quantitative feeding; the frequency converter adjusts the belt speed to ensure the material transmission efficiency; the stepper motor drives the feeding arm to extend and retract to control the cloth feeding position; based on the feedback correction, the actual state monitored by the sensor is compared with the target state in real time, and the deviation is calculated; among them, under normal working conditions, the PID controller is used to adjust the control parameters according to the proportional, integral, and differential links to quickly respond to the deviation; in complex working conditions, an adaptive parameter adjustment algorithm is enabled to dynamically optimize the PID parameters to ensure control stability, including timely adjusting the control strategy when the raw material characteristics change suddenly.

[0021] Furthermore, performing control execution and feedback based on the decision result also includes building a dual defense line to ensure system safety during the execution process, including hard limits and soft constraints. Among them, the hard limit restricts the stroke of the actuator through a mechanical switch, and the current overload protection cuts off the dangerous power supply; the soft constraint is based on the state prediction model to estimate the future state, and when approaching the safety boundary, the control parameters are adjusted in advance, and the risk assessment function quantifies the risk of control actions to guide the system to avoid danger.

[0022] Furthermore, performing control execution and feedback based on the decision result also includes taking the Markov decision process as the framework, designing a reward function to comprehensively consider the material surface uniformity, energy consumption efficiency, and safety status, encouraging optimization behaviors through positive rewards, avoiding bad operations through negative punishments, and updating the decision-making strategy through the policy gradient algorithm to gradually learn the global optimal control scheme.

[0023] Furthermore, outputting control instructions and operating status includes converting the finally determined control parameters into execution signals to drive each component of the feeder to achieve intelligent cloth feeding of the batch material on the surface of the kiln furnace, ensuring stable and efficient melting process; real-time feedback of kiln furnace operation indicators, including material surface uniformity, energy consumption efficiency, and equipment safety status data, providing visual monitoring information for operators.

[0024] In a second aspect, an intelligent control system for a highly compatible feeding machine of a glass all-electric melting furnace includes:

[0025] A data acquisition module, configured to acquire raw data, including process parameters and operating data of the glass all-electric melting furnace;

[0026] A preprocessing module, configured to perform data preprocessing on the acquired raw data;

[0027] A feature module, configured to perform state estimation and feature extraction on the preprocessed data;

[0028] A planning module, configured to plan the cloth feeding trajectory based on the extracted features;

[0029] A decision-making module, configured to make parameter decisions using the control parameters after the cloth feeding trajectory is planned by a neural network;

[0030] An execution module, configured to perform control execution and feedback based on the decision result;

[0031] An output module, configured to output control instructions and operating states.

[0032] In a third aspect, the present invention provides a computer-readable storage medium, in which multiple instructions are stored, and the instructions are adapted to be loaded and executed by a processor of a terminal device to perform the intelligent control method for a highly compatible feeding machine of a glass all-electric melting furnace.

[0033] In a fourth aspect, the present invention provides a terminal device, including a processor and a computer-readable storage medium. The processor is used to implement each instruction; the computer-readable storage medium is used to store multiple instructions, and the instructions are adapted to be loaded and executed by the processor to perform the intelligent control method for a highly compatible feeding machine of a glass all-electric melting furnace.

[0034] In summary, the present invention has the following beneficial technical effects:

[0035] 1. Through multi-source data acquisition and in-depth analysis, combined with reinforcement learning optimization and intelligent decision-making algorithms, the system can accurately plan the cloth feeding trajectory and dynamically adjust control parameters such as the rotation speed of the screw conveyor and the belt speed. This enables the feeding amount to be highly matched with the melting speed of the furnace, effectively solving the problem of uneven melting of the material surface caused by raw material and process differences. Compared with the traditional fixed feeding method, this solution can increase the melting efficiency by 20%-30%, reduce energy waste caused by material accumulation or insufficient supply, reduce the unit product energy consumption by 15%-20%, and significantly improve production efficiency and energy utilization rate.

