Method, device and equipment for controlling harmful gas of plasma oven and storage medium

Abstract

The invention discloses a plasma stove harmful gas control method, device and equipment and a storage medium, and relates to the technical field of plasma stove harmful gas discharge control, and the method comprises the steps: taking a current working parameter, a current environment parameter and a current harmful gas discharge concentration, which are obtained in real time, of a plasma stove as target input parameters; inputting the data into an emission prediction model trained through a historical time sequence data set; the target input parameters are processed through an emission prediction model, and the predicted emission concentration of the harmful gas at the next moment is obtained; and taking the current working parameter, the current environment parameter and the predicted emission concentration as input state data, processing the input state data based on a deep Q learning network to obtain a working parameter which enables the predicted emission concentration to be minimized and is adjusted by an action adjustment mechanism, and taking the working parameter as an optimal parameter. According to the invention, the working parameters of the plasma oven can be dynamically adjusted, and the emission of harmful gas is minimized.

Description

Technical Field

[0001] The present invention relates to the technical field of plasma stove harmful gas emission control, and in particular to a plasma stove harmful gas control method, device, equipment and storage medium. Background Art

[0002] A plasma cooker is a cooking device that uses plasma technology to generate flames. Its operating principle is based on the arc discharge phenomenon, using plasma technology to convert electrical energy into heat energy, which in turn generates flames. Specifically, this technology uses a high-frequency electric field to ionize the air, generating high-temperature plasma, which in turn forms an open flame. Unlike traditional combustion methods, plasma electric ignition technology does not require gas, eliminating the risk of gas leaks and explosions, and providing a new level of safety.

[0003] Plasma stoves, during their high-temperature discharge process, can produce harmful gases such as nitrogen oxides and carbon monoxide. These gases, when released untreated into the environment, can negatively impact air quality and the ecological environment. With increasing environmental protection requirements, reducing harmful gas emissions has become a key challenge for both domestic and industrial applications of plasma technology. Summary of the Invention

[0004] The present invention provides a plasma stove harmful gas control method, device, equipment and storage medium, which can dynamically adjust the working parameters of the plasma stove according to real-time data to ensure that the emission of harmful gases is minimized.

[0005] In order to achieve the above object, the present invention adopts the following technical solutions:

[0006] In a first aspect, the present invention provides a method for controlling harmful gases in a plasma stove, the method comprising:

[0007] Using the current operating parameters, current environmental parameters, and current harmful gas emission concentration of the plasma stove acquired in real time at the current moment as target input parameters, the target input parameters are input into an emission prediction model trained using a historical time series dataset; the historical time series dataset includes the historical operating parameters, historical environmental parameters, and historical harmful gas emission concentrations of the plasma stove acquired at multiple historical moments, as well as the actual harmful gas emission concentration corresponding to the next moment of each historical moment;

[0008] Processing the target input parameters through the emission prediction model to obtain the predicted emission concentration of the harmful gas at the next moment;

[0009] The current operating parameters, the current environmental parameters, and the predicted emission concentration are used as input state data, and the input state data is processed based on a deep Q-learning network to obtain the operating parameters adjusted by an action adjustment mechanism that minimizes the predicted emission concentration, and the operating parameters are used as optimal parameters; the action adjustment mechanism includes individual adjustment of each of the current operating parameters and simultaneous adjustment of multiple parameters;

[0010] The plasma cooker is controlled to operate with the optimal parameters.

[0011] In one possible implementation, the current operating parameters include the operating voltage, operating current, air flow rate, and electrode spacing of the plasma stove at the current moment; the current environmental parameters include the air pressure and humidity around the plasma stove at the current moment; the current harmful gas emission concentration includes the nitrogen oxide emission concentration and carbon monoxide emission concentration of the plasma stove at the current moment; the emission prediction model is an LSTM multi-task prediction model based on a spatiotemporal hybrid attention mechanism; the target input parameters are processed by the emission prediction model to obtain the predicted emission concentration of the harmful gas at the next moment, specifically:

[0012] The target input parameters are processed by an LSTM multi-task prediction model based on a spatiotemporal mixed attention mechanism to obtain the predicted nitrogen oxide emission concentration and carbon monoxide emission concentration of the plasma stove at the next moment.

[0013] In one possible implementation, the LSTM multi-task prediction model based on the spatiotemporal hybrid attention mechanism includes an input layer, a temporal attention layer, a spatial attention layer, a feature fusion layer, an LSTM shared feature layer, a multi-task prediction head, and an output layer connected in sequence;

[0014] The input layer is used to receive the historical time series dataset;

[0015] The time attention layer is used to determine the importance of a time step according to the changing trend of the actual harmful gas emission concentration in the historical time series data set, and to perform weighted processing on the sample data in the historical time series data set according to the importance of the time step to obtain weighted time series data;

[0016] The spatial attention mechanism is used to perform weighted processing on the historical working parameters according to the importance of each type of working parameters in the historical working parameters to obtain weighted working parameter data;

[0017] The feature fusion layer is used to fuse the time series data and the working parameter data to obtain feature fusion data;

[0018] The LSTM shared feature layer is used to learn the temporal dependency and the shared feature layer from the feature fusion data through multiple LSTM units to extract shared features;

[0019] The multi-task prediction head includes a nitrogen oxide prediction head and a carbon monoxide prediction head, wherein the nitrogen oxide prediction head predicts the predicted emission concentration of nitrogen oxide at the next moment through a first fully connected layer, and the carbon monoxide prediction head predicts the predicted emission concentration of carbon monoxide at the next moment through a second fully connected layer;

[0020] The output layer is used to output the predicted emission concentration of nitrogen oxides and the predicted emission concentration of carbon monoxide.

