Control method, device and equipment of boiler system and medium

Abstract

The invention discloses a control method, device and equipment of a boiler system and a medium, and relates to the technical field of plasmas. According to the scheme, the current operation state data of a boiler can be analyzed in real time and the current boiler control parameters can be automatically adjusted by combining a preset deep neural network, the high self-adaptive capacity is achieved, and the control efficiency is improved. The control strategy can be immediately optimized according to the change of the boiler state; besides, because the preset deep neural network can process a more complex nonlinear relationship, more accurate prediction can be made according to the multi-modal current operation state data, thereby improving the control precision.

Description

Technical Field

[0001] This application relates to the field of plasma technology, and particularly to a control method, device, equipment, and medium for a boiler system. Background Art

[0002] With the development of industrial automation technology, as an important device in the industrial field, the intelligence and automation level of the control system of the boiler system directly affect the operation efficiency, energy consumption, and environmental protection of the boiler. Traditional boiler control systems mostly rely on rule-based control methods. When facing a complex and dynamically changing operating environment, it is difficult to ensure the optimal operating state of the boiler system, and the adjustment process is usually relatively lagging, easily leading to problems such as slow response speed, energy waste, and equipment failures.

[0003] In recent years, with the development of deep learning and artificial intelligence technologies, especially the application of deep neural networks in multi-modal data processing, a new solution has been provided. However, currently, most boiler control systems still adopt traditional proportional-integral-derivative control or simple machine learning models, lacking deep learning applications for multi-modal data and being difficult to dynamically adjust according to the actual operating conditions, resulting in low boiler operation efficiency, insufficient system responsiveness, and prone to problems such as control lag and equipment damage.

[0004] Therefore, how to improve the intelligence level of the boiler control system is an urgent problem to be solved. Summary of the Invention

[0005] This application aims to at least solve the technical problems existing in the prior art. To this end, a first aspect of this application proposes a control method for a boiler system, and the method includes: Obtain the current operating state data of the plasma equipment during operation; wherein, the current operating state data includes current operating characteristic data, current equipment state data, current environmental data, and historical operating data of the boiler; Input the current operating state data into a preset deep neural network for processing to generate predicted values of boiler control parameters corresponding to each preset task branch; wherein, the preset deep neural network includes an input layer, a temporal chunking layer, a temporal mixture of experts layer, a fusion layer, and an output layer, and the preset task branches include an arc power adjustment value prediction task, an ignition operation instruction prediction task, an equipment maintenance prediction task, and a burner state prediction task; For each predicted value of the boiler control parameter, generate new boiler control parameters based on the predicted value of the boiler control parameter and the corresponding preset boiler control parameter target value, and adjust the current boiler control parameters based on the new boiler control parameters.

[0006] In a possible implementation, the current operating state data is input into a preset deep neural network for processing to generate predicted values of boiler control parameters corresponding to each preset task branch, including: Input the current operating state data into the preset deep neural network, and receive the current operating state data through the input layer and distribute it to the time series chunking layer; Through the time series chunking layer, the current operating state data is sliced according to a preset length to generate multiple target time blocks; Through the time series mixture of experts layer, each target time block is processed respectively according to a preset sparse mixture of experts mechanism to generate a processing result; Through the fusion layer, a preset cross-modal attention mechanism is used to fuse the processing results to generate a fused feature; Through the output layer, the fused feature is processed to generate predicted values of boiler control parameters corresponding to each preset task branch.

[0007] In a possible implementation, through the time series chunking layer, the current operating state data is sliced according to a preset length to generate multiple target time blocks, including: Through the time series chunking layer, the current operating state data is sliced according to a preset length to generate multiple initial time blocks; Through a preset masking function, data masking processing is performed on a preset number of time blocks in the multiple initial time blocks to obtain multiple target time blocks.

[0008] In a possible implementation, through the time series mixture of experts layer, each target time block is processed respectively according to a preset sparse mixture of experts mechanism to generate a processing result, including: Determine the experts corresponding to each target time block through the preset sparse mixture of experts mechanism in the time series mixture of experts layer; Each expert processes each target time block to obtain the output result of each expert; Perform weighted summation processing on the output results of each expert to generate a processing result.

[0009] In a possible implementation, through the fusion layer, a preset cross-modal attention mechanism is used to fuse the processing results to generate a fused feature, including: Through the fusion layer, calculate the attention weights corresponding to each processing result using a preset cross-modal attention mechanism; Based on each attention weight, perform weighted summation on each processing result to generate a fused feature.

[0010] In a possible implementation, new boiler control parameters are generated based on the predicted values of boiler control parameters and the corresponding preset boiler control parameter target values, including: Calculate the deviation value between the predicted value of the boiler control parameter and the corresponding preset target value of the boiler control parameter; Obtain the preset adjustment algorithm corresponding to each current boiler control parameter; Generate new boiler control parameters based on the preset adjustment algorithm and the deviation value.