[0036] 2. This solution constructs a complete safety constraint and adaptive mechanism, with dual guarantees of physical hard limits and soft constraints based on model prediction to prevent the equipment from exceeding the safe operating range; the adaptive algorithm can adjust the control strategy in a timely manner according to factors such as raw material characteristics changes and furnace aging. At the same time, precise batching control keeps the material surface uniform and stable, greatly reducing the generation of defects such as bubbles and streaks in glass products, increasing the product qualification rate to over 98%, meeting the production requirements of high-end glass products, and enhancing the product quality competitiveness of the enterprise.

[0037] 3. Advanced algorithms such as neural networks and reinforcement learning are used to achieve intelligent control, reducing the dependence on manual experience and lowering the cost of manual intervention. Through real-time monitoring and feedback, the system can automatically optimize control parameters, reduce equipment failures and downtime, and extend the service life of the furnace. In addition, the production report generation function provides comprehensive data analysis for the enterprise, helps optimize production management decisions, realizes refined management of the production process, comprehensively improves the production efficiency and market competitiveness of the enterprise, and brings significant economic and social benefits to the enterprise. BRIEF DESCRIPTION OF THE DRAWINGS

[0038] Figure 1 It is a schematic diagram of an intelligent control method for a highly compatible batching machine of a fully electric glass melting furnace according to Embodiment 1 of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0039] The present invention will be further described in detail below with reference to the accompanying drawings.

[0040] Embodiment 1

[0041] Referring to Figure 1 , an intelligent control method for a highly compatible batching machine of a fully electric glass melting furnace in this embodiment includes: obtaining original data, including the process parameters and operation data of the fully electric glass melting furnace;

[0042] Performing data preprocessing on the obtained original data;

[0043] Performing state estimation and feature extraction on the preprocessed data;

[0044] Performing batching trajectory planning based on the extracted features;

[0045] Making parameter decisions using the control parameters after batching trajectory planning by neural networks;

[0046] Performing control execution and feedback based on the decision results;

[0047] Outputting control instructions and operating states.

[0048] Specifically, it includes the following steps:

[0049] 1. Data input

[0050] Data input is the starting link of intelligent control, providing basic information for subsequent decision-making. This step requires obtaining two types of key data: basic process parameters and real-time operation data. Basic process parameters determine the production goals and basic operation characteristics of the kiln, such as the type of glass, raw material ratio, etc.; real-time operation data reflects the current working state of the kiln, which is collected in real time through various sensors, enabling the control system to perceive changes in temperature, material level, etc. inside the kiln and providing real-time basis for precise control.

[0051] Among them, (1) Setting of basic process parameters: Operators input parameters such as the type of glass (e.g., soda-lime-silica glass, borosilicate glass), raw material ratio (proportions of quartz sand, soda ash, feldspar, etc.), proportion of cullet added, electrode arrangement method, daily melting capacity (M), etc. through the human-machine interface or production management system. These parameters will serve as the basic settings for system operation.

[0052] (2) Temperature data acquisition: A K-type thermocouple array is arranged on the liquid surface of the kiln, distributed at a certain spatial interval, to collect temperature data T(x, y, z) at different positions on the liquid surface in real time and construct the temperature field distribution.

[0053] Material level data acquisition: Using ultrasonic or radar sensors, multiple measurement points are set on the four walls and the central area of the kiln to periodically measure the material layer height Li(t) and obtain the change of the material level over time.

[0054] Weight data acquisition: A strain gauge sensor is built-in at the idler roller of the feeding machine belt to monitor the cloth weight distribution W(t) in real time, which is used to judge the uniformity of cloth feeding.

[0055] Other data acquisition: At the same time, auxiliary parameters such as ambient temperature and humidity, electrode current / voltage, etc. are collected to provide more dimensional information for subsequent analysis.

[0056] (3) Data transmission: Using industrial Ethernet or Modbus protocol, the data collected by the sensors is transmitted to the data processing module of the control system in real time and stably to ensure the timeliness and accuracy of the data.

[0057] 2. Data acquisition and preprocessing

[0058] Since the original data collected by the sensors may be affected by environmental noise, signal interference, etc., there are outliers and fluctuations. Direct use will reduce the control accuracy and reliability. Therefore, in this step, through technologies such as filtering and noise reduction and outlier processing, the original data is cleaned and optimized, and effective data features are extracted to provide high-quality data support for subsequent state estimation and decision-making.