[0021] In one possible implementation, the input state data is processed based on a deep Q-learning network to obtain operating parameters adjusted by the action adjustment mechanism that minimize the predicted emission concentration, and the operating parameters are used as optimal parameters, specifically including:

[0022] The Q value after executing each action adjustment mechanism on the input state data is calculated through a deep learning network, the action adjustment mechanism with the largest Q value is used as the action adjustment mechanism that minimizes the predicted emission concentration, and the working parameters adjusted by the action adjustment mechanism with the largest Q value are output as the optimal parameters; the Q value is determined by the immediate reward of the current action adjustment mechanism and the expected value of the future reward obtained due to the state transfer caused by the action adjustment mechanism.

[0023] In a possible implementation, the Q value is calculated using the following formula:

[0024]

[0025] Where s represents the current environment state corresponding to the input state data, a represents the action adjustment mechanism taken under the current environment state, Q(s, a) represents the total return expected to be obtained by taking action adjustment mechanism a under the current environment state, Rt(a) represents the immediate reward obtained when taking action adjustment mechanism a, and γ represents the discount factor. Indicates that the Q value of the action adjustment mechanism that maximizes the Q value is selected from all possible action adjustment mechanisms a1 in the next environment state s1 as the expected value of the future reward;

[0026] The immediate reward is determined by a predefined reward function in a deep Q-learning network; the reward function is used to measure the size of the predicted emission concentration after executing the action adjustment mechanism, wherein the predicted emission concentration is inversely proportional to the immediate reward.

[0027] In a possible implementation, the predicted emission concentration includes the predicted emission concentration of nitrogen oxides and the predicted emission concentration of carbon monoxide; the reward function is specifically:

[0028] Rt=w1·(-NOx t -CO t )+w2·Efficiency t -w3·Concentration_Rate_Change t ;

[0029] Among them, Rt represents the reward function, NOx t and CO t They represent the predicted emission concentrations of nitrogen oxides and carbon monoxide at time t, respectively. t Indicates the discharge efficiency corresponding to the input state data, Concentration_Rate_Change t It represents the changing rate of harmful gas concentration, w1, w2 and w3 represent weighting factors respectively.

[0030] In one possible implementation, before using the current operating parameters, current environmental parameters and current harmful gas emission concentration of the plasma stove acquired in real time at the current moment as target input parameters, the method further includes: using sensors to collect the current operating parameters, current environmental parameters and current harmful gas emission concentration of the plasma stove in real time, and performing denoising, interpolation, filling and standardization on the current operating parameters, the current environmental parameters and the current harmful gas emission concentration respectively.

[0031] In a second aspect, the present invention provides a plasma stove harmful gas control device, the device comprising:

[0032] A real-time monitoring module is configured to use the current operating parameters, current environmental parameters, and current harmful gas emission concentration of the plasma stove acquired in real time at the current moment as target input parameters, and input the target input parameters into an emission prediction model trained using a historical time series dataset; the historical time series dataset includes the historical operating parameters, historical environmental parameters, and historical harmful gas emission concentrations of the plasma stove acquired at multiple historical moments, as well as the actual harmful gas emission concentration corresponding to the next moment after each historical moment;

[0033] An emission prediction module, configured to process the target input parameters using the emission prediction model to obtain a predicted emission concentration of the harmful gas at the next moment;

[0034] an optimization control module, configured to use the current operating parameters, the current environmental parameters, and the predicted emission concentration as input state data, process the input state data based on a deep Q-learning network, obtain operating parameters adjusted by an action adjustment mechanism that minimizes the predicted emission concentration, and use the operating parameters as optimal parameters; the action adjustment mechanism includes individual adjustment of each of the current operating parameters, as well as simultaneous adjustment of multiple parameters;

[0035] A feedback regulation module is used to control the plasma stove to operate at the optimal parameters.

[0036] In one possible implementation, when the current operating parameters include the operating voltage, operating current, air flow rate, and electrode spacing of the plasma stove at the current moment, the current environmental parameters include the air pressure and humidity around the plasma stove at the current moment, the current harmful gas emission concentration includes the nitrogen oxide emission concentration and carbon monoxide emission concentration of the plasma stove at the current moment, and the emission prediction model is an LSTM multi-task prediction model based on a spatiotemporal hybrid attention mechanism, the emission prediction module is specifically configured to execute:

[0037] The target input parameters are processed by an LSTM multi-task prediction model based on a spatiotemporal mixed attention mechanism to obtain the predicted nitrogen oxide emission concentration and carbon monoxide emission concentration of the plasma stove at the next moment.

[0038] In one possible implementation, the plasma stove harmful gas control device further includes a model construction module, within which the LSTM multi-task prediction model based on the spatiotemporal hybrid attention mechanism is configured to include an input layer, a temporal attention layer, a spatial attention layer, a feature fusion layer, an LSTM shared feature layer, a multi-task prediction head, and an output layer connected in sequence;

[0039] The input layer is used to receive the historical time series dataset;

[0040] The time attention layer is used to determine the importance of a time step according to the changing trend of the actual harmful gas emission concentration in the historical time series data set, and to perform weighted processing on the sample data in the historical time series data set according to the importance of the time step to obtain weighted time series data;

[0041] The spatial attention mechanism is used to perform weighted processing on the historical working parameters according to the importance of each type of working parameters in the historical working parameters to obtain weighted working parameter data;

[0042] The feature fusion layer is used to fuse the time series data and the working parameter data to obtain feature fusion data;

[0043] The LSTM shared feature layer is used to learn the temporal dependency and the shared feature layer from the feature fusion data through multiple LSTM units to extract shared features;

[0044] The multi-task prediction head includes a nitrogen oxide prediction head and a carbon monoxide prediction head, wherein the nitrogen oxide prediction head predicts the predicted emission concentration of nitrogen oxide at the next moment through a first fully connected layer, and the carbon monoxide prediction head predicts the predicted emission concentration of carbon monoxide at the next moment through a second fully connected layer;

[0045] The output layer is used to output the predicted emission concentration of nitrogen oxides and the predicted emission concentration of carbon monoxide.

