Waste incineration control and treatment method and system based on total waste calorific value

Abstract

This invention relates to the field of data processing technology, specifically to a waste incineration control and processing method and system based on the total calorific value of waste. This invention can accurately and timely determine the total calorific value label of waste, thereby obtaining accurate and forward-looking total calorific value, and ultimately achieving effective control of the waste incineration process. This invention combines the acquisition of multimodal information, the application of a total calorific value prediction model, and the generation of dynamic incineration control strategies. This not only improves the accuracy and timeliness of total calorific value prediction but also enables refined control of waste incineration based on the predicted total calorific value, maximizing energy efficiency while achieving stable output of thermal or electrical energy.

Description

Technical Field

[0001] The embodiments of the present invention relate to the field of data processing technology, and more specifically, to a waste incineration control and treatment method and system based on the total calorific value of waste. Background Technology

[0002] In the waste incineration process, accurate and real-time incineration control is crucial to ensuring incineration efficiency and resource utilization, which mainly relies on calorific value calculation. Regarding calorific value assessment, one existing technology determines the calorific value through sampling experiments; however, this method suffers from significant time lag and limited sample size. Another existing technology calculates the calorific value through energy balance methods, but this method also suffers from time lag. Therefore, determining the calorific value of the entire waste volume with forward-looking accuracy is a key technical issue to ensure incineration efficiency and resource utilization through controlled incineration. Summary of the Invention

[0003] In view of this, the present invention provides a waste incineration control and treatment method and system based on the total calorific value of waste.

[0004] To achieve the above objectives, embodiments of the present invention provide a waste incineration control and processing method based on the total calorific value of waste. The method is applied to a waste incineration control and processing system and includes: responding to a target control command by acquiring multimodal information of the waste incineration control task corresponding to the target control command; wherein the multimodal information includes image state factor data, environmental factor data, and waste attribute factor data; the environmental factor data includes seasonal factor data and weather factor data; the waste attribute factor data includes drainage duration factor data, location factor data, and morphological factor data; inputting the multimodal information into a target total calorific value prediction model to obtain a total waste calorific value label output by the target total calorific value prediction model; and generating a dynamic incineration control strategy for the waste incineration control task based on the total waste calorific value label.

[0005] This invention provides a waste incineration control and processing system, comprising: a modal information acquisition module, used to acquire multimodal information of a waste incineration control task corresponding to a target control command in response to the target control command; wherein the multimodal information includes image state factor data, environmental factor data, and waste attribute factor data; the environmental factor data includes seasonal factor data and weather factor data; the waste attribute factor data includes drainage duration factor data, location factor data, and morphological factor data; a full-volume tag output module, used to input the multimodal information into a target full-volume calorific value prediction model to obtain full-volume waste calorific value tags output by the target full-volume calorific value prediction model; and a control strategy generation module, used to generate a dynamic incineration control strategy for the waste incineration control task based on the full-volume waste calorific value tags.

[0006] The waste incineration control and processing method and system based on the total calorific value of waste provided in this invention can accurately and timely determine the total calorific value label of waste, thereby obtaining accurate and forward-looking total calorific value, and thus achieving effective control of the waste incineration process. This invention combines the acquisition of multimodal information, the application of a total calorific value prediction model, and the generation of dynamic incineration control strategies. This not only improves the accuracy and timeliness of total calorific value prediction, but also enables refined control of waste incineration based on the predicted total calorific value, maximizing energy efficiency while achieving stable output of thermal or electrical energy. Attached Figure Description

[0007] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the embodiments will be briefly introduced below. It should be understood that the following drawings only show some embodiments of the present invention and should not be regarded as a limitation on the scope. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort.

[0008] Figure 1 This is a flowchart illustrating a waste incineration control and treatment method based on the total calorific value of waste, provided in an embodiment of the present invention.

[0009] Figure 2 This is a block diagram of a waste incineration control and treatment system provided in an embodiment of the present invention. Detailed Implementation

[0010] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. The components of the embodiments of the present invention described and shown in the accompanying drawings can generally be arranged and designed in various different configurations.

[0011] Therefore, the following detailed description of the embodiments of the invention provided in the accompanying drawings is not intended to limit the scope of the claimed invention, but merely to illustrate selected embodiments of the invention. All other embodiments obtained by those skilled in the art based on the embodiments of the invention without inventive effort are within the scope of protection of the invention.

[0012] like Figure 1 As shown, this embodiment of the invention provides a waste incineration control and treatment method based on the total calorific value of waste, which is applied to the above-mentioned waste incineration control and treatment system. The specific steps to implement this method are as follows: steps 101-103.

[0013] Step 101: The waste incineration control and processing system responds to the target control command and acquires multimodal information of the waste incineration control task corresponding to the target control command.

[0014] In this embodiment of the invention, the multimodal information includes image state factor data, environmental factor data, and waste attribute factor data; the environmental factor data includes seasonal factor data and weather factor data; the waste attribute factor data includes drainage duration factor data, location factor data, and morphological factor data.

[0015] For example, when a waste incineration control and processing system receives a target control command, it begins acquiring multimodal information about the waste incineration control task corresponding to that command. Specifically, image-based dynamic factor data includes information such as the appearance of the waste. For instance, images of waste in a waste dumping area or on a waste conveyor belt, captured by a camera, can reflect the approximate composition of the waste and whether large non-incinerable materials are mixed in. If the image shows a large amount of white, plastic-like lumps in the waste, these lumps can provide a reference for subsequent calorific value estimation and incineration strategy formulation.

[0016] In addition, seasonal factors in environmental data have a certain impact on the calorific value of waste. In some seasons, due to lower temperatures, the moisture in some waste may freeze, which affects the overall calorific value of the waste. The moisture content of waste varies in different seasons; for example, season A may have 20% less moisture than season B, which may result in a relatively higher calorific value. Weather factors are also important. In humid weather, waste absorbs moisture more easily, thus lowering its calorific value. For example, after several consecutive days of rainfall, the moisture content of waste may increase significantly, making its calorific value significantly lower than that of the same type of waste under sunny and dry weather.

[0017] Furthermore, the drainage time factor in the waste attribute factor data is directly related to the moisture content of the waste. If the drainage time is short, there will be more residual moisture. For example, some kitchen waste that is only drained for one day may have a significantly higher moisture content than the same type of waste that has been drained for three days, which will greatly reduce the calorific value of the waste. Location factor data can reflect the source of the waste; waste generated in different regions and locations may differ in composition and calorific value. For example, the calorific value distribution of household waste in different areas, and even different types of waste (such as garden waste, sludge, etc.), varies considerably. Morphological factor data refers to the data obtained by further classifying and identifying waste (especially identifying foreign objects in the waste) through camera and image recognition technology, thereby providing a reference for predicting the calorific value of waste.

[0018] Step 102: The waste incineration control and processing system inputs the multimodal information into the target total calorific value prediction model to obtain the total calorific value label of the waste output by the target total calorific value prediction model.

[0019] In this step, the waste incineration control and processing system inputs the acquired multimodal information into the target total calorific value prediction model. This model, trained with data, accurately outputs the total calorific value label of the waste based on the input multimodal information.

[0020] For example, the target total calorific value prediction model collects different multimodal information samples and corresponding actual calorific value measurements of waste during its construction. For an exemplary combination of multimodal information, the target total calorific value prediction model can accurately determine the calorific value range of the waste. For instance, if the input image state factor data shows a large amount of dry paper in the waste, the environmental factor data indicates a dry season and sunny weather, and the waste attribute factor data shows a long drainage time, origin from a residential area, and the waste is in block form, the target total calorific value prediction model may output a higher total waste calorific value label, such as within the range of k1 to k2 kcal per kilogram, where k1 and k2 are positive numbers.

[0021] Step 103: The waste incineration control and processing system generates a dynamic incineration control strategy for the waste incineration control task based on the total calorific value tags of the waste.

[0022] Finally, the waste incineration control and processing system generates dynamic incineration control strategies based on the total calorific value tags of all waste. If the total calorific value tags indicate a high waste calorific value, the incineration control strategy can appropriately increase the feed rate of the incinerator while adjusting the supply of combustion air. For example, when the waste calorific value is high, the bed thickness can be reduced by a certain percentage to ensure complete combustion of the waste in the incinerator and a stable output of heat or electricity. If the waste calorific value is low, it may be necessary to increase the bed thickness and the input of primary air to maintain a stable incineration process. For example, when the waste calorific value is low, the bed thickness and primary air frequency can be increased by a certain percentage to ensure the normal operation of the incinerator and the stability of heat output.