[0011] In a possible implementation manner, the method further includes: Obtain the predicted value of the new boiler control parameter and the new preset target value of the boiler control parameter; Calculate the new deviation value between the predicted value of the new boiler control parameter and the corresponding new preset target value of the boiler control parameter; Obtain the preset adjustment algorithm corresponding to each current boiler control parameter; Generate new boiler control parameters based on the preset adjustment algorithm and the new deviation value.

[0012] A second aspect of the present application proposes a control device for a boiler system, and the device includes: An acquisition module, configured to acquire the current operation state data during the operation of the plasma device; wherein, the current operation state data includes current operation characteristic data, current device state data, current environmental data, and historical operation data of the boiler; A generation module, configured to input the current operation state data into a preset deep neural network for processing, and generate predicted values of boiler control parameters corresponding to each preset task branch; wherein, the preset deep neural network includes an input layer, a time series chunking layer, a time series mixture of experts layer, a fusion layer, and an output layer, and the preset task branches include an arc power adjustment value prediction task, an ignition operation instruction prediction task, a device maintenance prediction task, and a burner state prediction task; An adjustment module, configured to generate new boiler control parameters for each predicted value of the boiler control parameter based on the predicted value of the boiler control parameter and the corresponding preset target value of the boiler control parameter, and adjust the current boiler control parameter based on the new boiler control parameter.

[0013] In a possible implementation manner, the above generation module is specifically configured to: Input the current operation state data into the preset deep neural network, and receive the current operation state data through the input layer and distribute it to the time series chunking layer; Perform a segmentation process on the current operation state data by the time series chunking layer according to a preset length to generate a plurality of target time blocks; Process each target time block by the time series mixture of experts layer according to a preset sparse mixture of experts mechanism to generate a processing result; Fuse and process each processing result by the fusion layer using a preset cross-modal attention mechanism to generate a fusion feature; The fused features are processed by the output layer to generate predicted values of boiler control parameters corresponding to each preset task branch.

[0014] In a possible implementation manner, the above-mentioned generation module is further configured to: The current operating state data is segmented by the time series segmentation layer according to a preset length to generate a plurality of initial time blocks; A preset number of time blocks in the plurality of initial time blocks are subjected to data masking processing by a preset masking function to obtain a plurality of target time blocks.

[0015] In a possible implementation manner, the above-mentioned generation module is further configured to: Determine the experts corresponding to each target time block through a preset sparse mixture of experts mechanism in the time series mixture of experts layer; Each expert processes each target time block to obtain the output result of each expert; The output results of each expert are subjected to weighted summation processing to generate a processing result.

[0016] In a possible implementation manner, the above-mentioned generation module is further configured to: The fusion layer calculates the attention weights corresponding to each processing result by adopting a preset cross-modal attention mechanism; Based on each attention weight, the processing results are subjected to weighted summation to generate fused features.

[0017] In a possible implementation manner, the above-mentioned adjustment module is specifically configured to: Calculate the deviation value between the predicted value of the boiler control parameter and the corresponding preset target value of the boiler control parameter; Obtain the preset adjustment algorithm corresponding to each current boiler control parameter; Based on the preset adjustment algorithm and the deviation value, generate new boiler control parameters.

[0018] In a possible implementation manner, the control device of the above-mentioned boiler system is further configured to: Obtain a new predicted value of the boiler control parameter and a new preset target value of the boiler control parameter; Calculate the new deviation value between the new predicted value of the boiler control parameter and the corresponding new preset target value of the boiler control parameter; Obtain the preset adjustment algorithm corresponding to each current boiler control parameter; Based on the preset adjustment algorithm and the new deviation value, generate new boiler control parameters.

[0019] In a third aspect, the present invention provides a device using plasma heating, including: A controller; A memory for storing executable instructions of the controller; Wherein, the controller is configured to execute the instructions to implement the control method of the boiler system described in any one of the above.

[0020] In a fourth aspect, the present invention provides a computer-readable storage medium storing at least one instruction, at least one segment of program, a code set or an instruction set, which is loaded and executed by a processor to implement the control method of the boiler system described in any one of the above.

[0021] The embodiments of the present application have the following beneficial effects: The control method, device, equipment and medium of the boiler system provided by the embodiments of the present application, the method includes: obtaining the current operating state data of the plasma device during operation, inputting the current operating state data into a preset deep neural network for processing, generating predicted values of boiler control parameters corresponding to each preset task branch, and for each predicted value of the boiler control parameter, generating new boiler control parameters based on the predicted value of the boiler control parameter and the corresponding preset target value of the boiler control parameter, and adjusting the current boiler control parameters based on the new boiler control parameters. This solution can analyze the current operating state data of the boiler in real time and automatically adjust the current boiler control parameters by combining a preset deep neural network, has strong adaptability, and can optimize the control strategy immediately according to the change of the boiler state; in addition, since the preset deep neural network can process more complex non-linear relationships, it can make more accurate predictions based on multi-modal current operating state data, thereby improving the control accuracy. Description of the Drawings

[0022] Figure 1 It is a step flow chart of a control method of a boiler system provided by an embodiment of the present application; Figure 2 It is a step flow chart of generating predicted values of boiler control parameters provided by an embodiment of the present application; Figure 3 It is a step flow chart of generating a target time block provided by an embodiment of the present application; Figure 4 It is a step flow chart of generating a processing result provided by an embodiment of the present application; Figure 5 It is a step flow chart of generating a fusion feature provided by an embodiment of the present application; Figure 6 It is a step flow chart of generating new boiler control parameters provided by an embodiment of the present application; Figure 7 It is another step flow chart of generating new boiler control parameters provided by an embodiment of the present application; Figure 8It is a structural block diagram of a control device for a boiler system provided by an embodiment of the present application. Detailed implementation manners

[0023] Next, the technical solutions in the embodiments of the present application will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative efforts belong to the scope of protection of the present application.