[0046] In one possible implementation, when processing the input state data based on the deep Q-learning network to obtain operating parameters adjusted by the action adjustment mechanism that minimize the predicted emission concentration, and using the operating parameters as optimal parameters, the optimization control module is specifically configured to execute:

[0047] The Q value after executing each action adjustment mechanism on the input state data is calculated through a deep learning network, the action adjustment mechanism with the largest Q value is used as the action adjustment mechanism that minimizes the predicted emission concentration, and the working parameters adjusted by the action adjustment mechanism with the largest Q value are output as the optimal parameters; the Q value is determined by the immediate reward of the current action adjustment mechanism and the expected value of the future reward obtained due to the state transfer caused by the action adjustment mechanism.

[0048] In a possible implementation, within the optimization control module, the Q value in the deep learning network is specifically configured to be calculated using the following formula:

[0049]

[0050] Where s represents the current environment state corresponding to the input state data, a represents the action adjustment mechanism taken under the current environment state, Q(s, a) represents the total return expected to be obtained by taking action adjustment mechanism a under the current environment state, Rt(a) represents the immediate reward obtained when taking action adjustment mechanism a, and γ represents the discount factor. Indicates that the Q value of the action adjustment mechanism that maximizes the Q value is selected from all possible action adjustment mechanisms a1 in the next environment state s1 as the expected value of the future reward;

[0051] The immediate reward is determined by a predefined reward function in a deep Q-learning network; the reward function is used to measure the size of the predicted emission concentration after executing the action adjustment mechanism, wherein the predicted emission concentration is inversely proportional to the immediate reward.

[0052] In a possible implementation, within the optimization control module, the reward function in the deep learning network is specifically configured as follows:

[0053] Rt=w1·(-NOx t -CO t )+w2·Efficiency t -w3·Concentration_Pate_Chatge t ;

[0054] Among them, Rt represents the reward function, NOx t and CO t They represent the predicted emission concentrations of nitrogen oxides and carbon monoxide at time t, respectively. t Indicates the discharge efficiency corresponding to the input state data, Concentration_Rate_Change t It represents the changing rate of harmful gas concentration, w1, w2 and w3 represent weighting factors respectively.

[0055] In one possible implementation, the plasma stove harmful gas control device further includes a data preprocessing module. Before using the current operating parameters, current environmental parameters, and current harmful gas emission concentration of the plasma stove acquired in real time at the current moment as target input parameters, the data preprocessing module is configured to perform:

[0056] The current working parameters, current environmental parameters and current harmful gas emission concentration of the plasma stove are collected in real time by using sensors, and the current working parameters, current environmental parameters and current harmful gas emission concentration are denoised, interpolated, filled and standardized respectively.

[0057] In a third aspect, the present invention provides an electronic device comprising: a processor and a memory. The memory stores at least one instruction, at least one program, a code set, or an instruction set, wherein the at least one instruction, at least one program, or the code set or instruction set is loaded and executed by the processor to implement any of the aforementioned methods for controlling harmful gases in a plasma stove.

[0058] In a fourth aspect, the present invention provides a computer-readable storage medium, which stores at least one instruction, at least one program, a code set or an instruction set. The at least one instruction, the at least one program, the code set or the instruction set is loaded and executed by a processor to implement any of the above-mentioned methods for controlling harmful gases in a plasma stove.

[0059] In actual application, the plasma stove harmful gas control method provided by the embodiment of the present invention firstly inputs the current working parameters, current environmental parameters and current harmful gas emission concentration of the plasma stove obtained in real time into the emission prediction model as target input parameters; secondly, the target input parameters are processed by the emission prediction model to obtain the predicted emission concentration of the harmful gas at the next moment; thirdly, the current working parameters, current environmental parameters and predicted emission concentration are used as state input data, the state input data are processed by a deep Q learning network, and the working parameters adjusted by the action adjustment mechanism that minimizes the predicted emission concentration are used as the optimal parameters; finally, the plasma stove is controlled to operate with the optimal parameters; the present invention can dynamically adjust the working parameters of the plasma stove according to the collected real-time data to ensure that the emission of harmful gases is minimized, so as to achieve the goals of green environmental protection and low emissions; while maintaining the efficient discharge performance of the plasma stove. BRIEF DESCRIPTION OF THE DRAWINGS

[0060] Figure 1 A flowchart of a method for controlling harmful gases in a plasma stove provided by an embodiment of the present invention;

[0061] Figure 2 This is a structural block diagram of a plasma stove harmful gas control device provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0062] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. All other embodiments obtained by ordinary technicians in this field based on the embodiments of the present invention without making any creative efforts shall fall within the scope of protection of the present invention.

[0063] In the following, the terms "first" and "second" are used for descriptive purposes only and are not to be understood as indicating or implying relative importance or implicitly indicating the number of the indicated technical features. Thus, the features defined as "first" and "second" may explicitly or implicitly include one or more of the features. In the description of the embodiments of the present disclosure, unless otherwise specified, "multiple" means two or more. In addition, the use of "based on" or "according to" means openness and inclusiveness, because the process, steps, calculations or other actions "based on" or "according to" one or more of the conditions or values may be based on additional conditions or values beyond the stated in practice.

[0064] The embodiments of the present invention provide a plasma stove harmful gas control method, device, equipment and storage medium, which can dynamically adjust the operating parameters of the plasma stove according to the collected real-time data to ensure that the emission of harmful gases is minimized.

[0065] like Figure 1 As shown, in the first aspect, an embodiment of the present invention further provides a method for controlling harmful gases in a plasma stove, the method comprising:

[0066] Step 101: Using the current working parameters, current environmental parameters and current harmful gas emission concentration of the plasma stove acquired in real time at the current moment as target input parameters, the target input parameters are input into an emission prediction model trained using a historical time series dataset.

[0067] The historical time series data set includes the historical operating parameters, historical environmental parameters, and historical harmful gas emission concentrations of the plasma stove obtained at multiple historical moments, as well as the actual harmful gas emission concentration corresponding to the next moment of each historical moment.

