Vertical furnace welding parameter monitoring and adjusting method and device and computer equipment

Abstract

The invention discloses a vertical furnace welding parameter monitoring and adjusting method and device and computer equipment. The method comprises the steps that relevant data detected by a sensor and technological parameters of welding are obtained to obtain initial data; inputting the initial data into an LSTM prediction network for parameter prediction and determination of a confidence interval to obtain a prediction result; when the number of errors predicted by the LSTM prediction network does not exceed a set threshold value, processing the initial data and a prediction result by adopting a DDPG algorithm to obtain a decision result; and adjusting the parameters of the welding equipment according to the decision result. By implementing the method provided by the invention, accurate temperature and oxygen content prediction can be realized, energy consumption is optimized, the stability and energy efficiency of the system are improved, and meanwhile, the method has online learning and self-optimization capabilities.

Description

Technical Field

[0001] The present invention relates to a method for monitoring and adjusting parameters, and more specifically to a method, device, and computer equipment for monitoring and adjusting welding parameters of a vertical furnace. Background Art

[0002] During the vertical furnace welding process, temperature, oxygen content, and nitrogen content are key factors affecting welding quality. Since the oxygen content and nitrogen content during the welding process will continuously fluctuate with the change of temperature, real-time monitoring must be carried out to ensure that these parameters are always within the appropriate range. If the gas content is too high or the temperature is too high, a series of problems may occur, including temperature overshoot, oxidation reaction, etc., thus affecting the stability of the process and the quality of the product.

[0003] LSTM (Long Short-Term Memory) has important potential in real-time monitoring, especially in the welding process. By utilizing its powerful time series prediction ability, LSTM can effectively track and predict the change trends of key parameters such as oxygen content, nitrogen content, and temperature, thereby identifying potential abnormal situations and making adjustments in advance to ensure the stability of the welding process and improve the product quality. However, some challenges faced by the current technology include data quality issues, model training complexity, real-time requirements, and the difficulty of multi-parameter collaborative prediction. In addition, the lack of unified implementation standards makes the application effects of LSTM in different welding equipment and processes vary greatly, which puts higher requirements on the popularization and optimization of the technology.

[0004] Therefore, it is necessary to design a new method to achieve accurate temperature and oxygen content prediction, optimize energy consumption, improve system stability and energy efficiency, and at the same time have the ability of online learning and self-optimization. Summary of the Invention

[0005] The purpose of the present invention is to overcome the defects of the prior art and provide a method, device, and computer equipment for monitoring and adjusting welding parameters of a vertical furnace.

[0006] To achieve the above purpose, the present invention adopts the following technical solutions: A method for monitoring and adjusting welding parameters of a vertical furnace, including:

[0007] Obtain relevant data detected by sensors and welding process parameters to obtain initial data;

[0008] Input the initial data into the LSTM prediction network for parameter prediction and determination of the confidence interval to obtain a prediction result;

[0009] When the number of error times predicted by the LSTM prediction network does not exceed the set threshold, the initial data and the prediction result are processed by the DDPG algorithm to obtain a decision result;

[0010] Adjust the parameters of the welding equipment according to the decision result.

[0011] A further technical solution thereof is: after adjusting the parameters of the welding equipment according to the decision result, it further includes:

[0012] Obtain the relevant data detected by the sensor again;

[0013] Retrain the LSTM prediction network and the DDPG algorithm according to the relevant data obtained again.

[0014] A further technical solution thereof is: the LSTM prediction network is trained based on historical data.

[0015] A further technical solution thereof is: the prediction result includes temperature change, oxygen content trend and corresponding confidence interval.

[0016] A further technical solution thereof is: the processing the initial data and the prediction result by the DDPG algorithm to obtain a decision result includes:

[0017] Combine the relevant data detected by the sensor in the initial data with the prediction result, and combine the process parameters as the state space;

[0018] Use the DDPG algorithm to output an action space according to the state space to obtain a decision result; wherein, the reward function of the DDPG algorithm calculates the reward through the comprehensive performance of total energy consumption, temperature stability and oxygen content control.

[0019] A further technical solution thereof is: the decision result includes the adjustment amounts of heating power, nitrogen valve opening and fan speed.

[0020] A further technical solution thereof is: after inputting the initial data into the LSTM prediction network for parameter prediction and confidence interval determination to obtain a prediction result, it further includes:

[0021] When the number of error times predicted by the LSTM prediction network exceeds the set threshold, switch the control mode of the welding equipment to the traditional PID control mode.

[0022] A further technical solution thereof is: the decision result is physically limited.

[0023] The present invention also provides a vertical furnace welding parameter monitoring and adjustment device, including:

[0024] A data acquisition unit for acquiring relevant data detected by sensors and welding process parameters to obtain initial data;

[0025] A prediction unit for inputting the initial data into an LSTM prediction network for parameter prediction and determination of the confidence interval to obtain a prediction result;

[0026] A decision-making unit for, when the number of error times predicted by the LSTM prediction network does not exceed a set threshold, processing the initial data and the prediction result using the DDPG algorithm to obtain a decision result;

[0027] An adjustment unit for adjusting the parameters of the welding equipment according to the decision result.

[0028] The present invention also provides a computer device, which includes a memory and a processor. A computer program is stored on the memory, and when the processor executes the computer program, the above method is implemented.

[0029] The beneficial effects of the present invention compared with the prior art are as follows: By combining the LSTM prediction network and the DDPG algorithm, the present invention realizes accurate prediction of temperature and oxygen content, and on this basis, optimizes energy consumption, improves system stability and energy efficiency. First, welding parameters and process data are obtained through sensors and input into the LSTM model for prediction of temperature and oxygen concentration, and the prediction confidence interval is determined. When the prediction error remains within the set threshold, the DDPG algorithm is used for decision-making in combination with the initial data and the prediction result, and the parameters of the welding equipment are adjusted in real time to ensure process accuracy. In this way, not only can the changes in temperature and oxygen concentration be accurately predicted, but also dynamic adjustment can be made according to real-time data, thereby reducing energy consumption, improving the stability and energy efficiency of the system, and at the same time having the ability of online learning and self-optimization, and gradually optimizing the control strategy.

[0030] The following further describes the present invention with reference to the drawings and specific embodiments. Description of the Drawings

[0031] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings required for description in the embodiments will be briefly introduced below. Obviously, the drawings in the following description are some embodiments of the present invention. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative efforts.