[0023] As can be seen, based on the above steps 101-103, the calorific value of waste can be obtained in advance and in real time. In terms of process, this can greatly improve the level of fine control and energy efficiency of the incineration process and stably output heat or electricity.

[0024] By applying the above embodiments, through comprehensive acquisition of multimodal information and the use of a full-scale calorific value prediction model to derive accurate calorific value labels, a dynamic incineration control strategy can be generated. This allows for the real-time acquisition of the full calorific value of waste in advance, significantly improving the precision control and energy efficiency of the incineration process and ensuring stable output of heat or electricity. In other words, the embodiments of this invention can improve resource utilization efficiency, reduce energy waste, and simultaneously enhance the stability and safety of the waste incineration process.

[0025] It is understandable that, with the continuous changes in waste sources and the environment, the waste incineration control and treatment system will re-acquire multimodal information and perform subsequent operations each time it receives a new target control instruction. For example, with the changing seasons, the composition and calorific value of waste will change, and the waste incineration control and treatment system will adjust its incineration control strategy in a timely manner based on the new multimodal information. In municipal waste treatment plants, a large number of different types of waste enter the incineration plant every day. This ensures that each batch of waste receives optimal incineration treatment in terms of both calorific value utilization and environmental indicator control.

[0026] It's worth noting that in constructing the full-volume calorific value prediction model, in addition to collecting multimodal information samples and corresponding calorific value measurements, the data may also be classified and preprocessed. For example, waste data from different seasons and different sources can be labeled and analyzed separately so that the full-volume calorific value prediction model can better learn the changing patterns of waste calorific value under different conditions. Simultaneously, the full-volume calorific value prediction model may employ a fusion of multiple algorithmic techniques, such as combining neural network algorithms and decision tree algorithms. Neural network algorithms can automatically learn complex nonlinear relationships in multimodal information, while decision tree algorithms can provide highly interpretable decision logic. Therefore, the full-volume calorific value prediction model can more accurately output the calorific value label for all waste.

[0027] Furthermore, the generation of dynamic incineration control strategies may involve adjusting the feed rate and combustion air supply based on the waste's calorific value label, as well as controlling the temperature distribution within the incinerator, adjusting the incineration time, and optimizing exhaust gas treatment. If the waste has a high calorific value, the bed thickness, oxygen content, and incineration time can be adjusted, while ensuring complete combustion, to guarantee more stable main steam flow and furnace temperature, and reduce pollutant generation.

[0028] In other words, the waste incineration control and treatment system operates continuously in a cycle during the waste incineration process and constantly adjusts the incineration control strategy according to the actual situation of the waste, thereby achieving refined management of the waste incineration process and improving the overall efficiency of waste incineration treatment (including energy output efficiency, environmental protection efficiency, and equipment operation stability efficiency).

[0029] In the daily operation of waste treatment plants, regular maintenance and updates are necessary to ensure the reliability of the waste incineration control and treatment system. Equipment acquiring multimodal information, such as cameras, environmental sensors, and waste property detection devices, requires regular inspection and calibration to ensure data accuracy. Similarly, the total calorific value prediction model needs continuous optimization. As waste composition and environmental factors change, new data samples are continuously introduced to train the model and improve its generalization ability. For example, with the implementation of waste sorting policies, the composition of waste changes significantly, requiring timely updates to the total calorific value prediction model to adapt to new waste types and calorific value distributions. Dynamic incineration control strategies also need to be adjusted based on actual operational results and new environmental requirements to ensure optimal incineration performance under different waste treatment conditions.

[0030] From the perspective of the entire waste incineration industry chain, for upstream waste collection and sorting industries, more accurate waste calorific value information can promote more scientific and rational waste collection and sorting, increasing the recycling value of waste. For downstream energy utilization and environmental protection industries, stable heat or electricity output and better control of environmental indicators can improve the efficiency and quality of energy utilization and reduce negative environmental impacts. For example, stable heat output can better meet the needs of heating companies, reduce fluctuations in energy supply, and improve the stability and comfort of heating. In terms of environmental protection, better exhaust gas treatment can reduce pressure on the atmospheric environment and help improve regional environmental quality.

[0031] In practical applications, it is also necessary to consider the specific circumstances of different regions. For example, the quality of municipal solid waste in large cities is often better than that in small counties; even within the same city, the quality of municipal solid waste varies between different areas. Therefore, the construction of the total calorific value prediction model and the generation of dynamic incineration control strategies need to be optimized for the characteristics of different regions. For instance, the quality of waste in small counties is relatively poor, and the calorific value fluctuates more significantly. Therefore, special optimizations are needed to address calorific value fluctuations during dynamic control.

[0032] In summary, by applying steps 101-103, the total calorific value of waste can be accurately and promptly determined, thereby obtaining accurate and forward-looking total calorific value and enabling effective control of the waste incineration process. Specifically, the method combines the acquisition of multimodal information, the application of a total calorific value prediction model, and the generation of dynamic incineration control strategies. This not only improves the accuracy and timeliness of total calorific value prediction but also enables refined control of waste incineration based on the predicted total calorific value, maximizing energy efficiency while achieving stable output of thermal or electrical energy.

[0033] Furthermore, from an economic perspective, it can reduce labor and experimental costs, increase profits, and extend equipment lifespan. From an environmental perspective, it can help better control emissions, reduce pollutant generation, and comply with green production requirements. From a safety perspective, it can prevent equipment damage or fire risks caused by localized overheating. From a management perspective, it can improve the layout of a smart factory, enabling refined zoning management and scheduling of waste storage areas.

[0034] In this embodiment of the invention, the application of the target full calorific value prediction model is the core invention point. The target full calorific value prediction model can obtain the full calorific value data of the waste in advance, thereby providing an accurate and reliable decision basis for the generation of dynamic incineration control strategy. On this basis, the training steps of the target full calorific value prediction model include (1)-(4).

[0035] (1) Obtain sample multimodal information for sample waste incineration control task; wherein, the sample multimodal information includes sample image state factor data, sample environmental factor data and sample waste attribute factor data; the sample environmental factor data includes sample seasonal factor data and sample weather factor data; the sample waste attribute factor data includes sample drainage duration factor data, sample location factor data and sample morphology factor data.

[0036] (2) Input the sample multimodal information into the initial full calorific value prediction model to obtain the full calorific value prediction result output by the initial full calorific value prediction model.

[0037] (3) Construct a composite loss function using the full calorific value prediction results and the calorific value between time delay compensation.

[0038] The composite loss function includes a calorific value error loss function, a regularization loss function, and an imbalance loss function. For example, the composite loss function L = q1 * calorific value error loss function L1 + q2 * regularization loss function L2 + q3 * imbalance loss function L3. Here, q1, q2, and q3 are dynamic loss weights, and q1 + q2 + q3 = 1.

[0039] (4) Iteratively update the model parameters of the initial full calorific value prediction model in the direction of convergence of the composite loss function until the preset convergence condition is met, and obtain the target full calorific value prediction model.

[0040] For example, in the training process of a target full calorific value prediction model, the first step involves acquiring multimodal information from the samples. Sample image mode factor data is obtained through image acquisition devices, capturing information related to the appearance features of the waste. Taking common waste types as an example, image recognition algorithms can identify green and gray / black waste, as well as large foreign objects (such as sofas, large stones, gas cylinders, etc.) and ordinary household waste. Then, these different image features are quantified; for instance, object detection algorithms and segmentation algorithms can be used to determine the category labels of waste at different locations in the image.

[0041] For example, the seasonal factor data in the environmental factor data can be further refined based on seasonal characteristics such as temperature and humidity. The drainage duration factor data in the waste attribute factor data can be calculated in hours or days. The location factor is used to characterize the difference in moisture content of waste at different locations within the waste bin. For instance, because the fermentation of waste in zones A and B differs, the moisture content of waste in zones A and B will also differ. Furthermore, even within zone A, the moisture content of surface waste differs from that of waste at different depths; this difference in moisture content can lead to differences in the calorific value of the waste.