[0068] Specifically, the current operating parameters of the plasma stove refer to the operating parameters of the plasma stove collected in real time at the current moment, including data such as power, frequency, gas flow, pressure, electric field strength, and temperature. The current environmental parameters of the plasma stove refer to the environmental parameters of the plasma stove's surroundings collected in real time at the current moment, including data such as air pressure, humidity, light, and ventilation. The current harmful gas emission concentration of the plasma stove refers to the concentration of various harmful gases generated by the plasma stove during operation, including data such as the concentration of nitrogen oxides, carbon monoxide, and ozone.

[0069] A historical time series dataset refers to a collection of time series data collected during a historical period. Time series data refers to data collected at different times and is used to describe how a phenomenon changes over time. In other words, a historical time series dataset consists of operating parameters, environmental parameters, and harmful gas emission concentrations collected at different times, as well as the actual harmful gas emission concentrations at the next historical moment.

[0070] The actual harmful gas emission concentration at the next moment refers to the harmful gas emission concentration value actually measured at the next moment when the plasma stove maintains the operating state at the historical moment.

[0071] Step 102: Process the target input parameters through the emission prediction model to obtain the predicted emission concentration of the harmful gas at the next moment.

[0072] Specifically, the emission prediction model can output the predicted emission concentration of harmful gases at the next moment when the plasma stove maintains the current operating parameters based on the input current operating parameters, current environmental parameters and current harmful gas emission concentration.

[0073] Step 103: Using the current working parameters, current environmental parameters, and predicted emission concentration as input state data, the input state data is processed based on the deep Q learning network to obtain the working parameters adjusted by the action adjustment mechanism that minimizes the predicted emission concentration, and the working parameters are used as the optimal parameters.

[0074] The action adjustment mechanism includes individual adjustment of each parameter in the current working parameters, as well as simultaneous adjustment of multiple parameters.

[0075] The Deep Q Network (DQN) is an algorithm that combines deep learning with Q-learning and is primarily used to handle problems with high-dimensional state spaces. DQN uses a neural network as an approximator of the Q function and iteratively updates the Q function to find the optimal policy.

[0076] In an embodiment of the present invention, a deep Q-learning network includes an input layer, a spatial attention layer, a hidden layer, and an output layer. The input layer is used to receive input state data consisting of current operating parameters, current environmental parameters, and predicted emission concentrations; the spatial attention layer is used to determine the weighting based on the importance of the input state data received by the input layer and output the weighted operating parameter data; the hidden layer includes multiple fully connected layers, which are used to extract key features of the weighted operating parameter data through complex mathematical operations to calculate the Q value corresponding to each action adjustment mechanism after adjusting the input state data, and select the operating parameter adjusted by the action adjustment mechanism with the largest Q value as the optimal parameter output.

[0077] The Q value describes the expected reward from the plasma stove's current operating parameters to the parameters adjusted by the action adjustment mechanism. A larger Q value indicates the lowest predicted emission concentration when the plasma stove operates according to the adjusted operating parameters.

[0078] Step 104: Control the plasma stove to operate with optimal parameters.

[0079] The operation of the plasma stove is controlled according to the selected optimal parameters, so that the high efficiency performance of the plasma stove can be maintained during the operation of the plasma stove, and the concentration of harmful gases generated during the operation of the plasma stove can be minimized.

[0080] Through the above steps 101 to 104, the operating state of the plasma stove is monitored in real time and dynamically adjusted, thereby achieving the green, environmentally friendly and low-emission operating goals of the plasma stove.

[0081] In actual application of the plasma stove harmful gas control method provided by an embodiment of the present invention, first, the current working parameters, current environmental parameters and current harmful gas emission concentration of the plasma stove obtained in real time are input into the emission prediction model as target input parameters; secondly, the target input parameters are processed by the emission prediction model to obtain the predicted emission concentration of the harmful gas at the next moment; thirdly, the current working parameters, current environmental parameters and predicted emission concentration are used as state input data, and the state input data are processed by a deep Q learning network, and the working parameters adjusted by the action adjustment mechanism that minimizes the predicted emission concentration are used as optimal parameters; finally, the plasma stove is controlled to operate with the optimal parameters.

[0082] The present invention can dynamically adjust the working parameters of the plasma stove according to the collected real-time data, ensuring that the emission of harmful gases is minimized to achieve the goals of green environmental protection and low emissions; while maintaining the efficient discharge performance of the plasma stove.

[0083] Furthermore, the current operating parameters include the operating voltage, operating current, air flow rate and electrode distance of the plasma stove at the current moment.

[0084] The current environmental parameters include the air pressure and humidity around the plasma cooker at the current moment.

[0085] The current harmful gas emission concentration includes the nitrogen oxide emission concentration and carbon monoxide emission concentration of the plasma stove at the current moment.

[0086] Among them, the current working parameters, current environmental parameters and current harmful gas emission concentrations are all monitored in real time through high-precision sensors.

[0087] The emission prediction model is an LSTM multi-task prediction model based on the spatiotemporal hybrid attention mechanism.

[0088] At this time, the target input parameters are processed by the emission prediction model to obtain the predicted emission concentration of harmful gases at the next moment, specifically:

[0089] The target input parameters are processed through the LSTM multi-task prediction model based on the spatiotemporal mixed attention mechanism to obtain the predicted nitrogen oxide emission concentration and carbon monoxide emission concentration of the plasma stove at the next moment.

[0090] Specifically, LSTM combined with the attention mechanism can dynamically focus on the key information in the target input parameters, thereby improving the performance of the model when processing long sequences.

[0091] In the present invention, the working voltage, working current, air flow rate, electrode spacing, air pressure, humidity, nitrogen oxide emission concentration and carbon monoxide emission concentration of the plasma stove obtained in real time at the current moment are used as input, and the real-time data is processed by an LSTM multi-task prediction model based on the spatiotemporal mixed attention mechanism to output the predicted nitrogen oxide emission concentration and predicted carbon monoxide emission concentration at the next moment.