[0032] Figure 1 It is a flowchart of the vertical furnace welding parameter monitoring and adjustment method provided by the embodiment of the present invention;

[0033] Figure 2Schematic diagram of the sub - process of the vertical furnace welding parameter monitoring and adjustment method provided by an embodiment of the present invention;

[0034] Figure 3 Schematic diagram of the process of the vertical furnace welding parameter monitoring and adjustment method provided by another embodiment of the present invention;

[0035] Figure 4 Schematic block diagram of the vertical furnace welding parameter monitoring and adjustment device provided by an embodiment of the present invention;

[0036] Figure 5 Schematic block diagram of the decision - making unit of the vertical furnace welding parameter monitoring and adjustment device provided by an embodiment of the present invention;

[0037] Figure 6 Schematic block diagram of the vertical furnace welding parameter monitoring and adjustment device provided by another embodiment of the present invention;

[0038] Figure 7 Schematic block diagram of the computer device provided by an embodiment of the present invention. Detailed implementation manners

[0039] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present invention.

[0040] It should be understood that when used in this specification and the appended claims, the terms "comprises" and "comprising" indicate the presence of the described features, wholes, steps, operations, elements, and / or components, but do not preclude the presence or addition of one or more other features, wholes, steps, operations, elements, components, and / or their combinations.

[0041] It should also be understood that the terms used in this specification of the present invention are only for the purpose of describing specific embodiments and are not intended to limit the present invention. As used in this specification of the present invention and the appended claims, unless the context clearly indicates otherwise, the singular forms "a", "an", and "the" are intended to include the plural forms.

[0042] It should be further understood that the term "and / or" used in this specification of the present invention and the appended claims refers to any combination and all possible combinations of one or more of the associated listed items, and includes these combinations.

[0043] Please refer to Figure 1 , Figure 1Schematic flowchart of the vertical furnace welding parameter monitoring and adjustment method provided by the embodiments of the present invention. The vertical furnace welding parameter monitoring and adjustment method is applied to a server. The server exchanges data with sensors and welding equipment. By combining the LSTM and DDPG (Deep Deterministic Policy Gradient) algorithms, accurate prediction and real-time adjustment of the temperature, oxygen content, and nitrogen content during the vertical furnace welding process are achieved to improve system stability, energy efficiency, and welding quality. During the welding process, the LSTM model can predict the change trend of key parameters through time series analysis and timely adjust the working state of the welding equipment. The DDPG algorithm, based on the results predicted by the LSTM, comprehensively considers factors such as total energy consumption, temperature stability, and oxygen content control, and optimizes energy consumption by dynamically adjusting the heating power, nitrogen valve opening, and fan speed, etc. By continuously obtaining sensor data, the system has the ability of online learning and self-optimization, can self-adjust during the welding process and retrain the model according to new data to ensure that all parameters are always in the best state, thus solving the influence of temperature fluctuations and gas content changes on welding quality and improving the stability and energy efficiency of the welding process.

[0044] Figure 1 It is a schematic flowchart of the vertical furnace welding parameter monitoring and adjustment method provided by the embodiments of the present invention. As Figure 1 shown, the method includes the following steps S110 to S140.

[0045] S110. Obtain relevant data detected by sensors and welding process parameters to obtain initial data.

[0046] In this embodiment, the initial data refers to the relevant data detected by sensors and the process parameters.

[0047] In this step, the system first obtains real-time data from multiple sensors (such as pressure sensors, oxygen sensors, temperature sensors, etc.). These sensors monitor multiple parameters in the furnace, such as pressure, oxygen content, temperature, nitrogen flow rate, etc., and form initial data in combination with process parameters (such as target temperature, maximum oxygen content, etc.). The sensor data includes information such as furnace temperature, gas composition, pressure, etc., and the process parameters refer to the set requirements that need to be followed during the welding process. These initial data will serve as the basis for the entire control and adjustment process.

[0048] Specifically, the system obtains key data of the furnace environment in real time through multiple sensors. These sensors include but are not limited to the following types:

[0049] Pressure sensor: Monitors the gas pressure inside the furnace. Pressure is one of the important factors affecting the stability of the welding process. Excessive or insufficient pressure will affect the welding effect. Therefore, the system needs to monitor the gas pressure in the furnace in real time and use it as a key input.

[0050] Oxygen sensor: Monitors the oxygen content inside the furnace. The control of oxygen content is crucial because too high an oxygen concentration will cause overheating during the welding process and even oxidation reactions; while too low an oxygen content may lead to substandard welding quality. Therefore, the system ensures that the oxygen concentration is within an appropriate range through the oxygen sensor.

[0051] Temperature sensor: Monitors the temperature of the furnace. During the welding process, the temperature inside the furnace must be precisely controlled to ensure welding quality and energy efficiency. By monitoring the furnace temperature in real time, the system can adjust the heating power to maintain a stable temperature environment.

[0052] Gas flow sensor: For example, a nitrogen flow sensor, monitors the flow rates of different gases (such as nitrogen, argon, etc.). These gases play a protective role during the welding process, preventing metal oxidation and controlling the atmosphere in the welding area.

[0053] The data obtained by the sensors can be transmitted to the control system in real time for subsequent data analysis and decision-making.

[0054] Process parameters refer to the specific operating specifications or requirements that need to be followed during the welding process. These process parameters are generally preset by welding engineers and vary according to different welding tasks. Common welding process parameters include:

[0055] Target temperature: Refers to the ideal temperature that needs to be maintained inside the furnace during the welding process. The efficiency and quality of the welding process depend on the temperature stability.

[0056] Maximum oxygen content: Refers to the highest oxygen concentration allowed during the welding process. Too high an oxygen content may cause overheating or oxidation reactions, thus affecting the welding quality.

[0057] Atmosphere composition requirements: For example, the composition and flow rates of protective gases such as nitrogen and argon. Depending on the welding materials, the required atmosphere composition may vary.

[0058] Welding speed and current: These parameters directly affect the penetration depth, width, welding strength, etc. of the welding. The system needs to adjust these parameters based on real-time feedback.

[0059] Combining the real-time data collected by the sensors with the process parameters forms a complete initial data set. This data set includes:

[0060] Real-time sensor data: Temperature, pressure, gas flow rate, oxygen concentration, etc.

[0061] Process parameters: target temperature, maximum oxygen content, atmosphere requirements, welding current, welding speed, etc.

[0062] The system integrates these data and calibrates the numerical range of each parameter according to the requirements of the welding task. These initial data provide the basis for subsequent welding process control and optimization. For example, the difference between the target temperature and the actual furnace temperature helps the model determine whether to adjust the heating power; the difference between the oxygen concentration and the maximum oxygen content affects the control of the gas flow.

[0063] These initial data not only provide the basis for subsequent control decisions, but also provide a "benchmark value" for model prediction and optimization. For example, by comparing the real-time sensor data with the preset process parameters, it can be judged whether the current welding state meets the process requirements. If the sensor data deviates from the process requirements, the system can automatically make feedback adjustments. Therefore, the system must accurately obtain these initial data as the basis for judging whether to adjust the process and control parameters during the welding process.

