Heat energy regulation and control system and method for acid pickling co-production of machine-made carbon and activated carbon

Abstract

The invention provides a heat energy regulation and control system and method for acid pickling co-production of machine-made carbon and activated carbon. The system is applied to the technical field of comprehensive energy utilization and comprises a data acquisition module for acquiring heat source data in a machine-made charcoal production process and heat demand data in an activated carbon pickling process; the data preprocessing module is used for preprocessing the heat source data and the heat demand data to obtain a heat supply sub-sequence and a heat consumption sub-sequence; the heat regulation and control network model is used for respectively carrying out bidirectional modeling on the heat supply sub-sequence and the heat utilization sub-sequence to extract heat flow characteristics and carrying out dynamic weighting on the heat flow characteristics so as to optimize an energy transmission path and output an optimized dynamic heat distribution scheme; and the heat regulation and control module is used for regulating and controlling signals in real time according to the optimized dynamic heat distribution scheme so as to adjust heat energy matching in the machine-made charcoal production and activated carbon pickling processes. The energy utilization rate is effectively improved, the energy consumption cost is reduced, and the comprehensive benefits are improved.

Description

Technical Field

[0001] The present invention relates to the technical field of comprehensive energy utilization, and particularly to a heat energy regulation system and method for co-producing machine-made charcoal and activated carbon pickling. Background Art

[0002] With the increasing global attention to the efficient utilization of energy and environmental protection, the comprehensive utilization of energy in various links of industrial production has become a research hotspot. In the production process of machine-made charcoal, a large amount of combustible gas rich in components such as methane, hydrogen, and carbon monoxide is generated by the high-temperature pyrolysis of biomass raw materials, and at the same time, a considerable amount of waste heat is released. In the pickling process of activated carbon, in order to ensure the pickling effect and the smooth progress of subsequent treatment processes, a large amount of steam is required for acid solution heating and activated carbon drying. Theoretically, co-producing machine-made charcoal production and activated carbon pickling can achieve energy complementarity and efficient utilization, with significant economic and environmental benefits. However, in actual operation, there is a problem of asynchronous heat demand and supply in the machine-made charcoal production and pickling processes, which makes these valuable energies not fully utilized, resulting in energy waste and restricting the economic benefits of the co-production mode. Therefore, there is an urgent need for an intelligent energy allocation method to optimize the heat energy matching, achieve efficient heat recovery and reuse, and improve energy utilization efficiency. Summary of the Invention

[0003] The present invention provides a heat energy regulation system and method for co-producing machine-made charcoal and activated carbon pickling, aiming to optimize the heat energy matching in the co-production process of machine-made charcoal production and activated carbon pickling through an intelligent heat regulation method, so as to solve the technical problems that traditional heat energy regulation methods are difficult to accurately match the heat demands of different links, the heat is asynchronous, and energy is wasted.

[0004] To achieve the above object, it is realized through the following technical solutions: According to the first aspect of the present invention, there is provided a heat energy regulation system for co-producing machine-made charcoal and activated carbon pickling, including: A data acquisition module for real-time collecting heat source data in the machine-made charcoal production process and heat demand data in the activated carbon pickling process; A data preprocessing module for normalizing and energy feature mapping processing of the heat source data and the heat demand data, extracting energy-related features and then performing data segmentation to obtain two subsequences of heat supply and heat consumption; A heat regulation network model for bidirectional modeling of the two subsequences of heat supply and heat consumption respectively to extract heat flow features, dynamically weighting the heat flow features corresponding to the heat energy distribution in different regions to ensure the efficient matching of the heat released by machine-made charcoal and the demand of activated carbon pickling, thereby optimizing the energy transmission path and outputting an optimized dynamic heat distribution plan; A heat regulation module, which is used to adjust the signals in real time according to the optimized dynamic heat distribution scheme to adjust the thermal energy matching in the production of mechanism charcoal and the pickling process of activated carbon.

[0005] Further, among them, the heat regulation network model includes: A tokenized Mamba encoder, which is used to perform bidirectional temporal modeling on the two subsequences of heat supply and heat consumption, and extract basic heat flow features; A Gaussian decay mask, which dynamically assigns weights and optimizes thermal energy matching according to the temperature gradient and conduction resistance in the thermal energy transmission path, and obtains weighted heat flow features; A semantic token learner, which is used to adaptively learn key heat flow features; A semantic token fuser, which fuses the encoded key heat flow features with the original heat source data and outputs the fused high-efficiency heat flow features, and the high-efficiency heat flow features include the global thermal energy distribution and local optimization strategies; A feature weighted fusion unit, which is used to fuse the basic heat flow features and the high-efficiency heat flow features to obtain the final optimized features for realizing the accurate modeling and optimization of thermal energy transmission.

[0006] Further, among them, the tokenized Mamba encoder, which is used to perform bidirectional temporal modeling on the two subsequences of heat supply and heat consumption and extract basic heat flow features, includes: Apply forward and backward convolution operations to these two subsequences respectively through bidirectional convolution processing, and then use one-dimensional convolution to extract features to obtain basic heat flow features.

[0007] Further, among them, the Gaussian decay mask, which dynamically assigns weights and optimizes thermal energy matching according to the temperature gradient and conduction resistance in the thermal energy transmission path, includes: Calculate the spatial Gaussian distribution weight according to the index distance between each feature in the basic heat flow features and the central node, and the standard deviation of the spatial Gaussian distribution; Calculate the Gaussian distribution weight based on features according to the Euclidean distance between each feature in the basic heat flow features and the central node feature, and the standard deviation of the Gaussian distribution based on features; Multiply the spatial Gaussian distribution weight and the Gaussian distribution weight based on features to obtain the comprehensive weight of each feature; Use the calculated comprehensive weight to weight each original feature to obtain weighted heat flow features.

[0008] Further, among them, the semantic token learner, which is used to adaptively learn key heat flow features, includes: Receive the heat flow features weighted by the Gaussian decay mask; Merge two semi - directional sequences using a cross - scanning method to obtain the integrated heat - flow characteristics, including: integrating two subsequences from different sources into a sequence of the original length through specific partitioning and splicing operations; Use a heat - flow attention module to compress the integrated heat - flow characteristic data, extract representative semantic tokens, obtain key heat - flow characteristics, and perform adaptive downsampling to optimize the heat - energy transmission efficiency.