[0092] Furthermore, the LSTM multi-task prediction model based on the spatiotemporal hybrid attention mechanism includes an input layer, a temporal attention layer, a spatial attention layer, a feature fusion layer, an LSTM shared feature layer, a multi-task prediction head, and an output layer connected in sequence.

[0093] The input layer is used to receive historical time series datasets;

[0094] The temporal attention layer is used to determine the importance of the time step based on the changing trend of the real harmful gas emission concentration in the historical time series dataset, and to weight the sample data in the historical time series dataset according to the importance of the time step to obtain the weighted time series data.

[0095] In other words, the time weight is calculated through the attention mechanism to emphasize the time points that have an important impact on the prediction process, such as the peak periods of nitrogen oxide concentration and carbon monoxide concentration.

[0096] The spatial attention mechanism is used to weight the historical working parameters according to the importance of each type of working parameters in the historical working parameters to obtain weighted working parameter data.

[0097] The feature fusion layer is used to fuse time series data and working parameter data to obtain feature fusion data.

[0098] The LSTM shared feature layer is used to learn temporal dependencies and shared feature layers from feature fusion data through multiple LSTM units to extract shared features.

[0099] The multi-task prediction head includes a nitrogen oxide prediction head and a carbon monoxide prediction head. The nitrogen oxide prediction head predicts the predicted emission concentration of nitrogen oxide at the next moment through the first fully connected layer, and the carbon monoxide prediction head predicts the predicted emission concentration of carbon monoxide at the next moment through the second fully connected layer.

[0100] The output layer is used to output the predicted emission concentration of nitrogen oxides and the predicted emission concentration of carbon monoxide.

[0101] Furthermore, before using the current operating parameters, current environmental parameters, and current harmful gas emission concentration of the plasma stove acquired in real time at the current moment as target input parameters, the method further includes:

[0102] Sensors are used to collect the current working parameters, current environmental parameters and current harmful gas emission concentrations of the plasma stove in real time, and the current working parameters, current environmental parameters and current harmful gas emission concentrations are denoised, interpolated, filled and standardized respectively.

[0103] Denoising refers to removing noise from collected sensor data using a noise reduction autoencoder. Interpolation and filling refer to addressing data sparsity by using interpolation methods to fill in missing data. Standardization refers to normalizing input data to ensure consistent dimensions.

[0104] Furthermore, the input state data is processed based on the deep Q-learning network to obtain the working parameters adjusted by the action adjustment mechanism to minimize the predicted emission concentration, and the working parameters are taken as the optimal parameters, specifically including:

[0105] The Q value after executing each action adjustment mechanism based on the input state data is calculated through a deep learning network. The action adjustment mechanism with the largest Q value is used as the action adjustment mechanism that minimizes the predicted emission concentration, and the working parameters adjusted by the action adjustment mechanism with the largest Q value are output as the optimal parameters.

[0106] Among them, the Q value is determined by the immediate reward of the current action adjustment mechanism and the expected value of the future reward obtained due to the state transition caused by the action adjustment mechanism.

[0107] Furthermore, the Q value is calculated by the following formula:

[0108]

[0109] Where s represents the current environment state corresponding to the input state data, a represents the action adjustment mechanism taken under the current environment state, Q(s, a) represents the total return expected to be obtained by taking action adjustment mechanism a under the current environment state, Rt(a) represents the immediate reward obtained when taking action adjustment mechanism a, and γ represents the discount factor. It means that the Q value of the action adjustment mechanism that maximizes the Q value is selected from all possible action adjustment mechanisms a1 in the next environment state s1 as the expected value of the future reward.

[0110] The immediate reward is determined by a predefined reward function in the deep Q-learning network.

[0111] The reward function is used to measure the size of the predicted emission concentration after executing the action adjustment mechanism, where the predicted emission concentration is inversely proportional to the immediate reward.

[0112] In an embodiment of the present invention, the input of the deep Q learning network includes the working voltage, working current, air flow rate and electrode spacing of the plasma stove at the current moment, the humidity and air pressure at the current moment, and the predicted nitrogen oxide emission concentration and carbon monoxide emission concentration for the target input parameters output by the emission prediction model.

[0113] The operating parameters that can be adjusted by the action adjustment mechanism are defined in the workspace of the deep Q-learning network. These adjustable operating parameters include the plasma cooker's current operating voltage, current, air flow rate, and electrode spacing. The operating voltage, current, and air flow rate each have three settings: low, medium, and high, while the electrode spacing has three settings: near, medium, and far. The action adjustment mechanism adjusts the operating settings corresponding to the plasma cooker's current operating voltage, current, air flow rate, and electrode spacing.

[0114] The action adjustment mechanism selects actions based on the epsilon-greedy strategy.

[0115] Execute the action selected by the action adjustment mechanism and calculate the reward based on the state of the plasma stove before and after executing the action.

[0116] When designing the reward function, it is necessary to comprehensively consider the harmful gas emissions, the discharge efficiency of the plasma stove, and the smoothness of the rate of change of the harmful gas concentration.

[0117] Specifically, nitrogen oxide and carbon monoxide emissions are the core of the reward function. Specifically, the lower the emissions, the higher the reward.

[0118] When the concentrations of nitrogen oxides and carbon monoxide emissions exceed preset thresholds, negative rewards are set to penalize the plasma stove's hazardous gas control system for non-compliant behavior. For example, to prevent rapid adjustments from causing instability in the plasma stove's hazardous gas control system, a "smoothness" reward is added to the reward function to penalize the rate of change of hazardous gas emission concentrations.

[0119] Adjust the weights of each parameter in the reward function based on real-time environmental factors. For example, when nitrogen oxides are relatively easy to form, the weight of nitrogen oxides is increased, thereby strengthening the penalty for nitrogen oxides.