[0064] In addition to the initial data, the system also needs to continuously obtain the latest real-time data from the sensors to update the control model of the system. During the welding process, parameters such as the atmosphere, temperature, and gas flow in the furnace are dynamically changing, and the real-time feedback data can help the system make online adjustments.

[0065] In step S110, the system obtains the key data in the welding environment in real time through sensors and combines the preset process parameters to form an initial data set. These data are crucial for subsequent prediction, decision-making, and optimization. During the entire welding process, the system needs to continuously obtain and update these data to ensure the stability and efficiency of the welding quality.

[0066] S120: Input the initial data into the LSTM prediction network for parameter prediction and determination of the confidence interval to obtain the prediction result.

[0067] In this embodiment, the prediction result includes the temperature change and the oxygen content trend and the corresponding confidence interval.

[0068] Specifically, the LSTM prediction network is trained based on historical data.

[0069] After obtaining the initial data, the LSTM (Long Short-Term Memory) model starts to analyze the historical data (such as the sensor data in the past 60 minutes). The LSTM model predicts the change trend of the furnace temperature and oxygen content in the next period of time by learning the time series law of the historical data. The prediction results include the changes in temperature and oxygen content within the next 30 minutes and their confidence intervals. These prediction information can help adjust the current control strategy and respond to potential abnormal changes in advance.

[0070] In this embodiment, LSTM (Long Short-Term Memory) is a special type of recurrent neural network (RNN), which is very suitable for processing and predicting time series data. Compared with traditional RNNs, LSTM can better capture long-term dependencies, so it has significant advantages in predicting temperature changes and oxygen concentrations during the welding process.

[0071] LSTM controls the flow of information through its unique "gating mechanism", enabling it to effectively remember long-term information and forget irrelevant information. For the time series data involved in the welding process, LSTM can extract patterns from historical data and then predict the system state in the future for a period of time.

[0072] Before step S120, the LSTM model has been trained based on historical data. This historical data may include multiple parameters such as temperature, oxygen content, gas flow rate, and pressure collected by sensors over a past period of time (e.g., 60 minutes). These data constitute a large time series dataset, and the model learns the temporal relationships and dynamic change patterns between different parameters during the welding process through these data.

[0073] The core of the training process is to enable the LSTM network to learn the dependencies between each time point and learn how to predict future changes based on the current state. For example, LSTM can learn how the current temperature and oxygen concentration affect future temperature and oxygen changes, and how this change is affected by other process parameters (such as gas flow rate, pressure, etc.).

[0074] After obtaining the initial data, this real-time sensor data will be fed as input into the trained LSTM model. The initial data usually includes information such as current temperature, oxygen concentration, pressure, and gas flow rate. These data reflect the current welding state, providing a benchmark state for the LSTM model, and the model will predict future changes based on this.

[0075] The LSTM network will analyze the historical data and the current input data, and combine the trends over a past period of time (e.g., 60 minutes) to generate a prediction of the future state. This process is not just a simple prediction of the values of temperature or oxygen concentration, but a prediction of the change trends of these parameters according to the time series patterns.

[0076] The output of the LSTM network is not just a point estimate, but includes the predicted value and its confidence interval. Specifically, the LSTM prediction model will provide:

[0077] Temperature change trend: Predict the temperature change over a future period of time (e.g., 30 minutes). For example, predict whether the temperature will rise, fall, or remain stable.

[0078] Oxygen content trend: Similarly, predict the change trend of oxygen concentration within the next 30 minutes, including whether the oxygen concentration is too high or too low.

[0079] In addition, the LSTM also outputs the confidence interval for each predicted value. The confidence interval represents the uncertainty of the prediction result and is usually given in the form of a range. For example, the predicted temperature may be "90°C ± 2°C", indicating that the confidence level of the system for this predicted value is within the range of ±2°C. The confidence interval provides a measure of the reliability of the prediction result and helps the operator evaluate the credibility of the prediction result.

[0080] The confidence interval of the LSTM network prediction value is usually determined in the following way:

[0081] Error analysis: By analyzing the errors of the model prediction, the system can estimate the confidence level of the model in different prediction intervals. If the errors in the historical data are small, the confidence interval is small; if the errors are large, the confidence interval is wide.

[0082] Bayesian inference: Sometimes, the LSTM model can also be combined with the Bayesian inference method to quantify the prediction uncertainty of the model. The Bayesian method can provide a range of uncertainty for each prediction, making the prediction result not just a single value but a result with a probability distribution.

[0083] Based on the prediction results and confidence intervals of the LSTM model, the system can react to future change trends in advance. For example:

[0084] If it is predicted that the temperature will rise sharply within the next 30 minutes and the confidence interval is narrow, the system can adjust the heating power or gas flow in advance to prevent the temperature from being too high and affecting the welding quality.

[0085] If the predicted oxygen concentration may exceed the set range and the confidence interval is wide, the operator can observe and adjust the gas flow more carefully to maintain the stability of the atmosphere.

[0086] These prediction information not only provides the basis for the system's active control but also enables the operator to respond to potential abnormal changes in advance and prevent problems during the welding process.

[0087] According to the prediction results and confidence intervals of the LSTM prediction, the control system can make dynamic adjustments. For example, if the predicted oxygen concentration exceeds the predetermined range, the system can adjust the oxygen flow to maintain the target concentration. Through real-time feedback and adjustment, the system can improve efficiency and safety while ensuring the welding quality.

[0088] In step S120, the LSTM prediction network predicts the changing trends of temperature and oxygen content within a certain period in the future based on the initial data and historical data, and calculates the confidence interval. These prediction results provide a scientific basis for the real-time optimization and adjustment of the welding process, enabling the system to identify potential problems in advance and make adjustments, thereby ensuring the stability and quality of the welding process.

[0089] S130. When the number of error times predicted by the LSTM prediction network does not exceed the set threshold, the initial data and the prediction results are processed using the DDPG algorithm to obtain a decision result.

[0090] In this embodiment, the decision result includes the adjustment amounts of heating power, nitrogen valve opening, and fan speed, and the decision result is physically limited.

[0091] Based on the data predicted by the LSTM, the DDPG (Deep Deterministic Policy Gradient) algorithm makes decisions according to the current process requirements and prediction results. The DDPG algorithm dynamically adjusts control parameters (such as heating power, nitrogen flow rate, fan speed, etc.). The goal of the algorithm is to minimize energy consumption while meeting the process requirements. For example, if it is predicted that the temperature may overshoot, the DDPG will reduce the heating power in advance; if the oxygen content exceeds the standard, the DDPG will increase the nitrogen flow rate for adjustment.