[0024] Hereinafter, the terms "first" and "second" are only used for descriptive purposes, and cannot be understood as indicating or implying relative importance or implicitly specifying the quantity of the indicated technical features. Thus, the features defined with "first" and "second" may explicitly or implicitly include one or more of such features. In the description of the embodiments of the present disclosure, unless otherwise specified, the meaning of "a plurality" is two or more. In addition, the use of "based on" or "according to" means open and inclusive, because a process, step, calculation, or other action "based on" or "according to" one or more of the stated conditions or values may in practice be based on additional conditions or values beyond those stated.

[0025] Figure 1 It is a step flow chart of a control method for a boiler system provided by an embodiment of the present application. As Figure 1 shown, the method includes the following steps: Step 102, obtain the current operating state data of the plasma device during operation.

[0026] Among them, the current operating state data may include current operating characteristic data, current device state data, current environmental data, and historical operating data of the boiler. The current operating characteristic data is the core index of the boiler operating state. Optionally, after the initial operating characteristic data is collected, it can be processed to obtain the current operating characteristic data.

[0027] Specifically, the initial operating characteristic data may include temperature data of each area of the boiler, denoted as , the working pressure of the boiler, denoted as , the pulverized coal concentration, denoted as , the combustion state, denoted as . These data can be collected in real time through sensors. Among them, the pulverized coal concentration can reflect the combustion situation, and the combustion state can indicate whether the combustion is normal, usually represented as a discrete value , where 0 indicates abnormal combustion and 1 indicates normal combustion.

[0028] Next, the initially collected operating characteristic data can be vectorized at each time step t to form time series feature vectors , which is specifically defined as . Finally, after collection and vectorization, a first time series matrix can be obtained, denoted as .

[0029] And data preprocessing can be performed. Optionally, in order to remove high-frequency noise, a low-pass filter can be used to smooth each of the initially collected operating characteristic data, thereby obtaining a second time series matrix, denoted as . Next, in order to make each data have the same dimension and scale, the Z-Score normalization formula can be used to normalize each of the filtered initially collected operating characteristic data, thereby obtaining a third time series matrix, denoted as , that is, the current operating characteristic data is obtained.

[0030] The current device status data is about the operating status of the boiler and the plasma igniter. Optionally, after the initially collected device status data is collected, it can be processed to obtain the current device status data. The initially collected device status data can include the current flowing during the arc ignition process, denoted as , the arc voltage, denoted as , the cathode and anode status, denoted as , the burner wear degree, denoted as , and these data can be collected in real time through device sensors and a status monitoring system. Among them, the cathode and anode status represents the working status of the cathode and anode, , with a value of 0 indicating a fault status and a value of 1 indicating a normal status.

[0031] Next, the initially collected device status data can be vectorized at each time step t to form time series feature vectors , which is specifically defined as . Finally, after collection and vectorization, a fourth time series matrix is obtained, denoted as . And data preprocessing can be performed. Optionally, outlier detection and removal can be performed, and the principle is used for detection. Outlier detection and removal are respectively performed on each of the initially collected device status data. Then, linear interpolation can be used to supplement missing values. Next, normalization processing can be performed. After the above processing, a fifth time series matrix can be obtained, denoted as , that is, the current device status data is obtained.

[0032] The current environmental data are the external condition data of the boiler operation. Optionally, after the initial environmental data are collected, they can be processed to obtain the current environmental data. The initial environmental data may include the temperature difference between the inside and outside of the boiler, denoted as , the pressure of the gas inside the boiler, denoted as , and the air flow rate of the boiler, denoted as , which can be collected by environmental monitoring sensors.

[0033] Next, the above-collected initial environmental data can be vectorized at each time step t to form a time series feature vector , specifically defined as . Finally, after collection and vectorization, the sixth time series matrix is obtained, denoted as . And the denoising process and normalization process can be respectively performed on each initial environmental data to obtain the seventh time series matrix, denoted as , that is, the current environmental data are obtained.

[0034] The historical data may include the ignition success rate of the boiler, operation failure data, etc. These data need to be obtained from the industry. The unprocessed historical data may include the success rate during the ignition process, denoted as , and the operation failure data, denoted as . The operation failure data is used to record the boiler failure situation, , where 0 indicates no failure and 1 indicates a failure.