[0120] Furthermore, when the predicted emission concentration includes the predicted emission concentration of nitrogen oxides and the predicted emission concentration of carbon monoxide, the reward function is specifically:

[0121] Rt=w1·(-NOx t -CO t )+w2·Efficiency t -w3·Concentration_Rate_Change t ;

[0122] Among them, Rt represents the reward function, NOx t and CO t They represent the predicted emission concentrations of nitrogen oxides and carbon monoxide at time t, respectively. t Indicates the discharge efficiency corresponding to the input state data, Concentration_Rate_Change t It represents the changing rate of harmful gas concentration, w1, w2 and w3 represent weighting factors respectively.

[0123] This paper optimizes the LSTM model's processing of target input parameters based on a spatiotemporal hybrid attention mechanism, improving the model's focus on key time nodes and key features. Through feature fusion, the model simultaneously considers the impact of key time nodes and key features on harmful gas emissions during plasma stove operation.

[0124] The emission prediction model of the present invention not only enhances the adaptability of the model to different environmental conditions, but also can simultaneously predict the emission concentrations of multiple harmful gases.

[0125] The present invention combines reinforcement learning technology with the LSTM model, and uses the emission prediction model to predict the emission concentration of the plasma stove at the next moment as well as the current operating parameters and current environmental parameters of the plasma stove to guide the deep Q learning network to optimize the operating parameters of the plasma stove.

[0126] The present invention sets a dynamic reward and penalty mechanism based on the real-time acquired harmful gas emission concentration, plasma stove discharge efficiency, and the rate of change of harmful gas concentration; adjusts the weight of each parameter in the reward function according to environmental parameters, and enhances the adaptability of the plasma stove harmful gas control method of the present invention to harmful gas emission control in different environments.

[0127] like Figure 2 As shown, in a second aspect, an embodiment of the present invention further provides a plasma stove harmful gas control device, the device comprising:

[0128] The real-time monitoring module 201 is used to use the current working parameters, current environmental parameters and current harmful gas emission concentration of the plasma stove obtained in real time at the current moment as target input parameters, and input the target input parameters into the emission prediction model trained by the historical time series data set.

[0129] The historical time series data set includes the historical operating parameters, historical environmental parameters, and historical harmful gas emission concentrations of the plasma stove obtained at multiple historical moments, as well as the actual harmful gas emission concentration corresponding to the next moment of each historical moment.

[0130] The emission prediction module 202 is used to process the target input parameters through the emission prediction model to obtain the predicted emission concentration of the harmful gas at the next moment.

[0131] The optimization control module 203 is used to take the current working parameters, current environmental parameters and predicted emission concentration as input state data, process the input state data based on the deep Q learning network, obtain the working parameters adjusted by the action adjustment mechanism to minimize the predicted emission concentration, and use the working parameters as the optimal parameters.

[0132] The action adjustment mechanism includes individual adjustment of each parameter in the current working parameters, as well as simultaneous adjustment of multiple parameters.

[0133] The feedback regulation module 204 is used to control the plasma stove to operate at optimal parameters.

[0134] In one possible implementation, the current working parameters include the working voltage, working current, air flow rate and electrode spacing of the plasma stove at the current moment, the current environmental parameters include the air pressure and humidity around the plasma stove at the current moment, the current harmful gas emission concentration includes the nitrogen oxide emission concentration and carbon monoxide emission concentration of the plasma stove at the current moment, and the emission prediction model is an LSTM multi-task prediction model based on the spatiotemporal mixed attention mechanism.

[0135] The emission prediction module 202 is specifically configured to perform:

[0136] The target input parameters are processed through the LSTM multi-task prediction model based on the spatiotemporal mixed attention mechanism to obtain the predicted nitrogen oxide emission concentration and carbon monoxide emission concentration of the plasma stove at the next moment.

[0137] In one possible implementation, the plasma stove harmful gas control device also includes a model construction module. Within the model construction module, the LSTM multi-task prediction model based on the spatiotemporal hybrid attention mechanism is configured to include an input layer, a temporal attention layer, a spatial attention layer, a feature fusion layer, an LSTM shared feature layer, a multi-task prediction head and an output layer connected in sequence.

[0138] Among them, the input layer is used to receive historical time series datasets.

[0139] The temporal attention layer is used to determine the importance of the time step based on the changing trend of the real harmful gas emission concentration in the historical time series dataset, and to weight the sample data in the historical time series dataset according to the importance of the time step to obtain the weighted time series data.

[0140] The spatial attention mechanism is used to weight the historical working parameters according to the importance of each type of working parameters in the historical working parameters to obtain weighted working parameter data.

[0141] The feature fusion layer is used to fuse time series data and working parameter data to obtain feature fusion data.

[0142] The LSTM shared feature layer is used to learn temporal dependencies and shared feature layers from feature fusion data through multiple LSTM units to extract shared features.

[0143] The multi-task prediction head includes a nitrogen oxide prediction head and a carbon monoxide prediction head. The nitrogen oxide prediction head predicts the predicted emission concentration of nitrogen oxide at the next moment through the first fully connected layer, and the carbon monoxide prediction head predicts the predicted emission concentration of carbon monoxide at the next moment through the second fully connected layer.

[0144] The output layer is used to output the predicted emission concentration of nitrogen oxides and the predicted emission concentration of carbon monoxide.

[0145] In one possible implementation, when processing the input state data based on the deep Q-learning network to obtain the operating parameters adjusted by the action adjustment mechanism that minimize the predicted emission concentration, and using the operating parameters as the optimal parameters, the optimization control module 203 is specifically configured to execute:

[0146] The Q value after executing each action adjustment mechanism based on the input state data is calculated through a deep learning network. The action adjustment mechanism with the largest Q value is used as the action adjustment mechanism that minimizes the predicted emission concentration, and the working parameters adjusted by the action adjustment mechanism with the largest Q value are output as the optimal parameters.

[0147] Among them, the Q value is determined by the immediate reward of the current action adjustment mechanism and the expected value of the future reward obtained due to the state transition caused by the action adjustment mechanism.