[0092] In step S130, if the number of error times of the LSTM prediction network does not exceed the set threshold, it enters the next stage, and the DDPG (Deep Deterministic Policy Gradient) algorithm is used to process the initial data and prediction results to obtain the corresponding decision result. This process not only depends on the data predicted by the LSTM, but also combines the process requirements and physical limitations to dynamically adjust multiple control parameters, such as heating power, nitrogen flow rate, fan speed, etc.

[0093] Make decisions using the DDPG algorithm based on the LSTM prediction results and initial data

[0094] In this step, the DDPG algorithm dynamically adjusts the parameters of the control system during the welding process by processing the initial data and the results predicted by the LSTM. Specifically, the DDPG will make control decisions including heating power, nitrogen valve opening, and fan speed according to the current process requirements and prediction data, and all decisions need to comply with physical limitations to ensure the stability and safety of the welding process.

[0095] The goal of the DDPG algorithm is to minimize energy consumption while meeting the process requirements by adjusting the control parameters. For example, reduce unnecessary energy consumption while ensuring that the temperature does not overshoot.

[0096] At the same time, it is also necessary to ensure that the temperature stability, oxygen concentration control and other welding process parameters are kept within the set range during the process.

[0097] In one embodiment, see Figure 2 , the above-mentioned step S130 may include steps S131 to S132.

[0098] S131 , combining the relevant data detected by the sensor in the initial data with the prediction result, and combining the process parameters as a state space.

[0099] In this example, the sensor information in the initial data and the LSTM prediction results are combined to form a state space. This state space is the basis for the DDPG algorithm to make decisions, providing a comprehensive understanding of the current welding status and future trends.

[0100] Initial data: includes relevant data detected by sensors, such as current temperature, oxygen content, pressure, gas flow, etc.

[0101] LSTM prediction results: including future temperature change trends and oxygen concentrations predicted by the LSTM model.

[0102] Process parameters: These are the parameters required to define the welding process, such as heating power, gas flow, fan speed, etc.

[0103] These pieces of information together constitute the state space, and the DDPG algorithm will make corresponding decisions based on this state information.

[0104] The state space provides the DDPG algorithm with comprehensive current system information, enabling the algorithm to make precise control adjustments based on process requirements.

[0105] LSTM's prediction results help the system understand future trends, allowing decisions to be adjusted in advance to avoid overshoot or excessive adjustment.

[0106] S132. Use the DDPG algorithm to output the action space according to the state space to obtain a decision result; wherein the reward function of the DDPG algorithm calculates the reward through the comprehensive performance of total energy consumption, temperature stability and oxygen content control.

[0107] In this embodiment, the DDPG algorithm performs calculations based on the state space constructed above and outputs a decision result, which includes specific adjustments to control parameters (such as heating power, nitrogen flow rate, fan speed, etc.) during the welding process.

[0108] DDPG is an algorithm based on deep reinforcement learning, which is trained using the Actor-Critic structure. The Actor is used to generate actions (i.e., the amount of control parameter adjustment), while the Critic is used to evaluate the value of the actions and provide feedback.

[0109] This algorithm uses the current state space to select the optimal action, update the policy and execute it.

[0110] Heating power adjustment amount: According to the temperature trend predicted by LSTM, the DDPG algorithm can decide whether to increase or decrease the heating power to keep the temperature within an appropriate range.

[0111] Nitrogen flow rate adjustment amount: If it is predicted that the oxygen concentration exceeds the standard, DDPG can increase the nitrogen flow rate for adjustment.

[0112] Fan speed adjustment amount: Adjusting the fan speed helps to control the gas flow and temperature, thus maintaining stable welding conditions.

[0113] The DDPG algorithm not only makes decisions according to the process requirements, but also ensures that the adjustment results meet the physical limitations. For example, the heating power cannot exceed the maximum load capacity of the equipment, the nitrogen flow rate cannot exceed the maximum limit of the pipeline flow rate, and the fan speed needs to be within a safe range.

[0114] The reward function in the DDPG algorithm is very crucial. It is calculated based on multiple factors to ensure that the system's decisions meet the goals. The reward function comprehensively evaluates the system performance from the following aspects:

[0115] Total energy consumption: Optimizing energy use is one of the goals of DDPG, so the algorithm will evaluate whether the current decision leads to excessive energy consumption.

[0116] Temperature stability: The temperature must be kept stable within a set range, avoiding being too high or too low. If the temperature stability is poor, the reward will be reduced.

[0117] Oxygen content control: Excessive or too low oxygen concentration may affect the welding quality, so the algorithm will give corresponding rewards according to the effect of oxygen concentration control.

[0118] Based on these factors, the algorithm will calculate a comprehensive reward value and adjust the policy through this value. The higher the reward value, the closer the current decision is to the optimization goal, and the system will gradually learn more effective adjustment strategies.

[0119] In step S130, the DDPG algorithm utilizes the results predicted by the LSTM and the real-time sensor information in the initial data, combines the process requirements and physical limitations of the welding process, and dynamically adjusts control parameters (such as heating power, nitrogen flow rate, fan speed, etc.). Through the method of deep reinforcement learning, DDPG can minimize energy consumption and optimize the welding process on the premise of meeting process requirements such as temperature stability and oxygen concentration. The design of the reward function ensures that the algorithm can perform effective feedback learning in practical applications and gradually improve the adjustment ability of the system.

[0120] S140. Adjust the parameters of the welding equipment according to the decision result.

[0121] After the execution stage, the sensor monitors and feeds back information in real time, such as energy consumption, temperature, oxygen content, etc. The system performs error analysis based on the difference between the actual feedback data and the prediction result. If there is a large deviation between the actual energy consumption and the predicted energy consumption, or if the temperature and oxygen concentration deviate significantly from the target values, the system will update the LSTM and DDPG models for online learning. In this way, the system continuously optimizes the prediction accuracy and control strategy to adapt to different welding process conditions.

[0122] The core of this method is to predict the temperature and oxygen content through the LSTM model and use the DDPG algorithm for dynamic adjustment, so as to optimize the energy consumption while ensuring the process requirements. The closed-loop control system ensures the stable control of temperature and gas concentration through the process of real-time data acquisition, prediction, decision-making, and execution feedback, while reducing energy consumption and improving welding efficiency.

[0123] The design concept of this system can be widely applied to various control tasks in industrial automation, especially fine control in high-temperature and high-pressure environments.

[0124] During the welding process, the core purpose of step S140 is to adjust the control parameters of the welding equipment according to the decision result output by the DDPG algorithm to ensure that the temperature, oxygen concentration, and other key process indicators are always within the predetermined range, while optimizing energy consumption and welding efficiency. Specifically, the execution of S140 includes the following key steps:

[0125] According to the output of the DDPG algorithm in step S130, the control system will adjust the parameters of the welding equipment, mainly including the following aspects:

[0126] Heating power adjustment: According to the predicted temperature change trend of the LSTM model, the DDPG algorithm may adjust the heating power to ensure that the temperature does not exceed the target range.