[0042] In this embodiment of the invention, after the quantified multimodal information of the samples is input into the initial full-calorific-value prediction model, the model outputs the full-calorific-value prediction result. The full-calorific-value prediction result is obtained based on the algorithm structure within the initial full-calorific-value prediction model and the learning from the sample data. The initial full-calorific-value prediction model may contain multiple hidden layers, with neurons in each hidden layer performing different processing on the input data. For example, neurons in the first hidden layer may perform a weighted summation of color values ​​in the image state factor data, while neurons in the second hidden layer may perform nonlinear transformations on seasonal and weather-related values ​​in the environmental factor data. The results of these processing steps are then aggregated at the output layer to obtain the full-calorific-value prediction result.

[0043] Next, this embodiment of the invention constructs a composite loss function using a calorific value error term, a positive lateralization error term, and a sample equalization error term. The calorific value error loss function within the composite loss function measures the difference between the predicted calorific value and the time-delay compensated calorific value. For example, if the calorific value calculated by time-delay compensation is k3 kcal per kilogram, while the model predicts k4 kcal per kilogram, then the difference of (k4-k3) kcal can be included in the calculation of the calorific value error loss function. For instance, if the calorific value error loss function uses the mean squared error (MSE) calculation method, the square of this difference will be included as part of the calculation in the composite loss function.

[0044] It's important to note that the regularization loss function prevents overfitting. During model training, if the model overlearns certain features from the sample data, it may perform poorly on new data. For example, if there's a specific set of garbage samples in the sample data, which has an exemplary combination of image state factors, environmental factors, and garbage attribute factors, and this set of samples constitutes a large proportion of the sample set, the model might overfit to the features of this set. The regularization loss function is calculated based on the size and complexity of the model parameters. If a certain weight parameter of the model is too large, the value of the regularization loss function will increase, prompting the model to adjust this parameter during training to avoid over-reliance on the features of specific samples.

[0045] Furthermore, the imbalanced loss function primarily addresses the problem of imbalanced distribution of different categories of data in the sample data. For example, in the calorific value sample data of garbage, there might be a significant difference in the number of high-calorific-value garbage samples and low-calorific-value garbage samples. Or, for instance, high-calorific-value garbage samples might only account for 10% of the total samples, while low-calorific-value garbage samples might make up 90%. In this situation, the model might tend to better fit the larger proportion of low-calorific-value garbage samples, while reducing the prediction accuracy for high-calorific-value garbage samples. Therefore, the imbalanced loss function is calculated based on this imbalanced sample distribution, allowing the model to treat garbage samples with different calorific values ​​more evenly during training.

[0046] In this embodiment of the invention, the dynamic loss weights in the composite loss function L = q1 * calorific value error loss function L1 + q2 * regularization loss function L2 + q3 * imbalance loss function L3 are q1, q2, and q3. Different weight settings reflect the degree of importance attached to different loss functions. For example, when q1 = 0.6, it indicates that more emphasis is placed on the calorific value error loss function during model training, hoping that the model can predict the calorific value more accurately; when q2 = 0.2, it indicates that some importance is also attached to preventing overfitting; when q3 = 0.2, it indicates that the problem of sample imbalance is also taken into account.

[0047] Furthermore, the model parameters of the initial full-calorific-value prediction model can be iteratively updated in the direction of convergence of the composite loss function until a preset convergence condition is met, thus obtaining the target full-calorific-value prediction model. During the iterative update process, each parameter of the model is adjusted according to the gradient of the composite loss function. For example, a certain weight parameter in the model, initially valued at 0.05, may be increased by 0.01 in one iteration based on the calculation result of the composite loss function and the gradient descent algorithm, resulting in an updated parameter value of 0.06. This iterative process continues, with each iteration adjusting the model parameters based on new multimodal information input from samples, the model's prediction results, and the calculation result of the composite loss function. The preset convergence condition could be that the value of the composite loss function is less than a very small value, such as 0.001, or that the model's accuracy on the validation set reaches a certain quantitative standard, such as above 95%.

[0048] It is understandable that the target total calorific value prediction model constructed and trained through the above steps can accurately output the total calorific value label of waste based on the input multimodal information. During waste incineration, accurate waste calorific value data allows for fine-tuning of incineration control strategies. For example, if the waste calorific value is high, such as above 5 kcal per kilogram, the incineration control strategy can include "thinning the feed layer" to ensure the stability of the main steam flow and furnace temperature; if the waste calorific value is low, such as below 6 kcal per kilogram (k6 is less than k5), the incineration control strategy can include "thickening the feed layer" to ensure the stability of incineration.

[0049] Therefore, by applying the technical solutions described in (1)-(4) above, multimodal information of samples is constructed by integrating multiple factor data, and a composite loss function containing multiple loss functions is used to train the model, significantly improving the accuracy and generalization ability of the model. This enables the model to adapt to the forward-looking prediction needs of total calorific value under different regions, seasons, and waste sources and compositions. In addition, accurate calorific value prediction provides a key basis for waste incineration control, enabling the incineration process to be dynamically adjusted according to the actual calorific value of the waste, improving incineration efficiency and reducing energy waste. Thus, it can not only improve the economic benefits of waste incineration plants, but also stabilize the output of heat or electricity, thereby meeting the stability of energy demand and energy supply.

[0050] In a preferred embodiment, the calorific value error loss function is determined by the difference between the full calorific value prediction result and the calorific value between the time-delay compensation intervals; the determination method for the calorific value between the time-delay compensation intervals includes:

[0051] Step 201: Determine the time-delayed calorific value data y_delay(t) based on the sample incineration heat input data Q_in, system heat loss data Q_out, and waste quality data m_waste; the time-delayed calorific value data y_delay(t) is the calorific value of waste with time delay at time t.

[0052] Step 202: Determine the time delay variable delta_t based on the first time delay weight delta_v and the second time delay weight delta_bias; the first time delay weight delta_v is determined based on the feeder time delay delta_pusher and the grate time delay delta_grate, and the second time delay weight delta_bias is the bias value.

[0053] Step 203: Dynamically supplement and correct the time-delay waste calorific value data y_delay(t) using the time-delay variable delta_t to obtain the time-delay compensation interval calorific value; wherein, the time-delay compensation interval calorific value is y_adjusted(t)=y_delay(t+delta_t).

[0054] In this embodiment of the invention, the sample incineration heat input data Q_in, the system heat loss data Q_out, and the waste quality data m_waste are used to determine the time-delay waste calorific value data y_delay(t). In an exemplary waste incineration process, the above data, based on the principle of energy conservation, may include the following relationship: y_{delay}(t)=frac{Q_{in}-Q_{out}}{m_{waste}}.

[0055] In an optional but non-restrictive example, when the sample incineration heat input data Q_in is 5000 kcal, the system heat loss data Q_out is 1000 kcal, and the waste mass data m_waste is 2 kg, the relationship can be used to calculate y_{delay}(t)=frac{5000-1000}{2}=2000 kcal. 2000 kcal can be understood as the waste calorific value data y_delay(t) with time delay at time t, which reflects the waste calorific value under certain delay due to various factors during the incineration process.

[0056] In the following steps, the first time-delay weight delta_v is determined based on the feeder time delay delta_pusher and the grate time delay delta_grate, and the second time-delay weight delta_bias is a bias value. The feeder time delay delta_pusher and the grate time delay delta_grate are two key time-delay factors in a waste incineration system. For example, the first time-delay weight delta_v can be determined by the existence of a linear combination relationship between the feeder time delay delta_pusher and the grate time delay delta_grate, which can be exemplified as: delta_v = k_1 * delta_pusher + k_2 * delta_grate, where k_1 and k_2 are coefficients determined based on system characteristics. For example, if the feeder delay delta_pusher is 3 seconds and the grate delay delta_grate is 2 seconds, and k_{1} = 0.4 and k_{2} = 0.6, then delta_{v} = 0.4*3 + 0.6*2 = 2.4. The second delay weight delta_bias can be a fixed value, such as 0.5.

[0057] Then, the time delay variable delta_t is determined based on the two time delay weights mentioned above. An exemplary way to determine the time delay variable delta_t is as follows: delta_{t} = delta_{v} * delta_{bias}. Substituting the previously calculated delta_{v} = 2.4 and delta_{bias} = 0.5, we get delta_{t} = 2.4 * 0.5 = 1.2.