[0148] In one possible implementation, within the optimization control module 203, the Q value in the deep learning network is specifically configured to be calculated using the following formula:

[0149]

[0150] Where s represents the current environment state corresponding to the input state data, a represents the action adjustment mechanism taken under the current environment state, Q(s, a) represents the total return expected to be obtained by taking action adjustment mechanism a under the current environment state, Rt(a) represents the immediate reward obtained when taking action adjustment mechanism a, and γ represents the discount factor. It means that the Q value of the action adjustment mechanism that maximizes the Q value is selected from all possible action adjustment mechanisms a1 in the next environment state s1 as the expected value of the future reward.

[0151] The immediate reward is determined by a predefined reward function in the deep Q-learning network; the reward function is used to measure the size of the predicted emission concentration after executing the action adjustment mechanism, where the predicted emission concentration is inversely proportional to the immediate reward.

[0152] In one possible implementation, within the optimization control module 203, the reward function in the deep learning network is specifically configured as follows:

[0153] Rt=w1·(-NOx t -CO t )+w2·Efficiency t -w3·Concentration_Rate_Changet;

[0154] Among them, Rt represents the reward function, NOx t and CO t They represent the predicted emission concentrations of nitrogen oxides and carbon monoxide at time t, respectively. t Indicates the discharge efficiency corresponding to the input state data, Concentration_Rate_Change t It represents the changing rate of harmful gas concentration, w1, w2 and w3 represent weighting factors respectively.

[0155] In one possible implementation, the plasma stove harmful gas control device further includes a data preprocessing module. Before using the current operating parameters, current environmental parameters, and current harmful gas emission concentration of the plasma stove acquired in real time at the current moment as target input parameters, the data preprocessing module is configured to perform:

[0156] Sensors are used to collect the current working parameters, current environmental parameters and current harmful gas emission concentrations of the plasma stove in real time, and the current working parameters, current environmental parameters and current harmful gas emission concentrations are denoised, interpolated, filled and standardized respectively.

[0157] The plasma stove harmful gas control device provided by the embodiment of the present invention is used to execute the above-mentioned plasma stove harmful gas control method, and thus can achieve the same effect as the above-mentioned plasma stove harmful gas control method.

[0158] Those skilled in the art will clearly understand that, for the convenience and brevity of description, the specific working processes of the systems, devices and units described above can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0159] In a third aspect, an embodiment of the present invention further provides an electronic device, comprising a processor and a memory, wherein the memory stores at least one instruction, at least one program, a code set, or an instruction set, and the at least one instruction, at least one program, a code set, or an instruction set is loaded and executed by the processor to implement the plasma stove harmful gas control method in an embodiment of the present invention.

[0160] In a fourth aspect, an embodiment of the present invention further provides a computer-readable storage medium, in which at least one instruction, at least one program, code set or instruction set is stored, and the at least one instruction, at least one program, code set or instruction set is loaded and executed by a processor to implement the plasma stove harmful gas control method in an embodiment of the present invention.

[0161] In the above embodiments, all or part of the embodiments can be implemented by software, hardware, firmware, or any combination thereof. When implemented using software, all or part of the embodiments can be implemented in the form of a computer program product. The computer program product includes one or more computer instructions. When the computer program instructions are loaded and executed on a computer, all or part of the processes or functions according to the embodiments of the present invention are generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another computer-readable storage medium. For example, the computer instructions can be transmitted from one website, computer, server, or data center to another website, computer, server, or data center via a wired (e.g., coaxial cable, optical fiber, digital subscriber line (DSL)) or wireless (e.g., infrared, wireless, microwave, etc.) method. The computer-readable storage medium can be any available medium that can be accessed by a computer or a data storage device such as a server or data center that includes one or more available media. The available medium can be a magnetic medium (e.g., a floppy disk, a hard disk, a magnetic tape), an optical medium (e.g., a DVD), or a semiconductor medium (e.g., a solid-state drive (SSD)).

[0162] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any changes or substitutions within the technical scope disclosed by the present invention shall be covered by the scope of protection of the present invention. Therefore, the scope of protection of the present invention shall be subject to the scope of protection of the claims.

Claims

1. A method for controlling harmful gases in a plasma stove, characterized in that: include: Using the current operating parameters, current environmental parameters, and current harmful gas emission concentration of the plasma stove acquired in real time at the current moment as target input parameters, the target input parameters are input into an emission prediction model trained using a historical time series dataset; the historical time series dataset includes the historical operating parameters, historical environmental parameters, and historical harmful gas emission concentrations of the plasma stove acquired at multiple historical moments, as well as the actual harmful gas emission concentration corresponding to the next moment of each historical moment; Processing the target input parameters through the emission prediction model to obtain the predicted emission concentration of the harmful gas at the next moment; The current operating parameters, the current environmental parameters, and the predicted emission concentration are used as input state data, and the input state data is processed based on a deep Q-learning network to obtain the operating parameters adjusted by an action adjustment mechanism that minimizes the predicted emission concentration, and the operating parameters are used as optimal parameters; the action adjustment mechanism includes individual adjustment of each of the current operating parameters and simultaneous adjustment of multiple parameters; The plasma cooker is controlled to operate with the optimal parameters.

2. The plasma stove harmful gas control method according to claim 1, characterized in that: The current operating parameters include the operating voltage, operating current, air flow rate, and electrode spacing of the plasma stove at the current moment; the current environmental parameters include the air pressure and humidity around the plasma stove at the current moment; the current harmful gas emission concentration includes the nitrogen oxide emission concentration and carbon monoxide emission concentration of the plasma stove at the current moment; the emission prediction model is an LSTM multi-task prediction model based on a spatiotemporal hybrid attention mechanism; the target input parameters are processed by the emission prediction model to obtain the predicted emission concentration of the harmful gas at the next moment, specifically: The target input parameters are processed by an LSTM multi-task prediction model based on a spatiotemporal mixed attention mechanism to obtain the predicted nitrogen oxide emission concentration and carbon monoxide emission concentration of the plasma stove at the next moment.