[0009] Furthermore, the heat - flow attention module includes a channel attention mechanism and a spatial attention mechanism, which are used to capture different heat - flow patterns and spatial distribution characteristics; Through a series of heat - flow pooling, feature convolution, and non - linear activation operations, learn key energy - transmission patterns and generate semantic tokens; Perform adaptive downsampling on the generated semantic tokens, including screening or compressing the semantic tokens, to optimize the heat - energy transmission efficiency; Use a multi - layer perceptron to process the downsampled semantic tokens, and finally output semantic tokens representing heat - flow patterns to obtain key heat - flow characteristics. The key heat - flow characteristics include the heat - energy distribution in different regions, the change of heat - flow density, and the heat - conduction path information.

[0010] Furthermore, the semantic - token fuser receives semantic tokens representing heat - flow patterns and the original heat - source data and performs feature fusion to form an overall heat - energy transmission sequence, and outputs the fused high - efficiency heat - flow characteristics, making the fused high - efficiency heat - flow characteristics more in line with the actual energy demand, while taking into account local details and global transmission efficiency, and realizing precise heat - supply regulation for different process stages, including:

[0011] where, is the fused high - efficiency heat - flow characteristic; is the original input feature, that is, the original heat - source data; is a weighted pooling operation to extract key features; is a semantic - token learner operation; is a heat - flow attention module operation; is the semantic token output by the semantic - token learner.

[0012] Furthermore, the heat - flow attention module includes: a channel attention module and a heat - energy spatial attention module. The channel attention module and the heat - energy spatial attention module jointly act on the heat - flow matching optimization process, and through the combination of global and local information, realize the dynamic adjustment of the energy flow and reduce heat loss; The feature weighted fusion unit inputs the fused features into a multi-layer perceptron for training the heat regulation network model, realizes precise modeling and optimization of heat energy transmission, adjusts the local heat energy distribution in combination with the heat energy spatial attention module, and optimizes the loss function using the heat flow sensitive distance to improve the heat matching accuracy, and finally realizes the stability and efficiency of heat supply control.

[0013] Further, in the channel attention module, features are extracted by global average pooling and max pooling along the channel dimension, and a channel attention map is generated through processing by a multi-layer perceptron. The heat energy spatial attention module pools along the channel dimension to generate a feature map, splices and convolves to extract spatial attention features, and obtains a spatial attention map.

[0014] According to the second aspect of the present invention, there is also provided a heat energy regulation method for co-producing mechanism charcoal and activated carbon pickling, and the method includes: Real-time collecting heat source data in the production process of mechanism charcoal and heat demand data in the activated carbon pickling process; Performing normalization and energy feature mapping processing on the heat source data and the heat demand data, extracting energy-related features and then performing data segmentation to obtain two subsequences of heat supply and heat consumption; Constructing a heat regulation network model for bidirectionally modeling the two subsequences of heat supply and heat consumption to extract heat flow features, dynamically weighting the heat flow features corresponding to the heat energy distribution in different regions to ensure efficient matching of the heat released by the mechanism charcoal and the demand of the activated carbon pickling, thereby optimizing the energy transmission path and outputting an optimized dynamic heat distribution plan; According to the optimized dynamic heat distribution plan, real-time regulating signals are used to adjust the heat energy matching in the production of mechanism charcoal and the activated carbon pickling process.

[0015] Compared with the prior art, the present invention has the following beneficial effects: (1) The present invention proposes a tokenized Mamba encoder to perform bidirectional temporal modeling on heat flow data, starting from both ends of the heat source released by the cracking of mechanism charcoal and the heat demand of activated carbon pickling, and constructs a dynamic heat allocation mechanism. Through normalization processing and feature projection, key heat energy features are extracted, and a one-dimensional convolutional network is used to process the heat supply and heat consumption subsequences. This method can accurately depict the spatio-temporal change trend of heat energy transmission, reduce energy waste, and improve the heat utilization efficiency.

[0016] (2) Traditional heat energy regulation methods are difficult to accurately match the heat requirements of different links. The present invention proposes a Gaussian decay mask, which dynamically allocates the weights of each transmission step according to the temperature gradient and conduction resistance in the heat energy transmission path. By calculating the Gaussian decay weights of the spatial distribution and heat flow characteristics, the energy transmission path is optimized to ensure that the heat released by the mechanism charcoal can be efficiently transferred to the pickling link, thereby achieving accurate matching of heat.

[0017] (3) The present invention proposes a heat energy spatial attention module, which accurately identifies high-energy consumption areas and heat energy waste areas by constructing a local heat energy distribution model. The heat energy spatial attention module combines the channel attention mechanism and the spatial attention mechanism to extract features and perform weighted calculations on different heat flow areas, making the energy distribution more accurate, effectively reducing heat energy loss, and ensuring that heat is concentrated in key energy consumption areas, thereby improving the overall heat efficiency.

[0018] (4) Traditional heat optimization methods are difficult to effectively measure the heat flow matching effect, resulting in large errors in heat distribution. The present invention proposes a heat flow sensitive distance as the optimization target, which combines the heat flow form weight and the scale adjustment factor to measure the heat flow matching degree between different links. The heat flow sensitive distance comprehensively considers the heat gradient, heat flow balance degree, and heat load change of the heat energy transmission path, and combines the aspect ratio error and the scale error to optimize the heat supply and demand balance and improve the heat energy utilization rate of the co-production process of mechanism charcoal and activated carbon pickling.

[0019] It should be understood that the content described in the Summary of the Invention section is not intended to limit the key or important features of the embodiments of the present invention, nor is it used to limit the scope of the present invention. Other features of the present invention will become easily understood through the following description. Brief Description of the Drawings

[0020] In combination with the accompanying drawings and with reference to the following detailed description, the above and other features, advantages, and aspects of the embodiments of the present invention will become more apparent. The drawings are used to better understand the solution and do not constitute a limitation to the present invention. In the drawings, the same or similar reference numerals represent the same or similar elements, where: Figure 1 is a schematic diagram of the modules of a heat energy regulation system for the co-production of mechanism charcoal and activated carbon pickling according to an embodiment of the present invention; Figure 2 is a flowchart of a heat energy regulation system for the co-production of mechanism charcoal and activated carbon pickling according to an embodiment of the present invention; Figure 3 is a schematic diagram of the steps of a heat energy regulation method for the co-production of mechanism charcoal and activated carbon pickling according to an embodiment of the present invention; Figure 4 is a heat map of the visualization of the gradient distribution of the algorithm of the present invention; Figure 5It is a trend graph of the mean square error, coefficient of determination, accuracy, and loss function of the algorithm of the present invention with the number of training rounds. Specific Embodiments

[0021] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are some, but not all, of the embodiments of the present invention. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0022] In addition, the term "and / or" in this article is only a description of the association relationship of associated objects, indicating that three relationships may exist. For example, A and / or B may represent: A exists alone, A and B exist simultaneously, and B exists alone. In addition, the character " / " in this article generally represents an "or" relationship between the associated objects before and after.