[0035] Next, the above unprocessed historical operation data can be vectorized to form a time series feature vector , specifically defined as . And, in order to ensure the time stamp alignment of the historical operation data and the real-time data, time alignment processing is required. Since the real-time data is continuously collected at a certain time interval, while the historical operation data may have a long interval or irregular data points, the interpolation method must be used to make the historical operation data have the same time interval as the real-time data. Then, missing value filling can be performed. For the missing values, the mean value of the historical operation data can be used to fill them. Next, normalization processing can be performed, so as to obtain the eighth time series matrix, denoted as , that is, the historical operation data of the boiler are obtained.

[0036] Next, the above data sets can be integrated into a complete comprehensive input vector, that is, the current operation state data, for the convenience of training and inference of the deep neural network. The comprehensive input vector can be specifically expressed as .

[0037] Among them, the above .

[0038] Step 104: Input the current operating status data into a preset deep neural network for processing to generate predicted values of boiler control parameters corresponding to each preset task branch.

[0039] Among them, the preset deep neural network may include an input layer, a time series chunking layer, a time series mixture of experts layer, a fusion layer, and an output layer. The preset task branches may include an arc power adjustment value prediction task, an ignition operation instruction prediction task, an equipment maintenance prediction task, and a burner status prediction task.

[0040] After obtaining the current operating status data, the current operating status data can be input into a preset deep neural network for processing, so as to generate predicted values of boiler control parameters corresponding to each preset task branch. In some optional embodiments, as Figure 2 shown, Figure 2 FIG. 10 is a flowchart of steps for generating predicted values of boiler control parameters provided by an embodiment of the present application, including: Step 202: Input the current operating status data into a preset deep neural network, and receive the current operating status data through the input layer and distribute it to the time series chunking layer.

[0041] Step 204: Perform segmentation processing on the current operating status data by the time series chunking layer according to a preset length to generate a plurality of target time blocks.

[0042] Step 206: Process each target time block respectively by the time series mixture of experts layer according to a preset sparse mixture of experts mechanism to generate a processing result.

[0043] Step 208: Perform fusion processing on each processing result by the fusion layer using a preset cross-modal attention mechanism to generate a fusion feature.

[0044] Step 210: Process the fusion feature through the output layer to generate predicted values of boiler control parameters corresponding to each preset task branch.

[0045] Among them, when generating the target time blocks, in some optional embodiments, it can be directly obtained by performing segmentation processing on the current operating status data by the time series chunking layer according to a preset length. Specifically, in the time series chunking layer, the long-term current operating status data is cut into a plurality of target time blocks with a fixed preset length, so as to independently model each target time block and capture the time series pattern. Each target time block corresponds to the current operating status data within a time period The specific form is , represents the length of each target time block. In this way, the original long time series is transformed into a plurality of short time series, so as to effectively process long time series data and reduce the computational pressure.

[0046] In some alternative embodiments, such as Figure 3 shown, Figure 3 is a flowchart of steps for generating target time blocks provided by an embodiment of the present application, including: Step 302: Through a timing slicing layer, slice the current running state data according to a preset length to generate a plurality of initial time blocks.

[0047] Step 304: Through a preset masking function, perform data masking processing on a preset number of time blocks among the plurality of initial time blocks to obtain a plurality of target time blocks.

[0048] Among them, after the slicing process, data masking processing can also be performed on a preset number of time blocks among the plurality of initial time blocks through a preset masking function, so as to obtain a plurality of target time blocks, denoted as . Among them, the masking function can mask part of the initial time block data. By selectively masking part of the initial time blocks and only using the effective preset number of time blocks for the learning of the model, the generalization ability of the model can be enhanced.

[0049] When processing each target time block respectively through a timing mixture of experts layer according to a preset sparse mixture of experts mechanism, such as Figure 4 shown, Figure 4 is a flowchart of steps for generating a processing result provided by an embodiment of the present application, including: Step 402: Determine the experts corresponding to each target time block through the preset sparse mixture of experts mechanism in the timing mixture of experts layer.

[0050] Step 404: Process each target time block through each expert to obtain the output result of each expert.

[0051] Step 406: Perform weighted summation processing on the output results of each expert to generate a processing result.

[0052] Among them, the timing mixture of experts layer adopts a sparse mixture of experts mechanism to process the sliced timing data. This layer consists of multiple expert networks, and each expert is responsible for processing different patterns of timing features. The experts corresponding to each target time block can be determined through the preset sparse mixture of experts mechanism, and the specific calculation process is , is the output of the th expert network, is the total number of experts, is the output result of each expert.

[0053] Each expert network focuses on certain specific patterns in the data. The specific calculation process of the preset sparse mixture of experts mechanism is , is the weight matrix, The function is used to map the input data to the expert selection probability .

[0054] Finally, the output results of each expert can be weighted and summed to generate a processing result, denoted as , and the specific calculation process is , is the selection probability of the i-th expert. Through the sparse mixture-of-experts mechanism of the temporal mixture-of-experts layer, the network can automatically select the most suitable expert for the current input pattern, improving the computational efficiency and model accuracy.