[0058] Then, the time-delayed waste calorific value data y_delay(t) is dynamically supplemented and corrected using the time-delay variable delta_t to obtain the time-delayed compensation calorific value y_adjusted(t) = y_delay(t + delta_t). This process is an optimization adjustment of the time-delayed waste calorific value data. For example, if y_delay(t) is 2000 kcal at a certain moment, such as t = 5 seconds, then after correction by the time-delay variable delta_t, y_adjusted(5) = y_delay(5 + 1.2) is calculated. Here, the value of y_delay(6.2) needs to be determined based on the relationship between y_delay(t) and time (which can be based on the physicochemical changes during the incineration process). For example, under this relationship, the value of y_delay(6.2) is 2100 kcal, and this change in value can be understood as a dynamic supplement and correction considering the time-delay factor.

[0059] In this embodiment of the invention, the calorific value error loss function is determined by the difference between the full calorific value prediction result and the calorific value obtained above with time lag compensation. For example, if the full calorific value prediction result is y_predicted, the calorific value error loss function can be obtained based on a certain calculation method of the difference between the two. Alternatively, it can be constructed using the idea of ​​mean squared error (MSE), where the calorific value error loss function is L1, L1 = frac{1}{n}sum_{i=1}^{n}(y_{predicted}(i)-y_{adjusted}(i))^{2}, where n is the number of samples. This design allows the calorific value error loss function to more accurately reflect the deviation between the model prediction result and the actual calorific value after considering the time lag factor.

[0060] It's understandable that when iteratively updating the model parameters of the initial full-calorific-value prediction model, when performing gradient descent towards the direction of convergence of the composite loss function, the update formula for the model parameter θ can be θ = θ - η * frac{partialL}{partialθ}, where η is the learning rate. This process continues, with each iteration adjusting the model parameters based on new multimodal input sample information, the model's prediction results, and the calculated results of the composite loss function. Preset convergence conditions could be that the value of the composite loss function is less than a certain value, such as 0.001, or that the model's accuracy on the validation set reaches a preset accuracy, such as above 95%.

[0061] Based on steps 201-203 above, by determining the calorific value between time-delay compensation intervals and subsequently determining the calorific value error loss function, the accurate construction of the target full-volume calorific value prediction model can be achieved. Specifically, the determination of the calorific value between time-delay compensation intervals, the collection of multimodal information from the samples, and the regularization / imbalance loss function in the composite loss function ensure that the target full-volume calorific value prediction model can accurately predict the calorific value of waste under the influence of various factors, thereby accurately determining the calorific value of the entire waste volume.

[0062] In this embodiment of the invention, when collecting multimodal information of samples, image acquisition of waste includes not only recording its appearance but also quantifying features such as color and texture. For example, image recognition algorithms can identify greenish-brown waste and grayish-black waste, and can also distinguish between large foreign objects (such as sofas, large stones, gas cylinders, etc.) and ordinary household waste, and so on. The sample environmental factor data includes sample seasonal factor data and sample weather factor data, as different seasons and weather conditions affect the calorific value of waste. For example, in winter, due to low temperatures, the moisture in the waste may freeze, resulting in a relatively low overall calorific value; while in summer, during sunny weather, the calorific value is higher. Furthermore, when determining the calorific value of the entire amount of waste, the sample drainage time factor data, sample location factor data, and sample morphology factor data in the sample waste attribute factor data also need to be considered.

[0063] It is understandable that after accurately predicting the total calorific value of waste using a target total calorific value prediction model, a waste incineration control strategy can be dynamically generated based on the predicted total calorific value. During waste incineration, if the waste calorific value is high, the feed layer can be appropriately thinned in the incineration control strategy. For example, when the waste calorific value is high, reducing the feed layer thickness by 10% ensures complete combustion of the waste in the incinerator and a stable output of heat or electricity. If the waste calorific value is low, it may be necessary to increase the feed layer thickness and increase the primary air input to maintain a stable incineration process. For instance, when the waste calorific value is low, increasing the feed layer thickness by 15% and the primary air frequency by 20% ensures the normal operation of the incinerator and the stability of heat output.

[0064] By applying the embodiments of this invention, the calorific value error loss function is accurately calculated by precisely constructing a method for determining the calorific value between time-delay compensation intervals, taking into account various factors such as sample incineration heat input, system heat loss, waste quality, feeder time delay, and grate time delay. This improves the accuracy of the target total calorific value prediction model, thereby achieving accurate waste calorific value prediction. Accurate calorific value prediction can provide an execution reference for waste incineration control strategies, thereby achieving efficient and stable incineration, improving energy utilization efficiency, stabilizing heat or electricity output, and better controlling various parameters in the waste incineration process to adapt to different waste characteristics and incineration requirements.

[0065] In another preferred embodiment, the step of inputting the sample multimodal information into an initial full calorific value prediction model to obtain the full calorific value prediction result output by the initial full calorific value prediction model includes:

[0066] Step 301: Call the embedding layer of the initial full-volume calorific value prediction model to embed the sample multimodal information to obtain the sample multimodal embedding features corresponding to the sample multimodal information.

[0067] Step 302: Call the self-attention layer of the initial full-volume heat value prediction model to perform attention weighting on the sample multimodal embedding features, and obtain the attention-enhanced embedding features corresponding to the initial embedding features of each modality in the sample multimodal embedding features.

[0068] Step 303: Connect each attention-enhanced embedding feature and then perform a linear transformation to obtain the target fusion feature; wherein, the target fusion feature is X_fused=concat(attention_1,attention_2,...,attention_m), and attention_m is the attention-enhanced embedding feature of the m-th modality.

[0069] Step 304: Call the fully connected layer of the initial full calorific value prediction model to process the target fusion features and obtain the full calorific value prediction result.

[0070] For example, in step 301, the embedding layer of the initial full-volume calorific value prediction model is first invoked to embed the multimodal information of the samples, thereby obtaining the multimodal embedding features of the samples corresponding to the multimodal information. The multimodal information of the samples covers a variety of different types of data (such as sample image state factor data, sample environmental factor data, and sample garbage attribute factor data).

[0071] For sample image state factor data, it contains various visual information about the appearance of waste. This includes information such as the color distribution, shape, and texture of the waste. Taking waste color as an example, there may be a mixture of objects of various colors in the waste, such as white paper, black plastic, and green plant debris. This color information exists in its raw, intuitive visual form, but for the model, it needs to be converted into a form that can be computed and processed.

[0072] For example, in an image, the color value of the white paper portion is (255, 255, 255) in the RGB color space, while the black plastic portion is (0, 0, 0). When processed by the embedding layer, these color values ​​are mapped to an exemplary feature vector space. For instance, the color information of the white paper might be mapped to a vector containing multiple values, one of which might be 10. This 10 is obtained after complex mapping and transformation by the embedding layer, and it is associated with other features of the white paper in the overall multimodal information and its relationship with other waste components. Similarly, the color information of the black plastic, after processing by the embedding layer, might be mapped to another vector, where one value might be -5. This value reflects a unique embedding representation of the black plastic in the overall multimodal information of the sample. In terms of shape, blocky waste and filamentous waste differ significantly in shape features. Blocky waste might be mapped to a set of exemplary values ​​in the shape feature dimension, such as [1, 2, 3], while filamentous waste might be mapped to [-1, -2, -3]. These values ​​not only reflect the features of the shape itself but also have potential correlations with other modal information.

[0073] For example, the seasonal factor data within the environmental factor data is equally important, as different seasons significantly impact the characteristics of waste. For instance, in winter, moisture in waste may freeze due to low temperatures, altering its humidity and density, and consequently affecting its calorific value. For example, encoding the four seasons as spring, summer, autumn, and winter (spring = 1, summer = 2, autumn = 3, winter = 4) would convert these encodings into corresponding feature vectors after embedding. Winter, after embedding, might have multiple values ​​in a single feature vector, such as [-2, 3, -1]. These values ​​could be related to various factors such as winter temperature, sunshine duration, and precipitation. For example, -2 might be associated with lower winter temperatures, 3 with shorter winter sunshine duration, and -1 with relatively less winter precipitation.