[0127] Nitrogen flow rate adjustment: If the oxygen concentration shows an upward trend, the DDPG algorithm will increase the nitrogen flow rate to reduce the oxygen concentration and avoid oxidation problems.

[0128] Fan speed adjustment: The fan speed may be adjusted according to the current temperature and gas concentration to maintain the stability of gas flow and the balance of temperature.

[0129] Through these adjustments, the welding equipment can maintain the best state during operation, ensuring welding quality and production efficiency.

[0130] In addition, when the number of error times predicted by the LSTM prediction network exceeds the set threshold, the control mode of the welding equipment is switched to the traditional PID control mode.

[0131] When the prediction error of the LSTM model exceeds the set threshold, it indicates that the LSTM model cannot effectively process the current welding process, and the system will automatically switch to the traditional PID control mode. PID control is a classic control method, suitable for relatively simple feedback control tasks such as temperature and gas concentration. PID control adjusts the control quantity through three parameters (proportional, integral, differential), making the system response more stable and ensuring basic process control when the prediction accuracy is poor.

[0132] Frequent or large LSTM prediction errors indicate that the changes in the current welding process exceed the adaptation ability of the LSTM model.

[0133] When the system cannot quickly correct the prediction error through online learning, PID control can be used as an alternative solution to ensure basic control functions.

[0134] The design idea of this method can dynamically adjust control parameters by combining LSTM prediction, DDPG decision-making, and real-time feedback, achieving optimal energy consumption control while ensuring process requirements. The specific design ideas include:

[0135] Temperature and oxygen concentration control: Predict the future temperature and oxygen concentration trends through LSTM, and DDPG dynamically adjusts control parameters to ensure temperature stability and gas concentration within the target range.

[0136] Energy consumption optimization: The system continuously adjusts the control strategy according to the difference between real-time data and predicted data to reduce unnecessary energy consumption.

[0137] Closed-loop control: The system ensures precise control of the welding process through a closed-loop process of real-time data acquisition, prediction, decision-making, execution, and feedback, thereby improving welding efficiency.

[0138] In addition, this control method can be widely applied to other control tasks in industrial automation, especially suitable for fine control requirements in complex environments such as high temperature and high pressure.

[0139] The system performs real-time control of the welding equipment based on the LSTM prediction results and the DDPG decision output. The system conducts error analysis through real-time sensor feedback and continuously optimizes the LSTM and DDPG models through online learning. Through the closed-loop control system, the temperature, oxygen concentration, and energy consumption are effectively controlled, and the welding process is more efficient and stable. Switching to the PID control mode is an effective compensation for the failure of the LSTM model, ensuring the reliability and safety of the welding process.

[0140] The method of this embodiment is executed by the following hardware: a pressure sensor and an oxygen sensor.

[0141] When selecting a pressure sensor, the requirements of the high-temperature environment should be considered. It is recommended to use sensors with high temperature resistance (above 300 °C) and strong anti-interference ability, such as piezoelectric or MEMS (Micro-Electro-Mechanical System) sensors. For the range requirement, it should cover the furnace pressure range (e.g., ±5 kPa), and the accuracy should be ±0.1% FS (Full Scale). Common models include Honeywell's 26PC series (with high temperature resistance and anti-vibration characteristics) and Siemens' Sitrans P series.

[0142] For the oxygen sensor, it is recommended to select a tunable diode laser absorption spectroscopy (TDLAS) or zirconia (ZrO2) electrochemical sensor in a high-temperature environment. Its range is usually 0 - 1000 ppm (for semiconductor processes) or 0 - 21% O2 (for general processes), and the response time should be less than 5 seconds. Recommended models include Siemens' Oxymax series (laser spectroscopy type) and Figaro's KE series (electrochemical type).

[0143] The pressure sensor should be arranged at the top of the furnace (for monitoring thermal expansion pressure), the bottom (for monitoring air flow pressure), and the circulating air inlet (for dynamic pressure balance). Multi-sensor redundant deployment should be adopted to avoid data loss caused by single-point failure.

[0144] The oxygen sensor can be arranged at the center of the furnace (where the gas is relatively uniform) and the exhaust gas outlet (for monitoring the residual oxygen content). It is necessary to avoid placing the sensor in the high-temperature direct injection area, and the high-temperature resistance of the sensor can be improved by installing a heat insulation sleeve.

[0145] Industrial-grade DAQ (Data Acquisition Card), such as NI CompactDAQ, is selected for data acquisition, which supports multi-channel input. The signal conditioning module needs to filter and amplify weak signals. The communication protocol should be selected as industrial Ethernet (Profinet, EtherCAT) or RS-485 (Modbus RTU) to ensure the real-time nature of data transmission, with the delay controlled within 100 ms.

[0146] Edge computing uses an embedded industrial computer, such as Advantech ARK-1123, to run edge algorithms (such as pressure compensation calculation and oxygen concentration calibration) to further optimize the accuracy of sensor data.

[0147] In terms of anti-interference, the sensor signal line should use a twisted pair shielded cable to avoid electromagnetic interference from devices such as frequency converters or motors. For temperature effects, it is recommended to install a built-in temperature sensor (such as PT100) to correct the influence of ambient temperature on measurement data in real time.

[0148] After data acquisition, prediction is performed through LSTM (Long Short-Term Memory Network), and then decision-making is carried out through DDPG (Deep Deterministic Policy Gradient). Finally, control commands are executed and sensor feedback is obtained. The specific steps are as follows:

[0149] Input real-time sensor data (temperature, oxygen content, pressure, current, nitrogen flow rate) and process parameters (such as target temperature curve, maximum oxygen content threshold, energy consumption limit).

[0150] Use historical data from the past 60 minutes (sampled every 5 minutes) to predict the trends of temperature and oxygen content within the next 30 minutes and give a confidence interval.

[0151] Input the current sensor data, LSTM prediction results, and process target deviations, and output control commands (such as heating power, nitrogen flow rate, fan speed).

[0152] Design a reward function to balance the control of energy consumption, temperature stability, and oxygen content, and optimize the strategy.

[0153] The PLC control module performs adjustment operations, such as heating power adjustment, nitrogen injection, and variable frequency fan speed regulation.

[0154] Update sensor data in real time and use it for the rolling training of the LSTM model and the update of the DDPG strategy.

[0155] Predict the changes in furnace temperature and oxygen concentration within the next 10 - 30 minutes through LSTM and make advance judgments. Dynamically adjust the heating power, nitrogen flow rate, and fan speed through the DDPG algorithm to not only meet the process requirements but also reduce energy consumption. Complete the full-process automated control from prediction, decision-making, execution to feedback.