[0055] Thus, the preset cross-modal attention mechanism can be adopted through the fusion layer to fuse the processing results of each expert to generate a fusion feature. In some alternative embodiments, as Figure 5 shown,[[]]END]] Figure 5 is a flowchart of the steps for generating a fusion feature provided by an embodiment of the present application, including: Step 502, calculate the attention weights corresponding to the processing results of each expert through the fusion layer by adopting the preset cross-modal attention mechanism.

[0056] Step 504, perform weighted summation on the processing results of each expert based on the attention weights to generate a fusion feature.

[0057] Among them, in the fusion layer, data from different modalities are integrated through the cross-modal attention mechanism. This mechanism captures the mutual influence between different data sources and generates a globally unified representation, which is the fusion feature, denoted as . Specifically, the current operating state data of different modalities can be first mapped to the query matrix , the key matrix and the value matrix .

[0058] Among them, , are the learned weight matrices respectively.

[0059] Thus, the attention weights of each current operating state data can be obtained by calculating the similarity between the query matrix and the key matrix. Specifically, the dot product is first used to measure the similarity between the query matrix and the key matrix, denoted as , , is the dimension of the key matrix. Then, the softmax function can be used to normalize the scores to obtain the attention weights, denoted as , , which makes the attention weights of each current operating state data between 0 and 1, and the sum of all attention weights is 1.

[0060] Next, the weighted sum of each processing result can be performed based on each attention weight to generate a fused feature. Specifically, the values in the value matrix can be weighted by the attention weight to obtain the fused feature at each time step . , where is the attention weight and

[0061] is the value matrix. Finally, the fused feature is processed by the output layer to generate the predicted values of the boiler control parameters corresponding to each preset task branch. Among them, the output layer is the last layer of the deep neural network, and its main task is to generate the predicted values of the boiler control parameters according to the fused feature

[0062] . The basic structure of the output layer includes multiple preset task branches, and each preset task branch corresponds to a specific predicted value of the boiler control parameter. According to different task requirements, the design of the output layer will be different.

[0063] In some optional embodiments, the prediction function of the output layer can be defined as , where is the prediction function of the output layer;

[0064] and is the predicted value of the boiler control parameter generated. For different preset task branches, different output structures can be adopted. Optionally, the arc power adjustment value prediction task can use a regression model for point prediction, and the prediction formula of the output layer is , where

[0065] Optionally, the output of the ignition operation instruction prediction task is a discrete class label. Assuming the set of class labels for the ignition operation instruction is C = {arc starting, arc stopping, power increase, power decrease}, the prediction formula for the output layer is . Among them, is the weight matrix of the class output, that is, the weight matrix optimized during the training process; The function is used to convert the output of the model into a probability distribution, representing the probability of each class; is the predicted value of the boiler control parameter corresponding to the ignition operation instruction prediction task, and the final predicted class is the class with the highest probability.

[0066] Optionally, the equipment maintenance prediction task can use the sigmoid activation function to output a probability value indicating whether the equipment needs maintenance. Assuming the label for equipment maintenance prediction is , where 0 indicates no need for maintenance and 1 indicates need for maintenance, the output formula is . Among them, is the sigmoid activation function, which compresses the output into a value within the range [0,1], representing the maintenance probability; is the predicted value of the boiler control parameter corresponding to the equipment maintenance prediction task.

[0067] Optionally, the burner status prediction task can be a regression task or a classification task, depending on whether an accurate combustion efficiency prediction or a simple status classification is required. If it is a classification task, the output can be a class label indicating whether the burner has reached the optimal combustion efficiency; if it is a regression task, the output is a scalar representing the actual value of the combustion efficiency. Specifically, the predicted value of the boiler control parameter corresponding to the burner status prediction task can be denoted as , where, for the classification task, the above output formula is , is the weight matrix of the class output. For the regression task, the output formula is , is a linear activation function for the regression task.

[0068] In some alternative embodiments, to support point prediction and probability prediction, the output layer uses the mixture distribution log-likelihood. For regression tasks, such as the arc power adjustment value prediction task and the burner status regression prediction task, the mean squared error or the log-likelihood loss of the mixture Gaussian distribution can be used; for classification tasks, such as the ignition operation instruction prediction task and the equipment maintenance prediction task, the cross-entropy loss can be used. The specific loss function can be expressed as , where, is the loss function corresponding to the arc power adjustment value prediction task; is the loss function corresponding to the ignition operation instruction prediction task; is the loss function corresponding to the equipment maintenance prediction task; is the loss function corresponding to the burner status prediction task; are the weights corresponding to different preset task branches, respectively, and are used to adjust the influence of each preset task branch.

[0069] In this embodiment, by fusing multi-modal information such as current operation characteristic data, current equipment status data, current environmental data, and historical operation data of the boiler, a cross-modal attention mechanism is adopted to extract the dependency relationships of various types of data, so as to generate a unified time series representation. This multi-modal data fusion can comprehensively reflect the operation status of the boiler and improve the adaptability of the control system to various working conditions. And by using the time series chunking layer and the mixture of experts mechanism, the system can enhance the robustness of the model and avoid prediction errors caused by changes in data distribution or interference from local patterns, which is crucial for complex and dynamically changing boiler systems.