[0074] For the sample weather factor data, the humidity, temperature, and other characteristics of garbage will change under different weather conditions such as sunny, cloudy, and rainy days. Taking a sunny day as an example, garbage may experience faster moisture evaporation due to sunlight. For example, the original code for a sunny day might be 1. After processing by the embedding layer, its corresponding feature vector might contain values ​​such as [5, -3, 2]. Here, 5 might be related to sunlight intensity, -3 might be related to the decrease in humidity, and 2 might be related to the increase in temperature.

[0075] Furthermore, the sample draining time factor data in the sample waste attribute factor data has a direct impact on the moisture content of the waste. For example, waste drained for 1 hour and waste drained for 3 hours have significantly different moisture contents. Waste drained for 1 hour is represented as 1 in the original data. After processing by the embedding layer, its corresponding feature vector may contain values ​​[1, 0, -1]. Here, 1 may be related to the proportion of moisture in the waste after 1 hour of draining, 0 may be related to other physical properties during the draining process, and -1 may be related to the looseness of the waste after 1 hour of draining.

[0076] For example, the sample location factor data, such as whether the garbage comes from a residential area, commercial area, or industrial area, will have different garbage composition and calorific value characteristics. Taking garbage from a residential area as an example, it may contain more kitchen waste, paper, plastic and other daily household waste. For example, the original code for a residential area is 10. After the embedding layer is processed, its corresponding feature vector may have the value [2, 3, -1], where 2 may be related to the proportion of kitchen waste, 3 may be related to the proportion of paper and plastic, and -1 may be related to the proportion of other special garbage components.

[0077] In addition, different forms of waste, such as blocky, powdery, and filamentous waste, in the sample morphology factor data have different combustion characteristics and calorific value release. Blocky waste may be represented as 1 in the original representation, but after the embedding layer is processed, it may have a value of [3, -1, 2] in its corresponding feature vector. Among them, 3 may be related to the bulk density of blocky waste, -1 may be related to the ventilation performance of blocky waste, and 2 may be related to the calorific value release rate of blocky waste.

[0078] It is understandable that by processing the multimodal information of the samples through the embedding layer, multimodal embedding features of the samples are obtained. These multimodal embedding features contain rich information after the original multimodal information has been transformed, and the correlation between different types of data has been initially integrated.

[0079] Next, in step 302, the self-attention layer of the initial full-volume heat value prediction model is called to perform attention weighting on the sample multimodal embedding features, thereby obtaining the attention-enhanced embedding features corresponding to the initial embedding features of each modality in the sample multimodal embedding features.

[0080] In summary, the self-attention layer in this invention functions by applying targeted weighting to different features based on the interrelationships between elements in the multimodal embedding features of the samples. Taking the embedding features corresponding to the sample image state factor data as an example, the embedding features of the image state factor data exhibit certain correlations with the embedding features of other modalities. For instance, in the multimodal embedding features of the samples, a certain element in the embedding features corresponding to the image state factor data represents the proportion of white paper in the waste, and its value is 3 in the original embedding features. When the self-attention layer analyzes the entire multimodal embedding features of the samples, it finds a strong correlation between this element representing the proportion of white paper and the seasonal factor data in the sample environmental factor data (e.g., winter, because paper may be drier and more flammable in winter, significantly affecting its calorific value) and the drainage duration factor data in the sample waste attribute factor data (a longer drainage duration may reduce the moisture content of the paper, also affecting its calorific value). Therefore, the self-attention layer weights this element representing the proportion of white paper based on this correlation, adjusting its value from 3 to 4.

[0081] A similar attention-weighting process applies to the embedded features corresponding to the sample environmental factor data. For example, in the embedded features of the sample environmental factor data, one element represents the impact of summer sunshine duration on waste moisture. In the original embedded features, this element has a value of -2. When the self-attention layer analysis finds that this element is associated with the morphological factor data in the sample waste attribute factor data (e.g., powdery waste may be easier to dry under high summer temperatures and sunshine, changing its combustion characteristics and calorific value), the element will be weighted and its value adjusted to -1.

[0082] For the embedded feature element corresponding to the sample draining time factor data in the sample waste attribute factor data, for example, the original value of 1 indicates the degree of influence of short draining time on the moisture content of waste. When the self-attention layer analysis finds that this element is related to the waste color information in the sample image state factor data (for example, darker colored waste may still contain more moisture after draining because of the characteristic of color absorbing heat) and the weather factor data in the sample environmental factor data (such as rainy days may lead to poor draining effect), the element will be weighted and its value will be adjusted to 2.

[0083] By employing this self-attention layer to weight the initial embedding features of each modality, attention-enhanced embedding features corresponding to the initial embedding features of each modality are obtained. These attention-enhanced embedding features highlight the feature information that is associated with other modalities and has a significant impact on the prediction results of the total calorific value.

[0084] Then, in step 303, each attention-enhanced embedding feature is concatenated and then linearly transformed to obtain the target fusion feature. For example, there are three modal attention-enhanced embedding features, namely attention_1, attention_2, and attention_3. attention_1 is the attention-enhanced embedding feature of the sample image state factor data after the above processing, for example, it contains values ​​[1, 2, 3, 4]; attention_2 is the attention-enhanced embedding feature corresponding to the sample environment factor data, for example, its value is [5, 6, 7, 8]; attention_3 is the attention-enhanced embedding feature corresponding to the sample garbage attribute factor data, for example, its value is [9, 10, 11, 12]. The target fusion feature X_fused = concat(attention_1, attention_2, attention_3) is concatenated to obtain a new feature vector [1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12]. This target fusion feature integrates the important information of each modality after attention enhancement, forming a comprehensive feature representation.

[0085] Finally, in step 304, the fully connected layer of the initial full-calorific-value prediction model is invoked to process the target fusion features, obtaining the full-calorific-value prediction result. The fully connected layer consists of multiple neurons, each connected to every element in the target fusion features. The neurons in the fully connected layer have different weights, which are continuously adjusted during model training to achieve accurate processing of the target fusion features.

[0086] For example, a neuron in a fully connected layer might have a weight of 0.5, and an element in the target fusion feature might have a weight of 2. Multiplying them together yields 1 (which can be understood as the neuron's processing result for that element). Each neuron in the fully connected layer performs similar calculations on elements in the target fusion feature, and then these results are summarized and further processed. For instance, in a fully connected layer with 10 neurons, each neuron calculates the target fusion feature to obtain 10 intermediate results, such as [1, 2, 3, 4, 5, 6, 7, 8, 9, 10]. Then, through further calculations (such as summation, nonlinear transformations, etc.), the final full calorific value prediction result is obtained. This full calorific value prediction result can accurately reflect the calorific value of the waste corresponding to the multimodal information of the sample, providing an important basis for subsequent model training (such as calculating the composite loss function, etc.) and the formulation of waste incineration control strategies.

[0087] Steps 301-304 are applied to process the multimodal information of the samples through an embedding layer, a self-attention layer, and a fully connected layer to obtain the full calorific value prediction result. The embedding layer transforms the multimodal information into a form suitable for model processing, the self-attention layer captures the correlations in the multimodal embedded features and strengthens important features, and the fully connected layer performs comprehensive calculations on the fused features to obtain the prediction result. This effectively integrates different features from the multimodal information, improves the accuracy of full calorific value prediction, and provides a reliable basis for waste incineration control. It helps improve incineration efficiency, optimize energy utilization, stabilize heat or electricity output, and better adapt to the impact of different waste characteristics and environmental factors on calorific value.

[0088] Based on the above, in an independent embodiment, generating a dynamic incineration control strategy for the waste incineration control task based on the total calorific value tags of the waste includes:

[0089] Step 401: Determine the incineration stage division results based on the numerical range of the total waste calorific value label. The incineration stage division results include the heating stage, steady state stage, and cooling stage.

[0090] In this embodiment of the invention, when the calorific value of the total waste is in the higher calorific value range, the system determines that the current incineration stage is the heating stage, at which point the furnace temperature needs to be rapidly increased to a preset threshold. When the calorific value is in the middle calorific value range, the system determines that it is the steady-state stage, requiring the maintenance of combustion efficiency and furnace temperature stability. When the calorific value is in the lower calorific value range, the system determines that it is the cooling stage, requiring the furnace temperature to be gradually reduced to avoid equipment overheating or excessive pollutant emissions. The division of incineration stages based on the dynamic comparison between the calorific value label and the preset threshold ensures that the incineration process is accurately matched with the calorific value characteristics of the waste.