[0023] Figure 1 Shows a schematic module diagram of a thermal energy regulation system for co-production of machine-made charcoal and activated carbon pickling; Figure 2 Shows a flowchart of a thermal energy regulation system for co-production of machine-made charcoal and activated carbon pickling. As Figure 1 and Figure 2 shown, a system 100 for co-production of machine-made charcoal and activated carbon pickling includes: A data acquisition module 110 for acquiring heat source data in the production process of machine-made charcoal and heat demand data in the pickling process of activated carbon; Among them, the heat source end data includes time series data of temperature, heat flow, and combustible gas components (such as methane, hydrogen, carbon monoxide, etc.) generated in the production process of machine-made charcoal. The heat demand end data includes time series demand data such as steam temperature, heat load, and drying time required for the activated carbon pickling process. In the specific acquisition process, a thermocouple (accuracy ±1°C) is used to real-time acquire the temperature of the machine-made charcoal cracking furnace (range 300 - 800°C), and a gas flowmeter (measurement range 0 - 100 L / min) is used to monitor the flow rate of combustible gas ( , , CO), and a steam pressure sensor (0 - 10 bar) is used to acquire the steam demand data in the pickling link.

[0024] A data preprocessing module 120 for normalizing and energy feature mapping processing of the heat source data and the heat demand data, extracting energy-related features, and then performing data segmentation to obtain two subsequences of heat supply and heat consumption. For example, the acquired data is sampled at 5-second intervals and Z-score normalized (mean ±3σ truncation).

[0025] The heat regulation network model 130 is used to perform bidirectional modeling on the two subsequences of heat supply and heat consumption respectively to extract heat flow characteristics, dynamically weight the heat flow characteristics corresponding to the heat energy distribution in different regions, ensure the efficient matching of the heat released by the mechanism charcoal and the demand for activated carbon pickling, thereby optimizing the energy transmission path and outputting an optimized dynamic heat distribution plan; The heat regulation module 140 is used to adjust the heat energy matching in the process of mechanism charcoal production and activated carbon pickling in real time according to the optimized dynamic heat distribution plan.

[0026] For example, the PID controller outputs a 4-20 mA signal to adjust the opening of the waste heat boiler valve and the electric control valve of the steam distribution pipeline to control the heat energy transmission rate.

[0027] The working process of the heat energy regulation system is as follows: 1. The data acquisition module obtains the heat source and heat demand data in real time; 2. The data preprocessing module performs normalization and feature segmentation; 3. The heat regulation network model generates an optimization plan through encoding, weighting, token learning and fusion in sequence; 4. The heat regulation module outputs a valve control signal. 5. Adjust the heat energy matching in real time with regulation signals (such as valve opening, pump speed control instructions).

[0028] The co-production of mechanism charcoal and activated carbon pickling uses water as a carrier to construct an energy storage system, and three heat sources are provided by the high-temperature pyrolysis of mechanism charcoal to meet the energy consumption for pickling. The heat regulation network model is used to predict and optimize the time interval between energy supply and demand, effectively solve the problem of heat asynchrony, improve energy utilization efficiency, reduce energy consumption costs, realize the efficient collaborative production of mechanism charcoal and activated carbon pickling, and enhance the comprehensive benefits. This problem involves heat scheduling balance and is modeled based on heat supply and demand:

[0029] : At time , the heat provided by the production of mechanism charcoal.

[0030] : At time , the heat required for the pickling process.

[0031] In order to achieve the matching of heat supply and demand in the production of mechanism charcoal and pickling, the present invention designs the following objective function for optimization. The optimization goal is to minimize the difference between heat supply and demand, that is:

[0032] Among them, is the time period considered in the optimization process; represents the absolute value to ensure that the difference is positive.

[0033] Through the training of historical data, the system can intelligently adjust the matching between energy supply and demand, thereby eliminating the problem of heat asynchrony.

[0034] The data preprocessing module 120 is used to perform normalization and energy feature mapping processing (such as Fourier transform) on the heat source data and the heat demand data, extract energy-related features, and then perform data segmentation to obtain two subsequences of heat supply and heat consumption. Specifically, normalization uses Z-score standardization, and energy feature mapping extracts the frequency-domain energy distribution through Fourier transform, retaining the first 10 main frequency components to reduce the influence of noise.

[0035] This system adopts a centralized heat regulation method. Starting from the heat source released by the pyrolysis of mechanism charcoal and the heat demand of the pickling of activated carbon, a dynamic heat allocation mechanism is constructed to achieve the efficient utilization and synchronous regulation of energy. The system converts temperature and heat flow data into four sets of semi-directional time series features to more accurately capture the changing trends of heat supply and demand, reduce energy loss, and reduce the computational complexity through lightweight optimization, making it applicable to heat management under different working conditions. Specifically, the normalized heat source data and heat demand data are sliced in the time direction into: forward heat supply feature (forward time series, heat source end), forward heat consumption feature (forward time series, heat demand end), backward heat supply feature (backward time series, heat source end), and backward heat consumption feature (backward time series, heat demand end), that is, four sets of semi-directional time series features.

[0036] By combining the forward and backward features through bidirectional one-dimensional convolution, two subsequences of heat supply and heat consumption are split from the previous four sets of semi-directional time series features, that is, a heat supply subsequence (forward + backward heat supply) and a heat consumption subsequence (forward + backward heat consumption) are generated.