[0070] Step 106: For each predicted value of the boiler control parameter, generate a new boiler control parameter based on the predicted value of the boiler control parameter and the corresponding preset target value of the boiler control parameter, and adjust the current boiler control parameter based on the new boiler control parameter.

[0071] Among them, after obtaining the predicted values of each boiler control parameter, a new boiler control parameter can be generated based on the predicted value of the boiler control parameter and the corresponding preset target value of the boiler control parameter, and the current boiler control parameter can be adjusted based on the new boiler control parameter.

[0072] In some optional embodiments, as Figure 6 shown, Figure 6 is a flowchart of the steps for generating a new boiler control parameter provided by an embodiment of the present application, including: Step 602: Calculate the deviation value between the predicted value of the boiler control parameter and the corresponding preset target value of the boiler control parameter.

[0073] Step 604: Obtain the preset adjustment algorithm corresponding to each current boiler control parameter.

[0074] Step 606: Generate a new boiler control parameter based on the preset adjustment algorithm and the deviation value.

[0075] Among them, for the predicted values of the boiler control parameters corresponding to the above different preset branch tasks, the deviation values between the predicted values of the boiler control parameters and the corresponding preset target values of the boiler control parameters can be calculated respectively. The deviation value corresponding to the arc power adjustment value prediction task can be denoted as , and the deviation value corresponding to the ignition operation instruction prediction task can be denoted as , and the deviation value corresponding to the equipment maintenance prediction task can be denoted as , the deviation value corresponding to the burner status prediction task can be denoted as . Each of the above deviation values represents the deviation at time t. If the deviation value is too large, the system will trigger an adjustment strategy to reduce this deviation value.

[0076] Thus, based on the calculated deviation value, the system can correct the control behavior of the boiler through the control parameter adjustment function to obtain new boiler control parameters, denoted as , , represents the deviation value at time t.

[0077] Among them, in the process of adjusting and generating new boiler control parameters, it is necessary to first obtain the preset adjustment algorithms corresponding to each current boiler control parameter. Exemplarily, when generating new boiler control parameters corresponding to the arc power adjustment prediction task, the preset adjustment algorithm can be a linear model, that is , and are parameters obtained through training, representing the sensitivity and reference value of arc power adjustment respectively; is the new boiler control parameter corresponding to the arc power adjustment prediction task.

[0078] Exemplarily, when generating new boiler control parameters corresponding to the ignition operation instruction prediction task, the preset adjustment algorithm can be a classification algorithm, that is , is the ignition instruction generation function, which returns possible operation instruction categories, such as arc starting, arc stopping, etc., while selects the instruction category with the highest probability; is the new boiler control parameter corresponding to the ignition operation instruction prediction task.

[0079] In some alternative embodiments, after adjusting the current boiler control parameters, the system needs to monitor and feedback the adjustment result in real time. If the adjusted parameters effectively reduce the deviation and improve the boiler efficiency, the adjustment is considered successful; otherwise, readjustment is performed. Optionally, as Figure 7 shown, Figure 7 is another step flow chart for generating new boiler control parameters provided by the embodiments of the present application, including: Step 702, obtain the predicted value of the new boiler control parameter and the target value of the new preset boiler control parameter.

[0080] Step 704, calculate the new deviation value between the predicted value of the new boiler control parameter and the corresponding target value of the new preset boiler control parameter.

[0081] Step 706: Obtain the preset adjustment algorithms corresponding to the current boiler control parameters.

[0082] Step 708: Generate new boiler control parameters based on the preset adjustment algorithms and the new deviation values.

[0083] Among them, the predicted value of the new boiler control parameter can be the predicted value at time , can be the corresponding new preset target value of the boiler control parameter, so that the new deviation value between the predicted value of the new boiler control parameter and the corresponding new preset target value of the boiler control parameter can be calculated, denoted as , .

[0084] Then, the preset adjustment algorithms corresponding to the current boiler control parameters can be obtained, and new boiler control parameters can be generated based on the preset adjustment algorithms and the new deviation values. This process can refer to the above embodiments and will not be elaborated here.

[0085] In this embodiment, the control strategy is continuously optimized through a real-time feedback mechanism to ensure that the boiler operation is always in the best state. Through adaptive adjustment, the system can continuously correct the control parameters according to the actual deviation during operation, improving the stability and response speed of the system.

[0086] In some optional embodiments, to improve the accuracy and efficiency of adaptive adjustment, the system can adopt a reinforcement learning method to optimize the control strategy through a reward function. Assume that at time t, the reward function is calculated according to the deviation magnitude and unreasonable changes in the control parameters, where , is the coefficient of the penalty term, is the penalty term related to the unreasonable adjustment of the boiler control parameter , indicating the change range of the boiler control parameter. By maximizing the reward, the system will adjust the control strategy through backpropagation.