[0091] Step 402: Based on the incineration stage division results, match the corresponding combustion control sub-strategy set. The combustion control sub-strategy set includes the feed rate baseline parameter, primary air volume baseline parameter, and secondary air volume baseline parameter. Perform environmental compensation correction on the combustion control sub-strategy set according to the seasonal factor data and weather factor data in the environmental factor data to generate an environmentally compensated combustion control sub-strategy set. Perform material characteristic compensation correction on the environmentally compensated combustion control sub-strategy set according to the dewatering time factor data and morphology factor data in the waste attribute factor data to generate a material-compensated combustion control sub-strategy set.

[0092] In this embodiment of the invention, the system performs environmental compensation correction on the combustion control sub-strategy set based on seasonal and weather factor data from the environmental factor data. For example, in low-temperature or high-humidity environments, the system adds a primary air volume baseline parameter to compensate for the negative impact of the environment on combustion efficiency, generating an environmentally compensated combustion control sub-strategy set.

[0093] Furthermore, the system combines the drainage duration factor data and morphology factor data from the waste attribute factor data to perform material characteristic compensation and correction on the above strategies. For example, for waste with a short drainage duration and loose morphology, the system reduces the feed rate baseline parameter to extend the combustion residence time, generating a set of combustion control sub-strategies after material compensation, thereby adapting to the physical characteristics of the waste.

[0094] Step 403: Based on the flame brightness distribution data and combustion zone temperature data in the image state factor data, perform combustion state feedback correction on the material-compensated combustion control sub-strategy set to generate a feedback-corrected combustion control sub-strategy set; verify the time-series matching degree between the feedback-corrected combustion control sub-strategy set and the historical incineration control strategy, remove abnormal control parameters with matching degree lower than the preset threshold, and generate a time-series optimized combustion control sub-strategy set.

[0095] In this embodiment of the invention, the system performs combustion state feedback correction on the combustion control sub-strategy set after material compensation based on flame brightness distribution data and combustion zone temperature data from the image state factor data. Flame brightness distribution data reflects the differences in combustion intensity in different areas of the furnace, while combustion zone temperature data characterizes the real-time heat distribution state of each area. If incomplete combustion is detected in a local area, the system dynamically adjusts the primary air volume reference parameter for the corresponding area to enhance combustion uniformity, generating a feedback-corrected combustion control sub-strategy set.

[0096] Subsequently, the system verifies the timing matching degree between the revised strategy and the historical incineration control strategy. By comparing the parameter combinations in historical successful cases, abnormal control parameters that significantly deviate from the normal range are eliminated, and a set of timing-optimized combustion control sub-strategies is generated to ensure the timing continuity and reliability of the strategy.

[0097] Step 404: Based on the distribution characteristics of the waste dumping area corresponding to the location factor data, perform spatial distribution compensation on the time-optimized combustion control sub-strategy set to generate a spatially compensated combustion control sub-strategy set; input the spatially compensated combustion control sub-strategy set into the incineration equipment control parameter conversion model to generate an executable control instruction sequence containing feeder motion parameters, grate speed parameters, and fan speed parameters.

[0098] In this embodiment of the invention, the system performs spatial distribution compensation on the time-optimized combustion control sub-strategy set based on the distribution characteristics of the waste stacking area corresponding to the location factor data. The location factor data describes the distribution density and morphological characteristics of waste in the feeding area, such as the distribution differences between high-density and low-density areas. The system adjusts the feeder motion parameters and grate speed parameters according to the stacking characteristics of different areas, for example, slowing down the grate movement speed in high-density areas to prolong the combustion time, thereby generating a spatially compensated combustion control sub-strategy set.

[0099] Subsequently, the system inputs the spatially compensated strategy into the incineration equipment control parameter conversion model. This model, based on the mapping relationship between the equipment's physical characteristics and control parameters, converts the abstract parameters in the strategy into a sequence of executable control instructions, including feeder motion parameters, grate speed parameters, and fan speed parameters, ensuring that the instructions match the physical constraints of the equipment's actuators.

[0100] Step 405: Based on the actual execution effect data of the executable control command sequence in the target incineration equipment, perform online iterative optimization of the dynamic incineration control strategy to generate the final dynamic incineration control strategy.

[0101] In this embodiment of the invention, the system performs online iterative optimization of the dynamic combustion control strategy based on the actual execution effect data of the executable control command sequence on the target incineration equipment. The actual execution effect data includes real-time furnace temperature fluctuations, steam output stability, and pollutant emission concentrations. The system continuously collects operating data using a rolling time window algorithm, analyzes the deviation between the execution effect and the expected target, and dynamically fine-tunes the feed rate, air volume parameters, and grate speed to generate the final dynamic combustion control strategy. For example, if furnace temperature fluctuations exceed the allowable range, the system prioritizes adjusting the primary air volume parameters to stabilize combustion intensity, achieving closed-loop optimization of the strategy.

[0102] Step 406: Perform thermal balance verification between the final dynamic incineration control strategy and the total waste calorific value label. When the calorific value deviation exceeds the preset tolerance range, trigger the strategy reconstruction mechanism to re-execute steps 401-405.

[0103] In this embodiment of the invention, the system performs thermal balance verification between the final dynamic incineration control strategy and the calorific value labels of all waste. Thermal balance verification is achieved by comparing the deviation between the actual incineration heat output value and the predicted value from the calorific value labels. If the deviation exceeds a preset tolerance range, the system triggers a strategy reconstruction mechanism, re-executing the entire strategy generation process from incineration stage division to online optimization.

[0104] During the reconstruction process, the system prioritizes calibrating the input weights of the target full calorific value prediction model, enhances the fusion accuracy of environmental factor data and waste attribute factor data, and optimizes the mapping relationship of the incineration equipment control parameter conversion model until the heat balance verification is passed. This ensures that the dynamic incineration control strategy remains highly adapted to the calorific value characteristics of waste and real-time operating conditions.

[0105] As a scalable implementation, the specific steps of environmental compensation correction include: determining the environmental temperature compensation coefficient based on seasonal factor data, calculating the air density correction amount based on humidity data in weather factor data, and applying the environmental temperature compensation coefficient and air density correction amount to the primary air volume reference parameter and the secondary air volume reference parameter.

[0106] In the embodiments of the invention, seasonal factor data reflects the influence of external ambient temperature on the incinerator's inlet air temperature. For example, in low-temperature seasons, the ambient temperature compensation coefficient is used to increase the primary air volume reference parameter to compensate for the inhibitory effect of external cold air on combustion efficiency within the furnace. Humidity data in the weather factor data is collected in real time by a humidity sensor. The air density correction is calculated based on the characteristic that increased humidity leads to a decrease in oxygen content in the air. For example, in high-humidity weather, the system increases the secondary air volume reference parameter based on the air density correction to maintain the oxygen concentration required for combustion. By superimposing the combined effects of the ambient temperature compensation coefficient and the air density correction on the air volume parameters, the system dynamically adjusts the primary and secondary air volume reference parameters to ensure the adaptability of combustion air supply under different environmental conditions. For example, in a high-humidity environment in winter, the system simultaneously increases the primary and secondary air volume reference parameters to offset the combined negative impacts of low temperature and high humidity on the combustion reaction, maintaining combustion stability within the furnace.

[0107] As an scalable implementation, the specific steps of material property compensation correction include: calculating the material moisture content compensation value based on the draining time factor data, determining the material air permeability correction coefficient based on the crushed particle size distribution in the morphology factor data, and applying the moisture content compensation value and the air permeability correction coefficient to the feed rate reference parameter.

[0108] In this embodiment of the invention, the draining time factor data characterizes the degree of dehydration of waste during the pretreatment stage. For example, when the draining time is short, the system reduces the feed rate benchmark parameter by adjusting the material moisture content compensation value, thereby extending the residence time of waste in the incinerator to fully evaporate moisture. The particle size distribution in the morphology factor data is obtained through image recognition technology. For example, waste with larger particle sizes has a lower bulk density, and the system increases the feed rate benchmark parameter based on the material permeability correction coefficient to adapt to the rapid combustion characteristics caused by its loose structure.