[0037] Among them, the heat regulation network model 130 includes: The tokenized Mamba encoder is used to perform bidirectional time modeling on the two subsequences of heat supply and heat consumption, and extract basic heat flow features; The Gaussian decay mask dynamically allocates weights and optimizes heat energy matching according to the temperature gradient and conduction resistance in the heat energy transmission path to obtain weighted heat flow features; The semantic token learner is used to adaptively learn key heat flow features; The semantic token fuser fuses the encoded key heat flow features with the original heat source data and outputs the fused efficient heat flow features, and the efficient heat flow features include the global heat energy distribution and local optimization strategies; The feature weighted fusion unit is used to fuse the basic heat flow features and the efficient heat flow features to obtain the final optimized features, which are used to achieve accurate modeling and optimization of heat energy transmission.

[0038] The heat regulation network model includes a tokenized Mamba encoder, a Gaussian decay mask, a semantic token learner, a semantic token fuser, and a feature weighted fusing unit, which gradually constructs an accurate dynamic heat allocation mechanism from the perspectives of feature extraction, thermal energy modeling, semantic optimization, feature fusion, etc.

[0039] Preferably, in some embodiments, the tokenized Mamba encoder is used to perform bidirectional temporal modeling on two subsequences of heat supply and heat consumption, and extract basic heat flow features, including: Apply forward and backward convolution operations to these two subsequences respectively through bidirectional convolution processing, and then use one-dimensional convolution to extract features to obtain basic heat flow features.

[0040] Specifically, in terms of energy prediction and regulation, a tokenized Mamba encoder is used to perform bidirectional modeling on two subsequences of heat supply and heat consumption. The two subsequences of heat supply and heat consumption obtained by splitting the time series heat flow data are respectively passed through an activation function and one-dimensional convolution to extract features, and then the basic heat flow features are obtained and input into the Gaussian decay mask of the regulation network. Subsequently, the Gaussian decay mask is used to optimize the thermal energy matching strategy, focusing on efficient energy transmission paths, and combined with the semantic token learner to adaptively learn key heat flow features for dynamic optimization and downsampling. Finally, the semantic token fuser integrates multi-source energy information to achieve precise heat supply control, and the system stability is improved through residual connection.

[0041] Among them, the key operation of the tokenized Mamba encoder is to encode the sequence through bidirectional processing, which is the basic sequence processing. Specifically based on the bidirectional temporal convolutional network (Bi-TCN), the forward and backward convolutional kernel sizes are 3, the stride is 1, the number of output channels is 64, and the activation function is ReLU. The formula is defined as: The formula includes:

[0042] Among them, : The input feature sequence; : The weight matrices of the forward and backward convolution operations respectively, which are learnable parameters; : The corresponding bias term.

[0043] The tokenized Mamba encoder processes sequence data through bidirectional convolution, thereby capturing context information in different directions.

[0044] Use the tokenized Mamba encoder to extract the temporal features of the heat flow data, and the output is the normalized heat flow temporal features, which are transmitted to the Gaussian decay mask to calculate the spatial weighted weights and optimize the thermal energy matching.

[0045] The tokenized Mamba encoder is responsible for extracting the temporal features of the heat flux data, while the Gaussian decay mask is mainly used for spatial feature modeling. The relationship between them can be understood as joint time-space modeling. The tokenized Mamba encoder generates the heat flux temporal features for calculating the heat demand in different regions at the current moment. The Gaussian decay mask uses these temporal features as inputs to dynamically adjust the spatial heat energy distribution, optimize the transmission path, and improve the heat utilization efficiency.

[0046] Preferably, in some embodiments, the Gaussian decay mask dynamically assigns weights and optimizes the heat energy matching according to the temperature gradient and conduction resistance in the heat energy transmission path.

[0047] According to the characteristics of the temperature gradient and conduction resistance in the heat energy transmission path, weights are assigned to each transmission step (referring to the discretization stage of the heat energy in the heat supply path, that is, the propagation state of the heat energy at different positions and time points), with a focus on key heat interaction nodes. Since the heat flux has temporal dynamics (time series) and spatial diffusion characteristics (heat energy transmission path), it is necessary to perform optimization modeling in both the time domain (tokenized Mamba encoder) and the spatial domain (Gaussian decay mask). By calculating the Gaussian decay weights of the spatial distribution and heat flux characteristics, a more refined heat energy scheduling is achieved, enhancing the system's perception and optimization ability for the energy transmission process. In the Gaussian decay mask mechanism, according to the temperature change rate and conduction efficiency in the heat energy transmission path, the heat energy distribution in different regions is dynamically weighted to ensure that the heat released by the mechanism carbon is efficiently matched with the demand for activated carbon pickling, thereby optimizing the energy transmission path and improving the overall heat utilization efficiency. Assume the feature sequence is , where, is the length of the feature sequence, that is, the total number of features. For the th feature , its weight is the product of the spatial Gaussian distribution weight and the feature-based Gaussian distribution weight:

[0048] where, is the spatial Gaussian distribution weight, and the calculation formula is:

[0049] : the spatial Gaussian distribution weight of the th feature; : the index distance between the th feature and the central node; : the standard deviation of the spatial Gaussian distribution, controlling the speed of spatial decay, is set to 1 / 5 of the transmission path length to ensure that the weights cover the main heat conduction regions.

[0050] Among them, is the Gaussian distribution weight based on features, and the calculation formula is:

[0051] : the Gaussian distribution weight based on features of the th feature; : the Euclidean distance between the th feature and the feature of the central node; : the standard deviation of the Gaussian distribution based on features, which controls the attenuation speed of features.

[0052] Among them, the selection strategies of the central node include: (1) using the peak position of heat demand as the central node to dynamically track the highest heat load area of the pickling process; (2) using the geometric center of the heat energy transmission path as the benchmark to ensure the balance of spatial weight distribution. Specifically, the system dynamically selects the central node according to the real-time heat load distribution: when there is a significant heat demand peak in the pickling process, the peak position is used as the central node; when the heat load distribution is uniform, the geometric center of the transmission path is used as the benchmark. Dynamic selection of the central node: Detect the local maximum value of the pickling heat demand sequence through a sliding window (window size: 10 time steps). If the peak exceeds the threshold (such as the mean + 2 times the standard deviation), the peak position is used as the center; otherwise, the geometric center of the transmission path (the midpoint of the sequence) is used as the benchmark.

[0053] The weight is used for the heat energy distribution in the pickling process of mechanism carbon and activated carbon. As the final feature weight, it dynamically weights the heat flow features, making the heat flow scheduling more accurate and improving the system's optimization ability for heat transfer.