[0059] (1) Moving window filtering: For time series data, including temperature T(t)), set the window size N and calculate the mean value of the data within the window Smooth the data to remove high-frequency noise, making the data better reflect the true trend. The window size N can be dynamically adjusted according to the signal fluctuation characteristics, and the typical value is 5 - 10.

[0060] Kalman filtering: For multi-source sensor data (such as temperature and level data), use the Kalman filtering algorithm for fusion. First, perform the prediction step. According to the state estimate value at the previous moment and the control input uk, predict the state at the current moment through the state transition matrix Fk and the control matrix Bk and update the covariance matrix simultaneously Then, perform the update step. According to the observation value zk and the observation matrix Hk at the current moment, calculate the Kalman gain Kk, and then obtain a more accurate state estimate value and covariance matrix

[0061] (2) Outlier processing: Detect outliers in the data based on the 3σ criterion. Calculate the mean μ and standard deviation σ of the data. When the data x satisfies |x - μ| > 3σ, it is determined as an outlier. For the detected outliers, use the interpolation method or the neighborhood mean method for correction. In this embodiment,

[0062] where x i-1 , x i+1 are adjacent data points to ensure the reliability of the data.

[0063] 3. State Estimation and Feature Extraction

[0064] State estimation and feature extraction are based on the preprocessed data to deeply analyze the operating state of the kiln and extract key features that can reflect the dynamic characteristics of the melting process of the kiln. By processing information such as the temperature field and the change of the material surface, calculate feature quantities such as temperature gradient, hot spot distribution, material level change rate, and melting rate, and integrate these features with the original data into a state vector to provide comprehensive and accurate information for subsequent cloth trajectory planning and control parameter decision-making.

[0065] Among them, (1) Temperature field analysis:

[0066] Spatial gradient calculation: Use the finite difference method to approximately calculate the gradient of temperature in space By analyzing the temperature gradient, the temperature mutation region can be located, and the non-uniformity of the temperature distribution can be understood, providing a basis for judging the melting situation.

[0067] Hot spot identification: Set the threshold coefficient k (the empirical value is 23). According to the average temperature Tavg and the temperature standard deviation σT, use the formula

[0068]

[0069] Identify the hot spot area and determine whether there is local overheating.

[0070] (2) Dynamic modeling of the burden surface:

[0071] Calculation of the burden level change rate: Based on the burden level data at adjacent times, through the formula Reflect the change of the burden surface height over time and quantify the dynamic change trend of the burden surface.

[0072] Estimation of the melting rate: Combining the raw material density ρ (which can be obtained through on-line measurement or preset value), the burden surface area A, and the burden level change rates at each measurement point and the weights wi (assigned according to the importance of the sensor position), using the formula Estimate the melting rate Master the real-time melting capacity of the kiln.

[0073] (3) Construction of the comprehensive feature vector: Integrate the feature quantities such as the temperature gradient, hot spot distribution, burden level change rate, and melting rate obtained from the above calculations with the preprocessed original data (such as temperature, burden level, weight data, etc.) into a complete state vector

[0074]

[0075] Comprehensively describe the operating state of the kiln.

[0076] 4. Burden distribution trajectory planning

[0077] The burden distribution trajectory planning selects the appropriate type of burden distribution trajectory and determines the corresponding trajectory parameters according to factors such as the daily melting capacity of the kiln and the real-time burden surface state, so as to achieve uniform distribution of the batch material on the liquid surface of the kiln and make the feeding amount match the melting speed. Through reasonable trajectory planning, the problem of uneven melting of the burden surface can be improved, and the melting efficiency and product quality of the kiln can be enhanced.

[0078] Among them, (1) Trajectory type decision: Divide different trajectory type selection strategies according to the daily melting capacity M:

[0079] When M < 30 tons, select the D-type trajectory (sine scanning mode), which is suitable for the uniform burden distribution requirements of small melting capacity. If M < 20 tons, adopt the D1 + D2 combination; if 20 < M < 30 tons, adopt the D1 + D3 combination.

[0080] When 30 ≤ M ≤ 50 tons, select the AB-type trajectory (double sine superposition mode). According to the judgment of the burden surface unevenness, if the burden surface unevenness exceeds the set threshold, adopt the A + B combination; otherwise, adopt the A + A combination to meet the more complex burden distribution requirements under medium melting capacity.