[0109] It is understandable that by integrating the moisture content compensation value and the air permeability correction coefficient, the system makes bidirectional adjustments to the feed rate reference parameter. For example, for waste with high moisture content and uneven crushing particle size, the system simultaneously reduces the feed rate reference parameter to compensate for the moisture evaporation requirement and fine-tunes the feed rate reference parameter to adapt to the combustion rate fluctuation caused by particle size differences, thereby optimizing the synergy between material conveying and combustion processes.

[0110] As a scalable implementation, the specific steps of combustion state feedback correction include: identifying the coordinates of unburned areas through flame brightness distribution data, calculating the temperature gradient distribution based on combustion area temperature data, and generating secondary damper opening adjustment parameters based on the coordinates of unburned areas and the temperature gradient distribution.

[0111] In this embodiment of the invention, flame brightness distribution data is captured by an infrared camera to show the light intensity distribution of the flame within the furnace. For example, if the flame brightness on the left side of the grate is significantly lower than that on the right side, the system marks this area as the coordinates of an incompletely burned region. Combustion zone temperature data is collected using a distributed thermocouple array. The temperature gradient distribution reflects the combustion uniformity in different areas of the furnace; for example, the temperature gradient corresponding to the coordinates of an incompletely burned region exhibits localized low-temperature characteristics. Based on the coordinates of the incompletely burned region, the system locates the position of incomplete combustion and calculates the secondary damper opening adjustment parameters based on the temperature gradient distribution. For example, for low-temperature regions, the opening of the corresponding damper is increased to enhance oxygen supply and promote the secondary combustion of residual combustibles.

[0112] It is understandable that by matching the coordinates of the unburned area with the adjustment parameters of the secondary air damper opening in real time, the system dynamically corrects the combustion control strategy. For example, when the system detects continuous unburned coordinates in the rear furnace area, it gradually increases the opening of the secondary air damper in that area until the flame brightness and temperature gradient are restored to equilibrium, thereby eliminating the combustion blind zone and improving the overall combustion efficiency.

[0113] Based on the same inventive concept, a waste incineration control and processing system is also provided. The system includes: a modal information acquisition module, used to acquire multimodal information of the waste incineration control task corresponding to the target control command in response to the target control command; wherein the multimodal information includes image state factor data, environmental factor data, and waste attribute factor data; the environmental factor data includes seasonal factor data and weather factor data; the waste attribute factor data includes drainage duration factor data, location factor data, and morphological factor data; a full-volume tag output module, used to input the multimodal information into a target full-volume calorific value prediction model to obtain full-volume waste calorific value tags output by the target full-volume calorific value prediction model; and a control strategy generation module, used to generate a dynamic incineration control strategy for the waste incineration control task based on the full-volume waste calorific value tags.

[0114] Furthermore, the waste incineration control and treatment system also includes a prediction model training module, used for: acquiring sample multimodal information of sample waste incineration control tasks; wherein, the sample multimodal information includes sample image state factor data, sample environmental factor data, and sample waste attribute factor data; the sample environmental factor data includes sample seasonal factor data and sample weather factor data; the sample waste attribute factor data includes sample dewatering duration factor data, sample location factor data, and sample morphology factor data; inputting the sample multimodal information into an initial full calorific value prediction model to obtain the full calorific value prediction result output by the initial full calorific value prediction model; constructing a composite loss function using the full calorific value prediction result and the time-delay compensation intercalary value; iteratively updating the model parameters of the initial full calorific value prediction model in the direction of convergence of the composite loss function until a preset convergence condition is met to obtain the target full calorific value prediction model.

[0115] Furthermore, the composite loss function includes a calorific value error loss function, a regularization loss function, and an imbalance loss function.

[0116] Further, the calorific value error loss function is determined by the difference between the full calorific value prediction result and the calorific value between the time-delay compensation; the determination method of the calorific value between the time-delay compensation includes: determining the time-delay waste calorific value data y_delay(t) based on the sample incineration heat input data Q_in, the system heat loss data Q_out, and the waste quality data m_waste; the time-delay waste calorific value data y_delay(t) is the waste calorific value with time delay at time t; determining the time-delay variable delta_t based on the first time-delay weight delta_v and the second time-delay weight delta_bias; the first time-delay weight delta_v is determined based on the feeder time delay delta_pusher and the grate time delay delta_grate, and the second time-delay weight delta_bias is a bias value; using the time-delay variable delta_t to dynamically supplement and correct the time-delay waste calorific value data y_delay(t) to obtain the calorific value between the time-delay compensation; wherein, the calorific value between the time-delay compensation is y_adjusted(t) = y_delay(t + delta_t).

[0117] Further, the step of inputting the sample multimodal information into the initial full-calorie prediction model to obtain the full-calorie prediction result output by the initial full-calorie prediction model includes: calling the embedding layer of the initial full-calorie prediction model to perform embedding processing on the sample multimodal information to obtain the sample multimodal embedding features corresponding to the sample multimodal information; calling the self-attention layer of the initial full-calorie prediction model to perform attention weighting on the sample multimodal embedding features to obtain the attention-enhanced embedding features corresponding to the initial embedding features of each modality in the sample multimodal embedding features; concatenating each attention-enhanced embedding feature and then performing a linear transformation to obtain the target fusion feature; wherein, the target fusion feature is X_fused = concat(attention_1, attention_2, ..., attention_m), and attention_m is the attention-enhanced embedding feature of the m-th modality; calling the fully connected layer of the initial full-calorie prediction model to process the target fusion feature to obtain the full-calorie prediction result.

[0118] Figure 2 A block diagram of a waste incineration control and processing system 10 provided in an embodiment of the present invention is shown. The waste incineration control and processing system 10 in this embodiment of the present invention has data storage, transmission, and processing functions, such as... Figure 2 As shown, the waste incineration control and processing system 10 includes: a memory 11, a processor 12, a network module 13, and a waste incineration control and processing device 20 based on multimodal factors.

[0119] The memory 11, processor 12, and network module 13 are electrically connected directly or indirectly to enable data transmission or interaction. For example, these components can be electrically connected to each other via one or more communication buses or signal lines. The memory 11 stores a multimodal factor-based waste incineration control and processing device 20, which includes at least one software function module stored in the memory 11 in the form of software or firmware. The processor 12 executes various functional applications and data processing by running the software programs and modules stored in the memory 11, such as the multimodal factor-based waste incineration control and processing device 20 in this embodiment of the invention, thereby implementing the WEB-based data logic expression processing method in this embodiment of the invention.

[0120] The memory 11 may be, but is not limited to, Random Access Memory (RAM), Read Only Memory (ROM), Programmable Read-Only Memory (PROM), Erasable Programmable Read-Only Memory (EPROM), Electrically Erasable Programmable Read-Only Memory (EEPROM), etc. The memory 11 stores programs, and the processor 12 executes these programs upon receiving execution instructions.

[0121] The processor 12 may be an integrated circuit chip with data processing capabilities. The processor 12 can be a general-purpose processor, including a central processing unit (CPU), a network processor (NP), etc. It can implement or execute the methods, steps, and logic block diagrams disclosed in the embodiments of this invention. The general-purpose processor can be a microprocessor or any conventional processor.

[0122] The network module 13 is used to establish a communication connection between the waste incineration control and processing system 10 and other communication terminal equipment via the network, and to realize the transmission and reception of network signals and data. The aforementioned network signals may include wireless signals or wired signals.

[0123] Understandable. Figure 2 The structure shown is for illustrative purposes only; the waste incineration control and treatment system 10 may also include components that are more advanced than those shown. Figure 2 The more or fewer components shown, or having the same Figure 2 The different configurations shown. Figure 2 The components shown can be implemented using hardware, software, or a combination thereof.

[0124] This invention also provides a computer-readable storage medium, which includes a computer program. When the computer program runs, it controls the waste incineration control and processing system 10, where the readable storage medium is located, to execute the aforementioned waste incineration control and processing method based on the total calorific value of waste.