[0054] According to the temperature gradient and conduction resistance in the heat energy transmission path, the Gaussian attenuation mask calculates the weights of key heat flow interaction nodes to ensure the reasonable spatial distribution of heat flow features and optimize the perception ability of the heat energy regulation network. The output is the weighted weights of different feature points and is transmitted to the semantic token learner to extract key heat flow features.

[0055] Preferably, in some embodiments, the semantic token learner is used to adaptively learn key heat flow features, including: receiving the heat flow features weighted by the Gaussian attenuation mask; using the cross-scanning method to merge two semi-directed sequences, such as using bidirectional LSTM or temporal convolution to process the semi-directed sequences to ensure that the temporal dynamics are not destroyed, and obtaining the integrated heat flow features, including: through specific partitioning and splicing operations, integrating two subsequences from different sources into a sequence of the original length, maintaining the temporal dynamics of the features, and providing a suitable input for the subsequent processing layer; The heat flow attention module is used to compress the integrated heat flow feature data, extract representative semantic tokens, obtain key heat flow features, and perform adaptive downsampling to optimize the thermal energy transfer efficiency.

[0056] Furthermore, the heat flow attention module compresses the heat flow feature data, extracts representative semantic tokens (key heat flow features), and performs adaptive downsampling to optimize the heat energy transfer efficiency. Through a series of heat flow pooling, feature convolution, and nonlinear activation operations, key energy transfer patterns are learned and semantic tokens are generated to capture efficient heat flow features and provide support for dynamic heat matching. The semantic token learner adaptively learns and extracts key heat flow features based on the heat flow attention module to achieve adaptive thermal optimization of different process sections. Assume that the input sequence is ,in, is the length of the input sequence, that is, the total number of input data or the number of samples, and the output token of the semantic token learner It can be expressed as:

[0057] : The first elements; : Weighted features calculated by the heat flow attention module; MLP: Multi-layer perceptron, used to process the features output by the attention mechanism and generate the final semantic token.

[0058] Through cross-scanning and feature concatenation, the semantic token learner extracts representative semantic tokens and obtains key heat flow features to achieve efficient heat matching. The output is a semantic token (representing the heat flow pattern), which is passed to the semantic token fuser for feature fusion.

[0059] Preferably, in some embodiments, the heat flow attention module includes a channel attention mechanism and a spatial attention mechanism for capturing different heat flow patterns and spatial distribution characteristics; Through a series of heat flow pooling, feature convolution and nonlinear activation operations, key energy transfer patterns are learned and semantic tokens are generated; Adaptively downsampling the generated semantic tokens, including filtering or compressing the semantic tokens to optimize thermal energy transfer efficiency; The downsampled semantic tokens are processed using a multi-layer perceptron, and finally semantic tokens representing the heat flow pattern are output to obtain key heat flow features, which include thermal energy distribution in different regions, changes in heat flux density, and heat conduction path information.

[0060] Preferably, in some embodiments, the semantic token fuser receives semantic tokens representing the heat flow pattern and the original heat source data, performs feature fusion to form an overall thermal energy transfer sequence, and outputs the fused efficient heat flow features, making the fused efficient heat flow features more in line with the actual energy requirements, while taking into account local details and global transfer efficiency, and achieving precise heating control for different process stages, including: During the heating control process, the encoded semantic heat flow features are fused with the original heat source data such as temperature, flow rate, etc. Through operations such as adaptive pooling, weight assignment calculation, and weighted summation, the semantic token information is incorporated into the overall thermal energy transfer sequence, making the fused features more in line with the actual energy requirements, while taking into account local details and global transfer efficiency. The semantic token fuser finally integrates multi-source energy information, optimizes the heat flow matching, achieves precise heating control, and improves the system stability through residual connections, enhancing the adaptability to heat load fluctuations.

[0061] The formula for its fusion process includes:

[0062] Where, is the fused efficient heat flow feature; is the original input feature, that is, the original heat source data; is the weighted pooling operation to extract key features; is the semantic token learner operation; is the heat flow attention module operation; is the semantic token output by the semantic token learner.

[0063] Combining the semantic heat flow features with the original heat source data such as temperature, flow rate, etc., the semantic token fuser integrates multi-source heat flow information for global optimization matching to ensure that the heat flow transmission meets the actual requirements. The output is the fused efficient heat flow feature, which is transmitted to the feature weighted fusion unit for final training optimization.

[0064] In the thermal energy regulation task, the heat transfer is affected by both channel features (different heat flow patterns) and spatial features (the distribution of heat flow in different regions). Therefore, the heat flow attention module consists of a channel attention module and a thermal energy spatial attention module. The two work together in the heat flow matching optimization process to improve the heat matching efficiency in a global-local joint optimization manner, reduce heat loss, and achieve dynamic adjustment of the energy flow. Among them, the channel attention module extracts key heat flow features along the heat conduction path direction using global average pooling and max pooling, and calculates and generates heat flow distribution weights through a multi-layer perceptron to optimize the thermal energy utilization efficiency of different process links. The thermal energy spatial attention module accurately captures local hot spots and inefficient regions in heat transfer through local energy distribution modeling, realizes dynamic adjustment of the energy flow, and reduces heat loss.

[0065] The channel attention module extracts features along the channel dimension using global average pooling and max pooling, and generates a channel attention map through the processing of a multi-layer perceptron, including: Performing global average pooling and global max pooling merging them into , and generating a channel attention map through the processing of a two-layer multi-layer perceptron :

[0066] Among them, is the channel attention map, with the dimension of , serving as channel-level weight allocation, adjusting the heat flow contribution of each channel, evaluating the global importance of different heat flow features, and optimizing the thermal energy utilization efficiency of different process links; is the activation function; MLP is the multi-layer perceptron, including a dimensionality reduction layer and a dimensionality increase layer.

[0067] The input feature , is the number of channels, is the height of the feature map, is the width of the feature map, obtained through global average pooling (GAP) and max pooling (GMP) . The MLP structure is two fully connected layers. The first layer reduces the C dimension to C / 4 and activates with ReLU; the second layer restores to the C dimension and activates with Sigmoid, generating the channel attention map .