[0081] When 50 < M ≤ 100 tons, select the C-type trajectory (Bézier curve mode). By flexibly adjusting the control points of the Bézier curve, precise material distribution control under a large melting capacity can be achieved. Combinations of A+A, A+B, or B+B can be adopted.

[0082] When M > 100 tons, select the E-type trajectory (dual-machine collaborative mode), and complete the material distribution task through the collaborative work of two feeding machines.

[0083] (2) Trajectory parameter calculation:

[0084] For the D-type trajectory (sine scan): Use the formula

[0085] to calculate the trajectory coordinates, where the amplitude A = 0.5 + 0.01M, dynamically adjusting the lateral material distribution range according to the melting capacity; the angular velocity omega is adjusted according to the material distribution density to control the scanning frequency; the phase phi is used to adjust the starting position; and the longitudinal velocity v ensures the continuity of material distribution.

[0086] For the C-type trajectory (Bézier curve): Use the formula

[0087] to generate the trajectory. By dynamically generating the control points Pi according to the kiln size and the state of the material surface, the shape of the material distribution trajectory can be flexibly designed to adapt to different kiln shapes and material distribution requirements.

[0088] 5. Control parameter decision-making

[0089] For control parameter decision-making, a three-layer fusion neural network model is used. Combining various information such as historical data, current state, and process parameters, optimal control parameters such as the spiral conveyor rotation speed, belt speed, feeding arm telescopic speed, and trajectory change position are generated. This model is trained with a large amount of data to learn the mapping relationship between various factors and control parameters during the operation of the kiln, so as to accurately predict appropriate control parameters according to real-time input and achieve precise control of the feeding process.

[0090] Among them, (1) Neural network architecture design:

[0091] Sensor layer: Take the state vector mathbfx as the input, and perform a non-linear transformation through the activation function (such as the ReLU function) σ, weight matrix W1, and bias vector b1 to extract the data feature f sensor = σ(W1x + b1).

[0092] Time series layer: Process the historical state sequence H, and use the long short-term memory network (LSTM) and parameter W2 to capture the time series feature f time = LSTM(H; W2), learning the variation law of the state over time.

[0093] Decision-making layer: Fuse the features \(f_{sensor}\) output by the sensor layer, the features \(f_{time}\) output by the timing layer, and the process parameter vector \(\mathbf{p}\), and output the control parameter vector through the Softmax function, weight matrix \(W_3\), and bias vector \(b_3\).

[0094] \(U = softmax(W_3[f sensor ; f time ; p]+b_3)\),

[0095] \(U = [u screw , u belt , u arm , u track .

[0096] (2) Parameter mapping: Convert the normalized parameters output by the neural network into actual control parameters:

[0097] Screw rotation speed: \(n = n max \cdot u screw , where \(n_{max}\) is the maximum value of the screw rotation speed.

[0098] Belt speed: Considering the influence of the raw material density \(\rho\), \(k\) is a coefficient.

[0099] Arm telescopic rate: \(s = s max \cdot u arm , \(s_{max}\) is the maximum value of the feeding arm telescopic speed.

[0100] Track-changing position: \(p = p max \cdot u trac _k\), \(p_{max}\) is the maximum value of the track-changing position.

[0101] 6. Control execution and feedback

[0102] Control execution and feedback is to convert the control parameters obtained by decision-making into actual control instructions, drive the actuator of the feeder to act, and monitor the execution effect in real time through sensors, compare it with the target state, adjust the control parameters according to the deviation, form a closed-loop control, ensure that the actual operation state is consistent with the target state, and achieve stable control of the kiln feeding process.

[0103] Among them, (1) Control instruction output and actuator drive:

[0104] Servo motor (screw conveyor): Convert the calculated screw rotation speed control parameter into a pulse signal, drive the servo motor, and use its high-precision position control ability (accuracy ±0.01°) to accurately adjust the rotation speed of the screw conveyor to achieve quantitative feeding.

[0105] Variable Frequency Drive (Belt Drive): According to the belt speed control parameters, by adjusting the output frequency of the variable frequency drive (range 20 - 80Hz), the belt speed is controlled to vary within the range of 0.1 - 5m / s to ensure the material transfer efficiency.