[0156] Input the historical temperature curve, pressure fluctuations, nitrogen flow rate, and oxygen concentration sensor data. Train an LSTM model to predict the temperature change and oxygen concentration trend within the next 10 minutes. Adjust the nitrogen flow rate or heating power in advance according to the prediction results to avoid temperature overshoot or oxidation risk. Call the LSTM model every 5 minutes to predict the temperature and oxygen concentration trends in the next 30 minutes. If the predicted oxygen concentration exceeds the threshold (e.g., 100 ppm), take nitrogen intervention measures in advance.

[0157] Input real-time temperature, pressure, energy consumption data, and process goals (such as curing time, etc.). Use the DDPG algorithm to train the agent to dynamically adjust the fan speed, heating power, and nitrogen flow rate to achieve the lowest energy consumption goal. Design a reward function to comprehensively consider temperature stability and energy consumption. The reward function includes the total energy consumption (heating, nitrogen, fan), the difference between the actual temperature and the target temperature, and whether the oxygen content exceeds the limit. According to the LSTM prediction results, the DDPG algorithm generates control instructions. For example, if the predicted temperature rises too fast, reduce the heating power in advance; if the predicted oxygen concentration rises, increase the nitrogen flow rate in advance. The subsequent changes in nitrogen, oxygen content, etc. predicted by LSTM, based on this change trend, timely adjust the fan speed, heating power, and nitrogen flow rate, etc., to make the equipment achieve the lowest energy consumption under the condition of meeting the process standards. Simulate 10,000 heating-cooling cycles in the digital twin model, continuously optimize the initial strategy, and select the model with better training results for deployment.

[0158] During the adjustment process, since the parameters are changing in real time, the parameters also need to be adjusted in real time. Use reinforcement learning (DDPG) to train the fan speed, heating power, and nitrogen flow rate, etc., to achieve the lowest energy consumption under the condition of meeting the process requirements.

[0159] In addition, the reward function is: R = -(α·P energy +β|T actual -T target |+γ·max(O2 - O 2_max , 0), where P is the current total energy consumption (heating + nitrogen + fan), the second term after β is the difference between the actual temperature and the target temperature, the second term after γ is whether the oxygen content exceeds the limit, and α, β, γ are weight coefficients (adjusted according to process requirements).

[0160] The actual energy consumption and prediction deviation are used to update the LSTM and DDPG models in real time for online learning to continuously optimize the control strategy. If the prediction error of the LSTM model exceeds 5°C continuously for 3 times, switch to the traditional PID control mode and trigger an alarm. The control instructions of DDPG need to be limited by physical limits, for example, the heating power shall not be lower than the process lower limit.

[0161] The above vertical furnace welding parameter monitoring and adjustment method combines an LSTM prediction network and a DDPG algorithm to achieve accurate temperature and oxygen content prediction. On this basis, energy consumption is optimized, and system stability and energy efficiency are improved. First, welding parameters and process data are obtained through sensors and input into the LSTM model for temperature and oxygen concentration prediction, and the prediction confidence interval is determined. When the prediction error remains within the set threshold, the DDPG algorithm is used for decision-making by combining the initial data and the prediction results to adjust the welding equipment parameters in real time to ensure process accuracy. In this way, not only can the changes in temperature and oxygen concentration be accurately predicted, but also dynamic adjustment can be made according to real-time data, thereby reducing energy consumption, improving system stability and energy efficiency, and having the ability of online learning and self-optimization to gradually optimize the control strategy.

[0162] Figure 3 It is a schematic flowchart of a vertical furnace welding parameter monitoring and adjustment method provided by another embodiment of the present invention. As Figure 3 shown, the vertical furnace welding parameter monitoring and adjustment method of this embodiment includes steps S210 - S260. Among them, steps S210 - S240 are similar to steps S110 - S1'40 in the above embodiment and will not be elaborated here. The following will detail steps S260 - S270 added in this embodiment.

[0163] S250. Obtain the relevant data detected by the sensor again;

[0164] S260. Retrain the LSTM prediction network and the DDPG algorithm according to the relevant data obtained again.

[0165] During the welding process, the sensor will collect key process data in real time, such as temperature, oxygen concentration, gas flow rate, and energy consumption, etc. These data will be used as the feedback information of the system and transmitted to the control system for analysis and processing. The specific content of the feedback data includes:

[0166] Energy consumption: The system monitors the energy consumption during the welding process in real time and compares it with the energy consumption predicted by the LSTM to judge whether there are abnormal fluctuations.

[0167] Temperature: The temperature of the welding area is a key parameter in the control process and needs to be stabilized within a preset range to avoid the influence of too high or too low temperature on the welding quality.

[0168] Oxygen concentration: The oxygen concentration directly affects the welding quality and must be strictly controlled to prevent oxidation caused by too high oxygen.

[0169] After receiving the real-time feedback data, the system will perform error analysis on the difference between the actual feedback data and the prediction result to judge whether there is a large deviation:

[0170] Energy consumption error: If there is a large deviation between the actual energy consumption and the predicted energy consumption, the system will identify possible control problems, such as too high heating power or unreasonable gas flow.

[0171] Temperature error: If there is a large gap between the actual temperature and the target value, it may indicate a lag in the response of the temperature control system or inaccurate temperature prediction, and corresponding adjustments are required.

[0172] Oxygen concentration error: If the oxygen concentration fails to be maintained within the predetermined range, it indicates an error in gas flow control, and the nitrogen flow or fan speed needs to be corrected.

[0173] This error analysis can discover potential problems and correct them in a timely manner by comparing real-time data acquisition with predicted data.

[0174] According to the error analysis results of real-time feedback data, the system will dynamically adjust the LSTM and DDPG models to optimize the prediction accuracy and control strategy. This process is called online learning, and its specific steps include:

[0175] LSTM model update: If the number of error times predicted by LSTM exceeds the set threshold, it means that the accuracy of the prediction result needs to be improved. The system will use the latest feedback data to train and update the LSTM model, thereby improving the accuracy of predicting temperature and oxygen concentration.

[0176] DDPG algorithm update: When the prediction error of LSTM is large, the control strategy of the DDPG algorithm may also need to be adjusted. By combining new data feedback, the DDPG model will re-evaluate the decision-making strategy of control parameters to ensure more efficient welding control.

[0177] The process of online learning is a continuously optimized closed-loop control system. It continuously optimizes the prediction results and control strategies through new feedback data to ensure the stability and accuracy of the welding process.