[0068] The thermal energy spatial attention module pools along the channel dimension to generate a feature map, splices and convolves to extract spatial attention features. The thermal energy spatial attention module converts it into a bidirectional two-dimensional spatial decay format. Through the spatial decay matrix, the attention score attenuation degree of the surrounding markers farther away from the target center is greater, so as to focus more on the target center and its surrounding areas while paying attention to global information.

[0069] Specifically, the thermal energy spatial attention module pools along the channel dimension to generate a feature map, splices and convolves to extract spatial attention features, that is, performing channel average pooling and channel max pooling splicing them into , and then obtaining a spatial attention map through a convolutional layer :

[0070] Among them, is the spatial attention map, with the dimension of , as the attention weight at the spatial level, optimizes the matching effect of the heat energy flow in different regions, optimizes the local heat energy transmission, focuses on the heat energy hotspots and inefficient regions, and improves the heat energy utilization rate; For the feature map formed by concatenating channel average pooling and channel max pooling , where is the activation function, usually the Sigmoid function; Conv is the convolution operation.

[0071] Let the input feature map be , channel average pooling and channel max pooling are respectively:

[0072] where is the feature map after channel average pooling, with the dimension of , and is used to extract the spatial heat flow distribution features; is the cumulative operation of all elements in the rows and columns of the feature map; is the normalization factor, which calculates the pixel mean value to make the output independent of the size of the input feature map; represents the element value of the th channel, the th row, and the th column in the input feature map. Among them, ranges from 1 to ( is the number of channels), ranges from 1 to ( is the height of the feature map), ranges from 1 to ( is the width of the feature map); is the feature map after channel max pooling, with the dimension of ; and is used to extract the spatial heat flow distribution features; represents the double maximum operation to find the maximum value in the feature structure.

[0073] Concatenate channel average pooling and channel max pooling into , ; Process it with a 7×7 convolution kernel, output a single-channel feature map, and generate a spatial attention map through Sigmoid activation .

[0074] The feature weighted fusion unit inputs the fused features into a multi-layer perceptron for training the heat regulation network model, realizing precise modeling and optimization of heat energy transmission, adjusting the local heat energy distribution in combination with the heat energy spatial attention module, and using the heat flux sensitive distance to optimize the loss function, improving the heat matching accuracy, and finally realizing the stability and efficiency of heat supply control.

[0075] To optimize the heat energy matching in the co-production process of mechanism charcoal and activated carbon pickling, this step realizes precise modeling and optimization of heat energy transmission by fusing the original heat flux features and the efficient heat flux features extracted by the semantic token fuser. The fused features are input into a multi-layer perceptron for training the heat regulation network model to ensure that the heat released by the mechanism charcoal can be efficiently transmitted to the activated carbon pickling process section. In addition, the local heat energy distribution is adjusted in combination with the heat energy spatial attention module, and the heat flux sensitive distance is used to optimize the loss function, improving the heat matching accuracy, and finally realizing the stability and efficiency of heat supply control.

[0076] Among them, optimization of the heat matching loss function: Introduce the heat flux morphology weight and the dynamic scale adjustment factor, calculate the heat flux sensitive distance based on the heat gradient of the heat energy transmission path, the balance degree of heat flux distribution, and the heat load changes in different stages, and incorporate the heat matching error term to enhance the system's adaptability to heat imbalance problems and improve the heat energy utilization rate of the co-production process of mechanism charcoal and activated carbon pickling. To optimize the accuracy of the target, the loss function includes the following parts:

[0077] Among them, is the heat matching loss function, is the weight factor of the shape heat flux sensitive distance, used to adjust the influence of the loss; is the heat flux sensitive distance.

[0078] The calculation formula for defining the heat flux sensitive distance is:

[0079] Among them, is the generalized intersection over union loss; is the heat flux sensitive distance; is the heat flux morphology weight, controlling the importance of the aspect ratio error; is the scale adjustment factor, controlling the importance of the scale error; is the aspect ratio error; in some specific embodiments, for example, it is set as: , β = 0.5.

[0080]

[0081] Among them, P is B and The minimum bounding box, where IoU is the intersection over union.

[0082] Let the heat supply and demand time window be , and the heat supply and demand reference window be , where is the center coordinate of the box, is the width and height of the box. Define the aspect ratio error The calculation formula is:

[0083] represents the aspect ratio error, i.e., the scale error, which measures the difference in shape between the heat supply and demand time window and the heat supply and demand reference window, especially the change in the aspect ratio. represents calculating the aspect ratio of the heat supply and demand reference window and taking the larger value to ensure a unified scale for measuring the change in the aspect ratio. represents calculating the aspect ratio of the heat supply and demand time window and taking the maximum value among them to ensure numerical stability. is a very small constant (such as 10-7) to prevent division by zero errors.

[0084] Scale error The calculation formula is:

[0085] Among them, is the scale error, which measures the proportional difference in size between the heat supply and demand time window and the heat supply and demand reference window; is the maximum size (either width or height) of the heat supply and demand reference window; is the maximum size (either width or height) of the heat supply and demand time window.

[0086] In summary, the tokenized Mamba encoder extracts temporal heat flow features to form a preliminary input. The Gaussian decay mask calculates spatial weights to optimize the heat energy matching strategy. The semantic token learner extracts key heat flow features, generates semantic tokens, and enhances the learning of the energy transmission mode. The semantic token fuser combines the original input to optimize the global heat energy regulation strategy and improve the heating accuracy. The feature weighted fusion unit performs the final optimization and trains the MLP to achieve stable control of the heat regulation network model. Finally, these five parts work together to ensure the optimal heat energy matching of the mechanism charcoal and activated carbon pickling process and improve the overall heat utilization rate.

[0087] After the above processing process, the algorithm finally outputs an optimized heat supply-demand matching scheme to achieve efficient heat energy utilization in the production of mechanism charcoal and the pickling process of activated carbon. The algorithm performs dynamic energy management through the heat regulation network model 130, optimizes heat flow matching by combining Gaussian decay masks, and uses semantic token learners and semantic token fusers to integrate multi-source energy information, optimize the heat transfer path. Through Gaussian decay mask optimization, it ensures that heat energy flows along the efficient energy transfer path, reduces heat loss, and improves the overall heat energy utilization rate. Key heat flow features are extracted by the semantic token learner and combined with the semantic token fuser for multi-layer energy fusion to achieve precise heat supply regulation for different process stages and reduce heat waste. The system finally minimizes the heat matching loss function to the minimum, improves energy utilization efficiency, and reduces operating costs.