[0087] Exemplarily, if the system detects that a certain adjusted boiler control parameter exceeds the reasonable range, for example, the arc power exceeds the set safety value, then , where is the preset safety control parameter range.

[0088] The present application provides a control method for a boiler system. The method includes: obtaining current operating state data of a plasma device during operation, inputting the current operating state data into a preset deep neural network for processing to generate predicted values of boiler control parameters corresponding to each preset task branch, for each predicted value of the boiler control parameters, generating new boiler control parameters based on the predicted value of the boiler control parameters and the corresponding target value of the preset boiler control parameters, and adjusting the current boiler control parameters based on the new boiler control parameters. By combining the preset deep neural network, this solution can analyze the current operating state data of the boiler in real time and automatically adjust the current boiler control parameters, with strong adaptability, and can optimize the control strategy immediately according to the change of the boiler state. In addition, since the preset deep neural network can process more complex non-linear relationships and can make more accurate predictions based on multi-modal current operating state data, the control accuracy is improved.

[0089] It should be understood that although the steps in the flowcharts involved in the above-described embodiments are shown in sequence according to the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless there is a clear indication in this article, the execution of these steps has no strict order limit, and these steps can be executed in other orders. Moreover, at least a part of the steps in the flowcharts involved in the above-described embodiments may include multiple steps or multiple stages. These steps or stages are not necessarily executed at the same time, but can be executed at different times. The execution order of these steps or stages is not necessarily sequential, but can be executed alternately or alternately with at least a part of other steps or steps or stages in other steps.

[0090] Figure 8 It is a structural block diagram of a control device for a boiler system provided by an embodiment of the present application.

[0091] As Figure 8 shown, the control device 800 of the boiler system includes: An acquisition module 802, configured to acquire current operating state data of a plasma device during operation; wherein, the current operating state data includes current operating characteristic data, current device state data, current environmental data, and historical operating data of the boiler.

[0092] A generation module 804, configured to input the current operating state data into a preset deep neural network for processing to generate predicted values of boiler control parameters corresponding to each preset task branch; wherein, the preset deep neural network includes an input layer, a temporal chunking layer, a temporal mixture of experts layer, a fusion layer, and an output layer, and the preset task branches include an arc power adjustment value prediction task, an ignition operation instruction prediction task, an equipment maintenance prediction task, and a burner state prediction task.

[0093] An adjustment module 806 is configured to generate new boiler control parameters based on the predicted values of the boiler control parameters and the corresponding preset target values of the boiler control parameters for each predicted value of the boiler control parameters, and adjust the current boiler control parameters based on the new boiler control parameters.

[0094] Regarding the device in the above embodiments, the specific manners in which each module performs operations have been described in detail in the embodiments related to the method, and will not be elaborated here. Each module in the above graph-based retrieval enhanced generation device can be implemented in whole or in part by software, hardware, and their combination. Each of the above modules can be embedded in or independent of a processor in a computer device in the form of hardware, or stored in a memory in the computer device in the form of software, so that the processor can call and execute the operations of the above modules.

[0095] In an embodiment of the present application, a device using plasma heating is provided, including: A controller; A memory for storing executable instructions of the controller; Wherein, the controller is configured to execute the instructions to implement the control method of the boiler system described in any one of the above.

[0096] The device using plasma heating provided in the embodiments of the present application has a similar implementation principle and technical effect to the above method embodiments, and will not be repeated here.

[0097] In an embodiment of the present application, the embodiments of the present invention also provide a computer-readable storage medium, in which at least one instruction, at least one program, a code set or an instruction set is stored, and the at least one instruction, at least one program, the code set or the instruction set is loaded and executed by a processor to implement the control method of the boiler system in the embodiments of the present invention.

[0098] The computer-readable storage medium provided in this embodiment has a similar implementation principle and technical effect to the above method embodiments, and will not be repeated here.

[0099] In the above embodiments, it can be implemented in whole or in part by software, hardware, firmware, or any combination thereof. When implemented using software, it can be implemented in whole or in part 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, the processes or functions according to the embodiments of the present invention are generated in whole or in part. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable devices. The computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. 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 wired (such as coaxial cable, optical fiber, digital subscriber line (DSL)) or wireless (such as infrared, wireless, microwave, etc.) means. 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, data center, etc. that contains one or more integrated available media. The available medium can be a magnetic medium (such as a floppy disk, hard disk, magnetic tape), an optical medium (such as a DVD), or a semiconductor medium (such as a solid state disk (SSD)).

[0100] It is easy to understand that those skilled in the art can combine, split, reorganize, etc. the embodiments of the present application based on several embodiments provided in the present application to obtain other embodiments, and these embodiments do not exceed the protection scope of the present application.

[0101] The above specific implementation manners further elaborate on the purpose, technical solutions, and beneficial effects of the embodiments of the present application. It should be understood that the above are only the specific implementation manners of the embodiments of the present application and are not used to limit the protection scope of the embodiments of the present application. Any modifications, equivalent replacements, improvements, etc. made on the basis of the technical solutions of the embodiments of the present application should be included in the protection scope of the embodiments of the present application.