[0106] Stepper Motor (Feeding Arm): Convert the telescopic speed and orbit change position control parameters of the feeding arm into pulse signals to drive the stepper motor. Utilize the micro - stepping drive technology to achieve smooth movement, with a repeat positioning accuracy of ±0.1mm, accurately controlling the telescopic and orbit change movements of the feeding arm.

[0107] (2) Feedback Correction:

[0108] Deviation Calculation: The sensor continuously collects the operation state data of the kiln and compares it with the target state to calculate the deviation e(t).

[0109] PID Control: According to the deviation e(t), use the PID control algorithm to calculate the control quantity Among them, the proportional coefficient Kp quickly responds to the deviation, the integral coefficient Ki eliminates the steady - state error, and the derivative coefficient Kd suppresses the system overshoot. Through the adaptive formula Dynamically adjust the PID parameters, where J is the performance index, including the material surface uniformity error.

[0110] Model Predictive Control (MPC): Under complex working conditions, based on the state - space model, predict the state in the next N steps, and through the rolling optimization control sequence, adjust the control parameters in advance to make the system better adapt to changes.

[0111] 7. Safety Constraints and Adaptability

[0112] The safety constraints and adaptability link prevents the feeder and the kiln from exceeding the safe operating range by setting physical hard limits and model - prediction - based soft constraints, avoiding equipment damage and production accidents. At the same time, use the adaptive algorithm to adjust the control strategy and model parameters according to various state changes during the operation of the kiln, enabling the control system to adapt to the impacts brought by factors such as raw material characteristics changes and kiln aging, and maintaining good control performance.

[0113] Among them, (1) Safety Mechanism:

[0114] Hard Limit: Install mechanical limit switches at key parts of the feeder. When the actuator (such as the feeding arm) reaches the maximum stroke, the power supply is automatically cut off to limit its continued movement; set current overload protection. When the motor current I > Imax, immediately cut off the power supply to protect the motor and equipment safety.

[0115] Soft Constraint: Based on the state prediction model, through the formula

[0116] Predict the state at the next moment, where f(x_t, u_t) is the state prediction function, and Δx safe is the safety margin. When the predicted state approaches or exceeds the safety boundary, adjust the control parameters in advance. Design a risk assessment function to evaluate the risk level of the current control action, where is the probability function, is the mean of the safe state, and δ is the safety threshold. Optimize the control strategy according to the risk assessment results.

[0117] (2) Adaptive algorithm:

[0118] Parameter update: Use the forgetting factor recursive least squares method to update the model parameter theta online. The formula is theta_t = theta_{t - 1}+K_t(y_t - h(x_t, theta_{t - 1})), so that the model can adapt to the changes in the kiln operation state in a timely manner.

[0119] Raw material property identification: By analyzing data such as the cloth weight W and the change in material level, use the formula to estimate the raw material density rho; establish a fluidity index model to obtain the functional relationship through fitting of experimental data, which is used to characterize the flow characteristics of the raw materials and provide a basis for adjusting the control parameters.

[0120] 8. Reinforcement learning optimization

[0121] Principle content

[0122] Reinforcement learning optimization transforms the kiln feeding control problem into a sequential decision-making problem by constructing a Markov decision process (MDP). Design a reasonable reward function to give corresponding rewards or punishments according to the kiln operation state and control actions, and guide the control system to learn the optimal control strategy in the long-term operation. By continuously interacting with the environment and updating the policy parameters, the control system can achieve the control objectives of uniform material surface, lowest energy consumption, and safe and stable operation under various working conditions.

[0123] Algorithm and steps

[0124] 1. MDP modeling:

[0125] State space: Define the state space S = {x, trajectory parameters, process parameters}, which includes the state vector mathbfx, the cloth trajectory parameters, and the process parameters, and comprehensively describes the operation state of the kiln.

[0126] Action space: The action space A = {U}, that is, the control parameter vector, including control actions such as the screw conveyor rotation speed, the belt speed, the telescopic speed of the feeding arm, and the track change position.

[0127] Transition probability: Assume that the state transition follows a Gaussian distribution

[0128] Among them, f(s,a) is the state transition function, and σ is the covariance matrix, which describes the uncertainty of state transition.