[0125] In summary, the embodiments of the present invention can accurately and timely determine the calorific value label of all waste, thereby obtaining accurate and forward-looking total calorific value, and thus achieving effective control of the waste incineration process. The embodiments of the present invention combine the acquisition of multimodal information, the application of a total calorific value prediction model, and the generation of dynamic incineration control strategies. This not only improves the accuracy and timeliness of total calorific value prediction, but also enables refined control of waste incineration based on the predicted total calorific value, maximizing energy efficiency while achieving stable output of thermal or electrical energy.

[0126] Furthermore, a dynamic incineration control strategy based on the total calorific value of waste corresponding to the total calorific value label enables refined control of the waste incineration process. Various parameters during incineration, such as feed rate, combustion air volume, and incineration temperature, can be flexibly adjusted according to the actual calorific value of the waste. This not only improves incineration efficiency, allowing for more complete combustion of waste and reducing the emission of incompletely burned substances, but also ensures a stable output of heat or electricity, meeting energy supply demands while improving energy utilization efficiency, reducing negative environmental impacts, and enhancing the overall benefits of waste incineration.

[0127] The above description is merely a preferred embodiment of the present invention and is not intended to limit the invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A waste incineration control and treatment method based on the total calorific value of waste, characterized in that, The method is applied to a waste incineration control and treatment system, and the method includes: In response to a target control command, multimodal information of the waste incineration control task corresponding to the target control command is acquired; wherein, the multimodal information includes image state factor data, environmental factor data, and waste attribute factor data; the environmental factor data includes seasonal factor data and weather factor data; the waste attribute factor data includes drainage duration factor data, location factor data, and morphological factor data; The multimodal information is input into the target full calorific value prediction model to obtain the full calorific value label of the garbage output by the target full calorific value prediction model; Based on the total calorific value tags of the waste, a dynamic incineration control strategy is generated for the waste incineration control task. The training steps for the target total calorific value prediction model include: Acquire multimodal information of samples for waste incineration control tasks; wherein, the multimodal information of samples includes sample image state factor data, sample environmental factor data, and sample waste attribute factor data; the sample environmental factor data includes sample seasonal factor data and sample weather factor data; the sample waste attribute factor data includes sample drainage duration factor data, sample location factor data, and sample morphology factor data; The sample multimodal information is input into the initial full calorific value prediction model to obtain the full calorific value prediction result output by the initial full calorific value prediction model. A composite loss function is constructed using the full calorific value prediction results and the calorific value between time delay compensation. The model parameters of the initial full calorific value prediction model are iteratively updated in the direction of convergence of the composite loss function until the preset convergence condition is met, thereby obtaining the target full calorific value prediction model.

2. The method according to claim 1, characterized in that, The composite loss function includes the calorific value error loss function, the regularization loss function, and the imbalance loss function.

3. The method according to claim 2, characterized in that, The calorific value error loss function is determined by the difference between the full calorific value prediction result and the calorific value between the time-delay compensation intervals; the determination method for the calorific value between the time-delay compensation intervals includes: The time-delayed calorific value data y_delay(t) is determined based on the sample incineration heat input data Q_in, system heat loss data Q_out, and waste quality data m_waste; the time-delayed calorific value data y_delay(t) is the calorific value of waste with time delay at time t; The time delay variable delta_t is determined based on the first time delay weight delta_v and the second time delay weight delta_bias; the first time delay weight delta_v is determined based on the feeder time delay delta_pusher and the grate time delay delta_grate, and the second time delay weight delta_bias is a bias value. The time-delay waste calorific value data y_delay(t) is dynamically supplemented and corrected using the time-delay variable delta_t to obtain the time-delay compensation interval calorific value; wherein, the time-delay compensation interval calorific value is y_adjusted(t)=y_delay(t+delta_t).

4. The method according to claim 1, characterized in that, The step of inputting the sample multimodal information into the initial full calorific value prediction model to obtain the full calorific value prediction result output by the initial full calorific value prediction model includes: The embedding layer of the initial full-volume calorific value prediction model is invoked to embed the multimodal information of the sample, thereby obtaining the multimodal embedding features of the sample corresponding to the multimodal information of the sample. The self-attention layer of the initial full-volume heat value prediction model is invoked to perform attention weighting on the sample multimodal embedding features, thereby obtaining the attention-enhanced embedding features corresponding to the initial embedding features of each modality in the sample multimodal embedding features; each attention-enhanced embedding feature is concatenated and then linearly transformed to obtain the target fusion feature; wherein, the target fusion feature is X_fused = concat(attention_1, attention_2, ..., attention_m), and attention_m is the attention-enhanced embedding feature of the m-th modality; The target fusion features are processed by calling the fully connected layer of the initial full calorific value prediction model to obtain the full calorific value prediction result.

5. A waste incineration control and treatment system, characterized in that, The waste incineration control and treatment system includes: The modal information acquisition module is used to acquire multimodal information of the waste incineration control task corresponding to the target control command in response to the target control command; wherein, the multimodal information includes image state factor data, environmental factor data, and waste attribute factor data; the environmental factor data includes seasonal factor data and weather factor data; the waste attribute factor data includes drainage duration factor data, location factor data, and morphological factor data; The full-volume tag output module is used to input the multimodal information into the target full-volume calorific value prediction model to obtain the full-volume garbage calorific value tags output by the target full-volume calorific value prediction model. The control strategy generation module is used to generate a dynamic incineration control strategy for the waste incineration control task based on the total calorific value tags of the waste. The waste incineration control and treatment system also includes a predictive model training module, used for: Acquire multimodal information of samples for waste incineration control tasks; wherein, the multimodal information of samples includes sample image state factor data, sample environmental factor data, and sample waste attribute factor data; the sample environmental factor data includes sample seasonal factor data and sample weather factor data; the sample waste attribute factor data includes sample drainage duration factor data, sample location factor data, and sample morphology factor data; The sample multimodal information is input into the initial full calorific value prediction model to obtain the full calorific value prediction result output by the initial full calorific value prediction model. A composite loss function is constructed using the full calorific value prediction results and the calorific value between time delay compensation. The model parameters of the initial full calorific value prediction model are iteratively updated in the direction of convergence of the composite loss function until the preset convergence condition is met, thereby obtaining the target full calorific value prediction model.

6. The waste incineration control and treatment system according to claim 5, characterized in that, The composite loss function includes the calorific value error loss function, the regularization loss function, and the imbalance loss function.

7. The waste incineration control and treatment system according to claim 6, characterized in that, The calorific value error loss function is determined by the difference between the full calorific value prediction result and the calorific value between the time-delay compensation intervals; the determination method for the calorific value between the time-delay compensation intervals includes: The time-delayed calorific value data y_delay(t) is determined based on the sample incineration heat input data Q_in, system heat loss data Q_out, and waste quality data m_waste; the time-delayed calorific value data y_delay(t) is the calorific value of waste with time delay at time t; The time delay variable delta_t is determined based on the first time delay weight delta_v and the second time delay weight delta_bias; the first time delay weight delta_v is determined based on the feeder time delay delta_pusher and the grate time delay delta_grate, and the second time delay weight delta_bias is a bias value. The time-delay waste calorific value data y_delay(t) is dynamically supplemented and corrected using the time-delay variable delta_t to obtain the time-delay compensation interval calorific value; wherein, the time-delay compensation interval calorific value is y_adjusted(t)=y_delay(t+delta_t).

8. The waste incineration control and treatment system according to claim 5, characterized in that, The step of inputting the sample multimodal information into the initial full calorific value prediction model to obtain the full calorific value prediction result output by the initial full calorific value prediction model includes: The embedding layer of the initial full-volume calorific value prediction model is invoked to embed the multimodal information of the sample, thereby obtaining the multimodal embedding features of the sample corresponding to the multimodal information of the sample. The self-attention layer of the initial full-volume heat value prediction model is invoked to perform attention weighting on the sample multimodal embedding features, thereby obtaining the attention-enhanced embedding features corresponding to the initial embedding features of each modality in the sample multimodal embedding features; each attention-enhanced embedding feature is concatenated and then linearly transformed to obtain the target fusion feature; wherein, the target fusion feature is X_fused = concat(attention_1, attention_2, ..., attention_m), and attention_m is the attention-enhanced embedding feature of the m-th modality; The target fusion features are processed by calling the fully connected layer of the initial full calorific value prediction model to obtain the full calorific value prediction result.

Images