Claims

1. A control method for a boiler system, characterized in that: The method includes: Obtaining the current operating status data of the plasma device during operation; wherein, the current operating status data includes current operating characteristic data, current device status data, current environmental data, and historical operating data of the boiler; Inputting the current operating status data into a preset deep neural network for processing to generate predicted values of boiler control parameters corresponding to each preset task branch; wherein, the preset deep neural network includes an input layer, a time series chunking layer, a time series mixture of experts layer, a fusion layer, and an output layer, and the preset task branches include an arc power adjustment value prediction task, an ignition operation instruction prediction task, a device maintenance prediction task, and a burner status prediction task; For each of the predicted values of the boiler control parameters, generating new boiler control parameters based on the predicted values of the boiler control parameters and the corresponding preset boiler control parameter target values, and adjusting the current boiler control parameters based on the new boiler control parameters.

2. The method according to claim 1, characterized in that The step of inputting the current operating status data into a preset deep neural network for processing to generate predicted values of boiler control parameters corresponding to each preset task branch includes: Inputting the current operating status data into a preset deep neural network, and receiving the current operating status data through the input layer and distributing it to the time series chunking layer; Performing segmentation processing on the current operating status data by the time series chunking layer according to a preset length to generate a plurality of target time blocks; Processing each of the target time blocks respectively by the time series mixture of experts layer according to a preset sparse mixture of experts mechanism to generate a processing result; Performing fusion processing on each of the processing results by the fusion layer using a preset cross-modal attention mechanism to generate a fusion feature; Processing the fusion feature by the output layer to generate predicted values of boiler control parameters corresponding to each preset task branch.

3. The method according to claim 2, wherein The step of performing segmentation processing on the current operating status data by the time series chunking layer according to a preset length to generate a plurality of target time blocks includes: Performing segmentation processing on the current operating status data by the time series chunking layer according to a preset length to generate a plurality of initial time blocks; Performing data masking processing on a preset number of time blocks among the plurality of initial time blocks through a preset masking function to obtain the plurality of target time blocks.

4. The method according to claim 2 or 3, characterized in that The step of processing each of the target time blocks respectively by the time series mixture of experts layer according to a preset sparse mixture of experts mechanism to generate a processing result includes: Determining the experts corresponding to each of the target time blocks through the preset sparse mixture of experts mechanism in the time series mixture of experts layer; Processing each of the target time blocks by each expert to obtain the output result of each expert; Performing weighted summation processing on the output results of each expert to generate the processing result.

5. The method according to claim 2 or 3, characterized in that, The step of performing fusion processing on each of the processing results by the fusion layer using a preset cross-modal attention mechanism to generate a fusion feature includes: Calculating the attention weights corresponding to each of the processing results by the fusion layer using a preset cross-modal attention mechanism; Performing weighted summation on each of the processing results based on each of the attention weights to generate the fusion feature.

6. The method according to any one of claims 1 to 3, characterized in that Generating new boiler control parameters based on the predicted values of the boiler control parameters and the corresponding preset target values of the boiler control parameters includes: Calculating the deviation value between the predicted value of the boiler control parameter and the corresponding preset target value of the boiler control parameter; Obtaining the preset adjustment algorithms corresponding to the current boiler control parameters; Generating the new boiler control parameters based on the preset adjustment algorithm and the deviation value.

7. The method according to any one of claims 1 to 3, characterized in that The method further includes: Obtaining the predicted value of the new boiler control parameter and the new preset target value of the boiler control parameter; Calculating the new deviation value between the predicted value of the new boiler control parameter and the corresponding new preset target value of the boiler control parameter; Obtaining the preset adjustment algorithms corresponding to the current boiler control parameters; Generating the new boiler control parameters based on the preset adjustment algorithm and the new deviation value.

8. A control device for a boiler system, characterized in that, The device includes: An acquisition module, configured to acquire the current operation state data during the operation of the plasma device; wherein, the current operation state data includes current operation characteristic data, current device state data, current environmental data, and historical operation data of the boiler; A generation module, configured to input the current operation state data into a preset deep neural network for processing to generate the predicted values of the boiler control parameters corresponding to each preset task branch; wherein, the preset deep neural network includes an input layer, a temporal chunking layer, a temporal mixture of experts layer, a fusion layer, and an output layer, and the preset task branches include an arc power adjustment value prediction task, an ignition operation instruction prediction task, an equipment maintenance prediction task, and a burner state prediction task; An adjustment module, configured to generate new boiler control parameters based on the predicted values of the boiler control parameters and the corresponding preset target values of the boiler control parameters for each of the boiler control parameter predicted values, and adjust the current boiler control parameters based on the new boiler control parameters.

9. A device using plasma heating, characterized in that Includes: A controller; A memory for storing executable instructions of the controller; Wherein, the controller is configured to execute the instructions to implement the control method of the boiler system according to any one of claims 1-7.

10. A computer-readable storage medium, characterized in that, At least one instruction, at least one program, a code set, or an instruction set is stored in the storage medium, 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 control method of the boiler system according to any one of claims 1-7.

Images