[0178] Figure 4 It is a schematic block diagram of a vertical furnace welding parameter monitoring and adjustment device 300 provided by an embodiment of the present invention. As Figure 4 shown, corresponding to the above vertical furnace welding parameter monitoring and adjustment method, the present invention also provides a vertical furnace welding parameter monitoring and adjustment device 300. The vertical furnace welding parameter monitoring and adjustment device 300 includes units for executing the above vertical furnace welding parameter monitoring and adjustment method, and this device can be configured in a server. Specifically, please refer to Figure 4 . The vertical furnace welding parameter monitoring and adjustment device 300 includes a data acquisition unit 301, a prediction unit 302, a decision-making unit 303, and an adjustment unit 304.

[0179] The data acquisition unit 301 is configured to acquire the relevant data detected by the sensor and the process parameters of welding to obtain the initial data. The prediction unit 302 is configured to input the initial data into the LSTM prediction network for parameter prediction and determination of the confidence interval to obtain the prediction result. The decision-making unit 303 is configured to, when the number of error times predicted by the LSTM prediction network does not exceed the set threshold, process the initial data and the prediction result by using the DDPG algorithm to obtain the decision result. The adjustment unit 304 is configured to adjust the parameters of the welding equipment according to the decision result.

[0180] It further includes a traditional control unit, which is configured to, when the number of error times predicted by the LSTM prediction network exceeds the set threshold, switch the control mode of the welding equipment to the traditional PID control mode.

[0181] In one embodiment, as Figure 5 shown, the decision-making unit 303 includes a combination subunit 3031 and an action output subunit 3032.

[0182] The combination subunit 3031 is configured to combine the relevant data detected by the sensor in the initial data with the prediction result and combine the process parameters as the state space. The action output subunit 3032 is configured to output the action space according to the state space by using the DDPG algorithm to obtain the decision result. Wherein, the reward function of the DDPG algorithm calculates the reward through the comprehensive performance of the total energy consumption, temperature stability, and oxygen content control.

[0183] Figure 6 It is a schematic block diagram of a vertical furnace welding parameter monitoring and adjustment device 300 provided by another embodiment of the present invention. As Figure 6 shown, the vertical furnace welding parameter monitoring and adjustment device 300 in this embodiment adds a re-acquisition unit 305 and a training unit 306 on the basis of the above embodiment.

[0184] The re-acquisition unit 305 is configured to re-acquire the relevant data detected by the sensor. The training unit 306 is configured to re-train the LSTM prediction network and the DDPG algorithm according to the re-acquired relevant data.

[0185] It should be noted that those skilled in the art can clearly understand that the specific implementation processes of the above vertical furnace welding parameter monitoring and adjustment device 300 and each unit can refer to the corresponding descriptions in the foregoing method embodiments. For the convenience and conciseness of description, they will not be elaborated herein.

[0186] The above vertical furnace welding parameter monitoring and adjustment device 300 can be implemented in the form of a computer program, and this computer program can run on a computer device as Figure 7 shown.

[0187] Please refer to Figure 7 , Figure 7 which is a schematic block diagram of a computer device provided by an embodiment of the present application. The computer device 500 may be a server. Among them, the server may be an independent server or a server cluster composed of multiple servers.

[0188] Refer to Figure 7 , the computer device 500 includes a processor 502, a memory, and a network interface 505 connected through a system bus 501. Among them, the memory may include a non-volatile storage medium 503 and an internal memory 504.

[0189] The non-volatile storage medium 503 can store an operating system 5031 and a computer program 5032. The computer program 5032 includes program instructions. When the program instructions are executed, the processor 502 can be made to execute a vertical furnace welding parameter monitoring and adjustment method.

[0190] The processor 502 is used to provide computing and control capabilities to support the operation of the entire computer device 500.

[0191] The internal memory 504 provides an environment for the operation of the computer program 5032 in the non-volatile storage medium 503. When the computer program 5032 is executed by the processor 502, the processor 502 can be made to execute a vertical furnace welding parameter monitoring and adjustment method.

[0192] The network interface 505 is used for network communication with other devices. Those skilled in the art can understand that Figure 7 the structure shown in

[0193] is only a block diagram of a part of the structure related to the solution of the present application, and does not constitute a limitation on the computer device 500 to which the solution of the present application is applied. The specific computer device 500 may include more or fewer components than those shown in the figure, or combine some components, or have a different component layout.

[0194] Among them, the processor 502 is used to run the computer program 5032 stored in the memory to implement the following steps: obtaining relevant data detected by the sensor and the process parameters of welding to obtain initial data; inputting the initial data into the LSTM prediction network for parameter prediction and determination of the confidence interval to obtain a prediction result; when the number of error times predicted by the LSTM prediction network does not exceed a set threshold, processing the initial data and the prediction result using the DDPG algorithm to obtain a decision result; adjusting the parameters of the welding device according to the decision result.

[0195] Among them, the LSTM prediction network is trained based on historical data.

[0196] The predicted results include temperature changes, oxygen content trends, and corresponding confidence intervals.

[0197] The decision results include the adjustment amounts of heating power, nitrogen valve opening, and fan speed.

[0198] In one embodiment, after the processor 502 implements the step of adjusting the parameters of the welding device according to the decision results, the following steps are further implemented:

[0199] Obtain the relevant data detected by the sensor again; retrain the LSTM prediction network and the DDPG algorithm according to the relevant data obtained again.

[0200] In one embodiment, when the processor 502 implements the step of processing the initial data and the predicted results using the DDPG algorithm to obtain the decision results, the following steps are specifically implemented:

[0201] Combine the relevant data detected by the sensor in the initial data with the predicted results, and combine the process parameters as the state space; use the DDPG algorithm to output the action space according to the state space to obtain the decision results; wherein, the reward function of the DDPG algorithm calculates the reward through the comprehensive performance of total energy consumption, temperature stability, and oxygen content control.

[0202] Among them, the decision results are physically restricted.

[0203] In one embodiment, after the processor 502 implements the step of inputting the initial data into the LSTM prediction network for parameter prediction and determination of the confidence interval to obtain the predicted results, the following steps are further implemented:

[0204] Input the initial data into the LSTM prediction network for parameter prediction and determination of the confidence interval to obtain the predicted results

[0205] It should be understood that in the embodiments of the present application, the processor 502 may be a central processing unit (CPU), and the processor 502 may also be other general-purpose processors, digital signal processors (DSPs), application specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. Among them, the general-purpose processor may be a microprocessor or the processor may also be any conventional processor, etc.

[0206] Those of ordinary skill in the art can understand that all or part of the processes in the methods of the above embodiments can be completed by instructing relevant hardware through a computer program. The computer program includes program instructions, and the computer program can be stored in a storage medium, and the storage medium is a computer-readable storage medium. The program instructions are executed by at least one processor in the computer system to implement the process steps of the embodiments of the above methods.