[0088] In the above embodiments of the present invention, through bidirectional temporal modeling of the tokenized Mamba encoder, the spatio-temporal change trend of heat energy transmission is accurately characterized, reducing energy waste. The Gaussian decay mask dynamically assigns weights to optimize the energy transmission path and achieve precise heat matching. The heat energy spatial attention module combines the channel attention mechanism and the spatial attention mechanism to accurately identify high-energy consumption areas and reduce heat energy loss. Taking the heat flow sensitive distance as the optimization target, it improves the accuracy of heat supply-demand matching and enhances the overall heat efficiency.

[0089] The above is the introduction of the system embodiments. The following further illustrates the solution of the present invention through method embodiments.

[0090] Figure 3 The step schematic diagram of a method for co-producing mechanism charcoal and pickled activated carbon according to an embodiment of the present invention is shown. As Figure 3 shown, a method for co-producing mechanism charcoal and pickled activated carbon includes: S201: Real-time collect the heat source data in the production process of mechanism charcoal and the heat demand data in the pickling process of activated carbon; S202: Perform normalization and energy feature mapping processing on the heat source data and the heat demand data, extract energy-related features, and then perform data segmentation to obtain two subsequences of heat supply and heat consumption; S203: Construct a heat regulation network model to perform bidirectional modeling on the two subsequences of heat supply and heat consumption respectively to extract heat flow features, dynamically weight the heat flow features corresponding to the heat energy distribution in different regions, ensure the efficient matching of the heat released by mechanism charcoal and the demand for pickling activated carbon, thereby optimizing the energy transmission path, and output an optimized dynamic heat distribution scheme; S204: According to the optimized dynamic heat distribution scheme, adjust the signal in real time to adjust the heat energy matching in the production process of mechanism charcoal and the pickling process of activated carbon.

[0091] Those skilled in the art can clearly understand that for the convenience and conciseness of description, the specific working processes of the described steps can refer to the corresponding processes in the foregoing system embodiments and will not be elaborated herein.

[0092] The experimental results are as follows: As Figure 4 The heat map of the gradient distribution visualization shown, which shows the gradient changes in different regions, that is, the heat map of the Gaussian decay mask weight distribution, can show that the heat energy is preferentially distributed to the high-demand regions to evaluate the effect of feature learning. The color represents the intensity of the heat flow gradient, the red region represents a higher gradient change, and the blue region represents a lower gradient change. This visualization helps to analyze the key regions in the process of optimizing heat energy transmission.

[0093] The mean value of the overall gradient distribution is 0.045, and the standard deviation of the gradient is 0.012. The maximum gradient value is about 0.092; the minimum gradient value is about 0.008. The proportion of high-gradient regions, which are the key points of energy transmission (value > 0.8), is 13.2%, indicating that the heat energy change in this region is the most intense. The proportion of low-gradient regions, which are the heat stable regions (value < 0.02), is 27.5%, indicating that the heat in this region is relatively uniform and the optimization space is limited.

[0094] Under the action of the Gaussian decay mask optimization strategy, the heat flow gradient shows a trend of gradually decreasing from the high-gradient center outward, which conforms to the theoretical expectation of the gradual and balanced transmission of heat. The semantic token learner extracts the main key heat flow features, indicating that the dynamic optimization has a significant effect in these regions. During the process of integrating energy information, the semantic token fuser reduces the overall gradient variance by about 23.7%, indicating that the heat supply control is more stable.

[0095] Figure 5 Shows the changing trends of the mean squared error, coefficient of determination, accuracy, and loss function of the algorithm with the number of training rounds. Among them, the mean squared error uses yellow dots, the coefficient of determination uses orange squares, the accuracy uses red triangles, and the heat matching loss function uses pink diamonds. Among them, for network training: Adam optimizer (learning rate 0.001, decay rate × 0.1 every 50 rounds), batch size 32, training for 200 rounds, and the loss function is . The training curve shows that the loss function converges, and the accuracy of the validation set reaches 92%.

[0096] The convergence speed of the model is relatively fast and enters a stable state around 30 rounds. Continuing training has limited benefits. The key learning stage is between 10 - 25 rounds, where the mean squared error drops the fastest, and the accuracy and coefficient of determination increase the most significantly, indicating that the learning effect in this stage is the most remarkable. The final accuracy reaches 98%, the coefficient of determination is close to 0.9, and the heat matching loss function Lower than 0.05 and the mean square error is lower than 0.1, indicating that the model already has high prediction accuracy and stability. For practical applications, it is recommended to train for about 30 rounds because at this time the model performance has tended to be stable, and increasing the number of training rounds will have little improvement on the results, which can save computing resources.

[0097] According to Figure 5 The experimental results show that the algorithm tends to be stable after 30 rounds of training, improving the energy utilization rate and realizing the efficient collaborative production of mechanism charcoal and activated carbon pickling.

[0098] It should be understood that various forms of the processes shown above can be used, steps can be reordered, added or deleted. For example, the steps described in the present invention can be executed in parallel, sequentially or in a different order, as long as the desired results of the technical solutions disclosed in the present invention can be achieved, and no limitations are imposed herein.

[0099] The above specific embodiments do not constitute a limitation on the protection scope of the present invention. Those skilled in the art should understand that various modifications, combinations, sub - combinations and substitutions can be made according to design requirements and other factors. Any modifications, equivalent substitutions and improvements made within the spirit and principle of the present invention shall be included within the protection scope of the present invention.

Claims

1. A thermal energy control system for the co-production of machine-made carbon and activated carbon pickling, characterized in that: include: Data acquisition module, used to collect real-time heat source data of the carbon production process and heat demand data of the activated carbon pickling process; A data preprocessing module is used to normalize and perform energy feature mapping on the heat source data and the heat demand data, extract energy-related features, and then perform data segmentation to obtain two subsequences of heat supply and heat use; The heat control network model is used to perform bidirectional modeling on the two subsequences of heat supply and heat use to extract heat flow characteristics, dynamically weight the heat flow characteristics corresponding to the heat energy distribution in different regions, ensure that the heat released by the machine-made charcoal is efficiently matched with the demand for activated carbon pickling, thereby optimizing the energy transmission path and outputting the optimized dynamic heat distribution plan; The heat control module is used to control the signal in real time according to the optimized dynamic heat distribution plan to adjust the heat energy matching between the machine-made carbon production and the activated carbon pickling process.