[0129] 2. Design of reward function: Design the reward function

[0130] r1 = w1r uniform + w2r energy + w3r safety , and the achievement of different control objectives is comprehensively considered through weighted summation. Among them, w1, w2, and w3 are weight coefficients, which can be dynamically adjusted according to production requirements and management strategies to highlight the importance of different objectives.

[0131] Uniformity reward: Among them, σL is the standard deviation of the material level, which reflects the degree of fluctuation of the material surface height; is the average material level. This formula encourages the control system to achieve more uniform material distribution by minimizing the ratio of the standard deviation of the material level to the average material level, and avoids the influence of too high or too low local material surface on the melting effect.

[0132] Energy consumption reward: Among them, P is the energy consumption, is the melting rate. This reward mechanism encourages the system to minimize energy consumption, improve energy utilization efficiency, and reduce production costs on the premise of ensuring a certain melting rate.

[0133] Safety reward: Among them, S safe is the set of safety states. A positive reward is given when the operating state of the kiln is within the safe range, and a large negative reward is given otherwise, ensuring that the system always follows the safety constraint conditions during operation and preventing equipment damage and production accidents.

[0134] 3. Policy optimization: The Proximal Policy Optimization (PPO) algorithm is used for policy update. This algorithm improves the learning efficiency while ensuring the stability of policy update through importance sampling technology.

[0135] Calculate the advantage function used to evaluate the advantage degree of taking action a under the current policy relative to the average value. The formula is

[0136]

[0137] , where γ is the discount factor, which is used to balance the importance of future rewards and current rewards; rt is the reward at time t; is the value function of state s under policy πθ.

[0138] According to the advantage function and policy probability πθ(a|s), calculate the policy gradient This gradient indicates the update direction of the policy parameter theta.

[0139] Update the policy parameter

[0140] where alpha is the learning rate, which controls the step size of each parameter update. Through multiple iterative updates, the policy pi(theta) gradually converges to the optimal policy to adapt to the feeding control requirements under different working conditions.

[0141] 9. Data Output

[0142] Data output is the terminal link of the intelligent control process, undertaking the important task of transmitting control results and operation information to relevant devices and personnel. On the one hand, it outputs the final control parameter instructions to drive the feeder to operate in the planned manner to achieve precise batching; on the other hand, it outputs the evaluation data of the kiln operation status, providing real-time monitoring information for operators and decision-making basis for production management, helping to optimize the production process and improve production efficiency.

[0143] Among them, (1) Control instruction output: The control parameter vector U = [u screw , u belt , u arm , u track determined through reinforcement learning optimization and control parameter decision-making is converted into physical signals that can be received by each actuator of the feeder through a signal conversion module.

[0144] For the screw rotation speed control parameter uscrew, according to the mapping relationship n = n max ·u screw the actual rotation speed n is calculated, and then the rotation speed instruction is converted into the pulse signal frequency and quantity that can be recognized by the servo motor to accurately control the speed and quantity of the screw conveyor to convey materials.

[0145] For the belt speed control parameter ubelt, according to the formula the actual running speed vbelt of the belt is calculated and converted into the frequency control signal of the frequency converter to adjust the speed of the belt to convey materials.

[0146] For the control parameters uarm and utrack of the telescopic speed and track-changing position of the feeding arm, according to s = s max ·u arm and p = p max ·u track the actual telescopic length and track-changing position are calculated, and converted into the pulse signal of the stepping motor to achieve the accurate telescopic and track-changing movement of the feeding arm, ensuring that the batch material is evenly distributed on the liquid surface of the kiln according to the planned batching trajectory.

[0147] (2) Operation status output:

[0148] Calculation and Display of Key Indicators: Calculate and output key indicators such as the uniformity U of the burden surface, the energy consumption efficiency eta, and the safety risk level in real time. The uniformity U of the burden surface is calculated by analyzing the burden level data and reflects the flatness of the burden surface; the energy consumption efficiency eta is calculated based on the energy consumption P and the melting rate dotm to measure the effectiveness of energy utilization; the safety risk level is evaluated based on safety constraints and the real-time operating status to provide intuitive safety warnings for operators. These indicators are displayed in the form of numbers, charts, etc. on the monitoring interface to facilitate operators to grasp the operating conditions of the furnace in real time.