3. The plasma stove harmful gas control method according to claim 2, characterized in that: The LSTM multi-task prediction model based on the spatiotemporal hybrid attention mechanism includes an input layer, a temporal attention layer, a spatial attention layer, a feature fusion layer, an LSTM shared feature layer, a multi-task prediction head, and an output layer connected in sequence; The input layer is used to receive the historical time series dataset; The time attention layer is used to determine the importance of a time step according to the changing trend of the actual harmful gas emission concentration in the historical time series data set, and to perform weighted processing on the sample data in the historical time series data set according to the importance of the time step to obtain weighted time series data; The spatial attention mechanism is used to perform weighted processing on the historical working parameters according to the importance of each type of working parameters in the historical working parameters to obtain weighted working parameter data; The feature fusion layer is used to fuse the time series data and the working parameter data to obtain feature fusion data; The LSTM shared feature layer is used to learn the temporal dependency and the shared feature layer from the feature fusion data through multiple LSTM units to extract shared features; The multi-task prediction head includes a nitrogen oxide prediction head and a carbon monoxide prediction head, wherein the nitrogen oxide prediction head predicts the predicted emission concentration of nitrogen oxide at the next moment through a first fully connected layer, and the carbon monoxide prediction head predicts the predicted emission concentration of carbon monoxide at the next moment through a second fully connected layer; The output layer is used to output the predicted emission concentration of nitrogen oxides and the predicted emission concentration of carbon monoxide.

4. The plasma stove harmful gas control method according to claim 1, characterized in that: The input state data is processed based on a deep Q-learning network to obtain operating parameters adjusted by the action adjustment mechanism that minimize the predicted emission concentration, and the operating parameters are used as optimal parameters, specifically including: The Q value after executing each action adjustment mechanism on the input state data is calculated through a deep learning network, the action adjustment mechanism with the largest Q value is used as the action adjustment mechanism that minimizes the predicted emission concentration, and the working parameters adjusted by the action adjustment mechanism with the largest Q value are output as the optimal parameters; the Q value is determined by the immediate reward of the current action adjustment mechanism and the expected value of the future reward obtained due to the state transfer caused by the action adjustment mechanism.

5. The plasma stove harmful gas control method according to claim 4, characterized in that: The Q value is specifically calculated by the following formula: Where s represents the current environment state corresponding to the input state data, a represents the action adjustment mechanism taken under the current environment state, Q(s, a) represents the total return expected to be obtained by taking action adjustment mechanism a under the current environment state, Rt(a) represents the immediate reward obtained when taking action adjustment mechanism a, and γ represents the discount factor. Indicates that the Q value of the action adjustment mechanism that maximizes the Q value is selected from all possible action adjustment mechanisms a1 in the next environment state s1 as the expected value of the future reward; The immediate reward is determined by a predefined reward function in a deep Q-learning network; the reward function is used to measure the size of the predicted emission concentration after executing the action adjustment mechanism, wherein the predicted emission concentration is inversely proportional to the immediate reward.

6. The plasma stove harmful gas control method according to claim 5, characterized in that: The predicted emission concentration includes the predicted emission concentration of nitrogen oxides and the predicted emission concentration of carbon monoxide; the reward function is specifically: Rt=w1·(-NOx t -CO t )+w2·Efficiency t -w3·Concentration_Rate_Change t ; Among them, Rt represents the reward function, NOx t and CO t They represent the predicted emission concentrations of nitrogen oxides and carbon monoxide at time t, respectively. t Indicates the discharge efficiency corresponding to the input state data, Concentration_Rate_Change t It represents the changing rate of harmful gas concentration, w1, w2 and w3 represent weighting factors respectively.

7. The plasma stove harmful gas control method according to claim 1, characterized in that: Before using the current working parameters, current environmental parameters and current harmful gas emission concentration of the plasma stove obtained in real time at the current moment as target input parameters, the method also includes: using sensors to collect the current working parameters, current environmental parameters and current harmful gas emission concentration of the plasma stove in real time, and denoising, interpolating, filling and standardizing the current working parameters, the current environmental parameters and the current harmful gas emission concentration respectively.

8. A plasma stove harmful gas control device, characterized in that: include: A real-time monitoring module is configured to use the current operating parameters, current environmental parameters, and current harmful gas emission concentration of the plasma stove acquired in real time at the current moment as target input parameters, and input the target input parameters into an emission prediction model trained using a historical time series dataset; the historical time series dataset includes the historical operating parameters, historical environmental parameters, and historical harmful gas emission concentrations of the plasma stove acquired at multiple historical moments, as well as the actual harmful gas emission concentration corresponding to the next moment after each historical moment; An emission prediction module, configured to process the target input parameters using the emission prediction model to obtain a predicted emission concentration of the harmful gas at the next moment; an optimization control module, configured to use the current operating parameters, the current environmental parameters, and the predicted emission concentration as input state data, process the input state data based on a deep Q-learning network, obtain operating parameters adjusted by an action adjustment mechanism that minimizes the predicted emission concentration, and use the operating parameters as optimal parameters; the action adjustment mechanism includes individual adjustment of each of the current operating parameters, as well as simultaneous adjustment of multiple parameters; A feedback regulation module is used to control the plasma stove to operate at the optimal parameters.

9. An electronic device, characterized in that: The electronic device includes a processor and a memory, wherein the memory stores at least one instruction, at least one program, a code set, or an instruction set, and the at least one instruction, at least one program, the code set, or the instruction set is loaded and executed by the processor to implement the plasma stove harmful gas control method according to any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that The storage medium stores at least one instruction, at least one program, a code set, or an instruction set, and the at least one instruction, the at least one program, the code set, or the instruction set is loaded and executed by a processor to implement the plasma stove harmful gas control method according to any one of claims 1 to 7.

Images