2. The thermal energy control system for the co-production of machine-made carbon and activated carbon pickling according to claim 1 is characterized in that: in, The heat regulation network model includes: A tokenized Mamba encoder is used to perform bidirectional temporal modeling on the two subsequences of heat supply and heat consumption to extract basic heat flow features; Gaussian attenuation mask, dynamically assigns weights and optimizes thermal energy matching according to the temperature gradient and conduction resistance in the thermal energy transfer path to obtain weighted heat flow characteristics; Semantic token learner for adaptive learning of key heat flow features; A semantic token fuser fuses the encoded key heat flow features with the original heat source data and outputs a fused efficient heat flow feature, which includes global heat energy distribution and local optimization strategy; The feature weighted fusion unit is used to fuse the basic heat flow feature with the efficient heat flow feature to obtain the final optimized feature for realizing accurate modeling and optimization of thermal energy transmission.

3. The thermal energy control system for the co-production of machine-made carbon and activated carbon pickling according to claim 2 is characterized in that: in, The tokenized Mamba encoder is used to perform bidirectional time modeling on the two subsequences of heat supply and heat consumption to extract basic heat flow features, including: The two subsequences are processed by bidirectional convolution, and forward and backward convolution operations are applied to them respectively. Then, one-dimensional convolution is used to extract features to obtain basic heat flow features.

4. The thermal energy control system for the co-production of machine-made carbon and activated carbon pickling according to claim 2 is characterized in that: in, The Gaussian attenuation mask dynamically allocates weights and optimizes thermal energy matching according to the temperature gradient and conduction resistance in the thermal energy transfer path, including: Calculating the spatial Gaussian distribution weight according to the index distance between each feature in the basic heat flow feature and the central node and the standard deviation of the spatial Gaussian distribution; Calculate a feature-based Gaussian distribution weight according to the Euclidean distance between each feature in the basic heat flow feature and the central node feature, and the standard deviation of the feature-based Gaussian distribution; Multiply the spatial Gaussian distribution weight and the feature-based Gaussian distribution weight to obtain the comprehensive weight of each feature; The calculated comprehensive weight is used to weight each original feature to obtain the weighted heat flow feature.

5. The thermal energy control system for the co-production of machine-made carbon and activated carbon pickling according to claim 2 is characterized in that: in, The semantic token learner is used to adaptively learn key heat flow features, including: receiving heat flow features weighted by Gaussian attenuation mask; The two semi-directional sequences are merged using a cross-scanning method to obtain an integrated heat flow feature, including: integrating two subsequences from different sources into a sequence of original length through specific partitioning and splicing operations; The integrated heat flow feature data is compressed using a heat flow attention module, representative semantic tokens are extracted, key heat flow features are obtained, and adaptive downsampling is performed to optimize the heat energy transfer efficiency.

6. The thermal energy control system for the co-production of machine-made carbon and activated carbon pickling according to claim 5 is characterized in that: in, The heat flow attention module includes a channel attention mechanism and a spatial attention mechanism for capturing different heat flow patterns and spatial distribution characteristics; Through a series of heat flow pooling, feature convolution and nonlinear activation operations, key energy transfer patterns are learned and semantic tokens are generated; Adaptively downsampling the generated semantic tokens, including filtering or compressing the semantic tokens to optimize thermal energy transfer efficiency; The downsampled semantic tokens are processed using a multi-layer perceptron, and finally semantic tokens representing the heat flow pattern are output to obtain key heat flow features, which include thermal energy distribution in different regions, changes in heat flux density, and heat conduction path information.

7. The thermal energy control system for the co-production of machine-made carbon and activated carbon pickling according to claim 2 is characterized in that: in, The semantic token fuser receives the semantic tokens representing the heat flow pattern and the original heat source data and performs feature fusion to form an overall heat energy transmission sequence and outputs the fused efficient heat flow features, including: ， in, It is the efficient heat flow characteristic after fusion; is the original input feature, i.e. the original heat source data; It is a weighted pooling operation to extract key features; Operates for semantic token learners; Operates for the heat flow attention module; The semantic tokens output by the semantic token learner.

8. The thermal energy control system for the co-production of machine-made carbon and activated carbon pickling according to claim 7 is characterized in that: in, The heat flow attention module includes: a channel attention module and a thermal energy space attention module, which work together in the heat flow matching optimization process, and realize dynamic adjustment of energy flow and reduce heat loss through the combination of global and local information; The feature weighted fusion unit inputs the fused features into the multi-layer perceptron to train the heat control network model, thereby achieving accurate modeling and optimization of heat energy transmission, combining the thermal energy spatial attention module to adjust the local heat energy distribution, and using the heat flow sensitive distance to optimize the loss function to improve the heat matching accuracy, ultimately achieving stability and efficiency of heating control.

9. The thermal energy control system for the co-production of machine-made carbon and activated carbon pickling according to claim 8 is characterized in that: in, The channel attention module extracts features along the channel dimension using global average pooling and maximum pooling, and generates a channel attention map through multi-layer perceptron processing; The thermal energy spatial attention module generates a feature map by pooling along the channel dimension and concatenates and convolves to extract spatial attention features to obtain a spatial attention map.

10. A method for regulating heat energy of co-production of machine-made carbon and activated carbon pickling, characterized in that: The method includes: Real-time collection of heat source data in the carbon production process and heat demand data in the activated carbon acid washing process; Normalizing and energy feature mapping the heat source data and the heat demand data, extracting energy-related features and performing data segmentation to obtain two subsequences of heat supply and heat use; Construct a heat control network model to perform bidirectional modeling on the two subsequences of heat supply and heat use to extract heat flow characteristics, dynamically weight the heat flow characteristics corresponding to the heat energy distribution in different regions, ensure that the heat released by the machine-made charcoal is efficiently matched with the demand for activated carbon pickling, thereby optimizing the energy transmission path and outputting an optimized dynamic heat distribution plan; According to the optimized dynamic heat distribution scheme, the signal is controlled in real time to adjust the heat energy matching between the machine-made carbon production and the activated carbon pickling process.

Images