[0149] Generation of Visualization Data: Use the collected data such as temperature and burden level to generate visualization content such as a temperature cloud map, a burden surface height curve, and a real-time animation of the charging trajectory. The temperature cloud map intuitively presents the temperature distribution of the furnace liquid surface, helping operators quickly locate high-temperature or low-temperature areas; the burden surface height curve shows the change trend of the burden level over time, facilitating the analysis of the melting speed and the feeding effect; the real-time animation of the charging trajectory dynamically displays the operating trajectory of the feeder and the charging process, enabling operators to clearly understand the uniformity and accuracy of charging. These visualization data are displayed through an industrial monitoring system or a remote terminal, providing comprehensive and intuitive operating information for operators.

[0150] Generation of Production Reports: Automatically collect and organize various types of data in the production process, including control parameters, operating status, product quality, etc., according to the set time periods (shifts, days, months), and generate detailed production data reports. The report content covers aspects such as production output, raw material consumption, energy consumption statistics, equipment operating time, and fault records. Through data analysis and statistics, it provides decision-making support for production management, such as optimizing production plans, adjusting process parameters, and arranging equipment maintenance, helping to improve production management levels and production efficiency.

[0151] Embodiment 2

[0152] This embodiment provides an intelligent control system for a highly compatible feeder of a fully electric glass melting furnace, including:

[0153] A data acquisition module, configured to

[0154] A computer-readable storage medium, in which multiple instructions are stored, and the instructions are adapted to be loaded and executed by a processor of a terminal device for the intelligent control method of a highly compatible feeder of a fully electric glass melting furnace.

[0155] A terminal device, including a processor and a computer-readable storage medium, where the processor is used to implement each instruction; the computer-readable storage medium is used to store multiple instructions, and the instructions are adapted to be loaded and executed by the processor for the intelligent control method of a highly compatible feeder of a fully electric glass melting furnace.

[0156] The above are all preferred embodiments of the present invention, and the protection scope of the present invention is not limited thereby. Therefore, all equivalent changes made according to the structure, shape, and principle of the present invention shall be covered within the protection scope of the present invention.

Claims

1. An intelligent control method for a high-compatibility feeding machine of a fully electric glass melting furnace, characterized in that Including: Obtain original data, including the process parameters and operation data of the all-electric glass melting furnace; Perform data preprocessing on the obtained original data; Perform state estimation and feature extraction on the preprocessed data; Plan the cloth feeding trajectory based on the extracted features; Make parameter decisions using the control parameters after cloth feeding trajectory planning by neural network; Perform control execution and feedback based on the decision result; Output control instructions and operation status.

2. The intelligent control method of a highly compatible feeding machine for a fully electric glass melting furnace according to claim 1, characterized in that, The data preprocessing of the obtained original data includes using sliding window filtering to remove high-frequency noise in the original data, adopting Kalman filtering to fuse multi-source sensor data to improve data accuracy, detecting and removing outliers based on the 3σ criterion to ensure the reliability of the input data, and providing high-quality data for subsequent analysis.

3. An intelligent control method for a highly compatible feeding machine of a fully electric glass melting furnace according to claim 2, characterized in that, The state estimation and feature extraction of the preprocessed data include calculating the temperature gradient and identifying the hot spot distribution based on the preprocessed data, and then analyzing the temperature field characteristics; Use the model for estimating the change rate of the material level and the melting rate to model the dynamic material surface, calculate the cloth feeding uniformity and heat flux density as feature quantities through modeling, integrate the feature quantities with the original data, and construct a state vector reflecting the operation status of the furnace to describe the operation status of the furnace.

4. The intelligent control method of a highly compatible feeding machine for a fully electric glass melting furnace according to claim 3, wherein, The cloth feeding trajectory planning based on the extracted features includes dynamically planning the optimal cloth feeding path according to the production scale and real-time status of the furnace. Among them, the D-type trajectory is adopted, and small-range uniform cloth feeding is realized through the sine scanning mode. The AB combination trajectory is selected for 3050 tons, and the double sine superposition enhances the cloth feeding coverage; the C-type Bezier curve trajectory is adopted for 50-100 tons to flexibly adapt to large-size furnaces; for more than 100 tons, the E-type dual-machine cooperation is started, and the two feeding machines cooperate; the trajectory parameters are dynamically adjusted based on the real-time material surface status.