[0207] Therefore, the present invention also provides a storage medium. The storage medium may be a computer-readable storage medium. The storage medium stores a computer program, and when the computer program is executed by a processor, the processor executes the following steps:

[0208] Obtain relevant data detected by the sensor and welding process parameters to obtain initial data; input the initial data into the LSTM prediction network for parameter prediction and determination of the confidence interval to obtain a prediction result; when the number of error times predicted by the LSTM prediction network does not exceed a set threshold, process the initial data and the prediction result using the DDPG algorithm to obtain a decision result; adjust the parameters of the welding equipment according to the decision result.

[0209] Among them, the LSTM prediction network is trained based on historical data.

[0210] The prediction result includes temperature change, oxygen content trend and corresponding confidence intervals.

[0211] The decision result includes the adjustment amounts of heating power, nitrogen valve opening and fan speed.

[0212] In an embodiment, after the processor executes the computer program to implement the step of adjusting the parameters of the welding equipment according to the decision result, the following steps are also implemented:

[0213] Obtain the relevant data detected by the sensor again; retrain the LSTM prediction network and the DDPG algorithm according to the obtained relevant data again.

[0214] In one embodiment, when the processor executes the computer program to implement the step of processing the initial data and the prediction result by the DDPG algorithm to obtain a decision result, the following steps are specifically implemented:

[0215] Combine the relevant data detected by the sensor in the initial data with the prediction result, and combine the process parameters as the state space; use the DDPG algorithm to output the action space according to the state space to obtain a decision result; wherein, the reward function of the DDPG algorithm calculates the reward through the comprehensive performance of the total energy consumption, temperature stability and oxygen content control.

[0216] Among them, the decision result is restricted by physics.

[0217] In one embodiment, after the processor executes the computer program to implement the step of inputting the initial data into the LSTM prediction network for parameter prediction and confidence interval determination to obtain a prediction result, the following steps are also implemented:

[0218] When the number of error times predicted by the LSTM prediction network exceeds the set threshold, switch the control mode of the welding equipment to the traditional PID control mode.

[0219] The storage medium can be various computer-readable storage media such as a USB flash drive, a mobile hard disk, a read-only memory (ROM), a magnetic disk, or an optical disc that can store program codes.

[0220] Those of ordinary skill in the art can realize that the units and algorithm steps of each example described in combination with the embodiments disclosed herein can be implemented by electronic hardware, computer software, or a combination of the two. To clearly illustrate the interchangeability of hardware and software, the components and steps of each example have been generally described according to their functions in the above description. Whether these functions are executed in a hardware or software manner depends on the specific application and design constraints of the technical solution. Professional technicians can use different methods to implement the described functions for each specific application, but such implementation should not be considered to exceed the scope of the present invention.

[0221] In several embodiments provided by the present invention, it should be understood that the disclosed devices and methods can be implemented in other ways. For example, the device embodiments described above are merely illustrative. For example, the division of each unit is only a logical function division, and there may be other division methods in actual implementation. For example, multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed.

[0222] The steps in the method embodiments of the present invention can be adjusted, combined, and deleted according to actual needs. The units in the device embodiments of the present invention can be combined, divided, and deleted according to actual needs. In addition, the functional units in each embodiment of the present invention can be integrated into one processing unit, or each unit can exist physically alone, or two or more units can be integrated into one unit.

[0223] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a storage medium. Based on such an understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. The computer software product is stored in a storage medium and includes several instructions for causing a computer device (which can be a personal computer, a terminal, or a network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention.

[0224] As described above, the above are only the specific embodiments of the present invention, but the protection scope of the present invention is not limited thereto. Any person skilled in the art within the technical scope disclosed by the present invention can easily think of various equivalent modifications or substitutions, and these modifications or substitutions should be covered by the protection scope of the present invention. Therefore, the protection scope of the present invention should be subject to the protection scope of the claims.

Claims

1. A method for monitoring and adjusting welding parameters of a vertical furnace, characterized in that, Including: Obtain relevant data detected by the sensor and welding process parameters to obtain initial data; Input the initial data into the LSTM prediction network for parameter prediction and determination of the confidence interval to obtain a prediction result; When the number of error times predicted by the LSTM prediction network does not exceed a set threshold, process the initial data and the prediction result using the DDPG algorithm to obtain a decision result; Adjust the parameters of the welding equipment according to the decision result.

2. The vertical furnace welding parameter monitoring and adjustment method according to claim 1, characterized in that After adjusting the parameters of the welding equipment according to the decision result, it further includes: Obtain relevant data detected by the sensor again; Retrain the LSTM prediction network and the DDPG algorithm according to the relevant data obtained again.

3. The vertical furnace welding parameter monitoring and adjustment method according to claim 1, characterized in that, The LSTM prediction network is trained based on historical data.

4. The vertical furnace welding parameter monitoring and adjustment method according to claim 1, wherein The prediction result includes temperature change, oxygen content trend, and corresponding confidence intervals.

5. The vertical furnace welding parameter monitoring and adjustment method according to claim 1, wherein Processing the initial data and the prediction result using the DDPG algorithm to obtain a decision result includes: Combine the relevant data detected by the sensor in the initial data with the prediction result, and combine the process parameters as the state space; Use the DDPG algorithm to output the action space according to the state space to obtain a decision result; wherein, the reward function of the DDPG algorithm calculates the reward through the comprehensive performance of total energy consumption, temperature stability, and oxygen content control.

6. The vertical furnace welding parameter monitoring and adjustment method according to claim 5, characterized in that The decision result includes the adjustment amounts of heating power, nitrogen valve opening, and fan speed.

7. The vertical furnace welding parameter monitoring and adjustment method according to claim 1, characterized in that, After inputting the initial data into the LSTM prediction network for parameter prediction and determination of the confidence interval to obtain a prediction result, it further includes: When the number of error times predicted by the LSTM prediction network exceeds the set threshold, switch the control mode of the welding equipment to the traditional PID control mode.

8. The vertical furnace welding parameter monitoring and adjustment method according to claim 5, characterized in that, The decision result is physically restricted.

9. Vertical furnace welding parameter monitoring and adjustment device, characterized in that, Including: A data acquisition unit for obtaining relevant data detected by the sensor and welding process parameters to obtain initial data; A prediction unit for inputting the initial data into the LSTM prediction network for parameter prediction and determination of the confidence interval to obtain a prediction result; A decision unit for processing the initial data and the prediction result using the DDPG algorithm to obtain a decision result when the number of error times predicted by the LSTM prediction network does not exceed a set threshold; An adjustment unit for adjusting the parameters of the welding equipment according to the decision result.

10. A computer device, characterized in that, The computer device includes a memory and a processor, and a computer program is stored on the memory. When the processor executes the computer program, the method described in any one of claims 1 to 8 is implemented.

Images