5. An intelligent control method for a highly compatible feeding machine of a fully electric glass melting furnace according to claim 4, characterized in that, The parameter decision using the control parameters after cloth feeding trajectory planning by neural network includes constructing an intelligent decision-making center based on a three-layer neural network. Among them, the sensor layer performs nonlinear transformation on the original data through the activation function ReLU to extract key features; the time series layer uses the LSTM network to process the historical data sequence to capture the changing trends and laws of temperature and material level over time; the decision layer fuses the outputs of the previous two layers and the process parameters, and outputs the probability distribution of the control parameters through the Softmax function; then through the mapping formula, the neural network output is converted into the actual control parameters, including the matching calculation of the screw rotation speed and the belt speed, fully considering the influence of the raw material density, ensuring the coordination of the screw conveying volume and the belt transmission speed, and realizing uniform cloth feeding.

6. The intelligent control method of a highly compatible feeding machine for a fully electric glass melting furnace according to claim 5, characterized in that, The control execution and feedback based on the decision result include converting digital decisions into physical actions and forming a closed-loop correction. During the driving of the actuator, a servo motor is used to precisely control the rotation speed of the screw conveyor to achieve quantitative feeding. A frequency converter adjusts the belt speed to ensure the material transmission efficiency. A stepper motor drives the telescopic movement of the feeding arm to control the cloth placement position. Based on the feedback correction, the actual state monitored by the sensor is compared with the target state in real time, and the deviation is calculated. Among them, under normal working conditions, a PID controller is used to adjust the control parameters according to the proportional, integral, and differential links to quickly respond to the deviation. In complex working conditions, an adaptive parameter adjustment algorithm is enabled to dynamically optimize the PID parameters to ensure stable control, including timely adjusting the control strategy when the raw material characteristics change suddenly.

7. An intelligent control method for a highly compatible feeding machine of a fully electric glass melting furnace according to claim 6, characterized in that, The control execution and feedback based on the decision result also include building a dual defense system to ensure system safety during the execution process, including hard limits and soft constraints. Among them, the hard limit restricts the stroke of the actuator through a mechanical switch, and the current overload protection cuts off the dangerous power supply. The soft constraint is based on a state prediction model to estimate the future state. When approaching the safety boundary, the control parameters are adjusted in advance, and the risk assessment function quantifies the risk of control actions to guide the system to avoid danger.

8. An intelligent control method for a highly compatible feeding machine of a fully electric glass melting furnace according to claim 7, characterized in that, The control execution and feedback based on the decision result also include using the Markov decision process as a framework to design a reward function that comprehensively considers the material surface uniformity, energy consumption efficiency, and safety status. Positive rewards are used to encourage optimization behaviors, and negative punishments are used to avoid bad operations. The decision-making strategy is updated through the policy gradient algorithm to gradually learn the global optimal control scheme.

9. An intelligent control method for a highly compatible feeding machine of a fully electric glass melting furnace according to claim 8, characterized in that, The output of the control instruction and the operating state includes converting the finally determined control parameters into execution signals for driving each component of the feeder to achieve intelligent cloth placement of the batch material on the surface of the furnace, ensuring stable and efficient melting. Real-time feedback of the furnace operation indicators, including the material surface uniformity, energy consumption efficiency, and equipment safety status data, provides visual monitoring information for the operator.

10. An intelligent control system for a highly compatible feeding machine of a fully electrically melted glass furnace, characterized in that, It includes: A data acquisition module configured to acquire raw data, including the process parameters and operating data of the all-electric glass melting furnace. A preprocessing module configured to perform data preprocessing on the acquired raw data. A feature module configured to perform state estimation and feature extraction on the preprocessed data. A planning module configured to perform cloth placement trajectory planning based on the extracted features. A decision-making module configured to make parameter decisions using the control parameters after the neural network cloth placement trajectory planning. An execution module configured to perform control execution and feedback based on the decision result. An output module configured to output control instructions and operating states.

Images