Adaptive control system and method for fly ash washing

Abstract

The present application relates to the field of intelligent control, and provides an adaptive control system and method for fly ash washing, which uses AI-based data processing and encoding technology to group the washing liquid status data in the washing liquid storage tank, and then maps the grouped parameter data sets to the Gram angle field and performs characteristic analysis of the time series pattern, so as to adaptively adjust the water-cement ratio based on the dynamic memory attention fusion between the time series pattern characteristics of the washing liquid pH value and the time series pattern characteristics of the washing liquid conductivity. In this way, the changes in the washing process can be understood in real time, and the water-cement ratio can be adaptively adjusted according to the state of the washing liquid after the current washing, so as to ensure that the chlorine and heavy metal removal can be maintained stably when facing fly ash with large composition fluctuations. At the same time, the need for manual adjustment is reduced, and the error is reduced.

Description

Technical Field

[0001] The present application relates to the field of intelligent control, and more specifically, to an adaptive control system and method for fly ash washing. Background Art

[0002] Fly ash washing is a technology used to treat fly ash generated during the incineration of domestic waste and hazardous waste. The purpose is to remove harmful substances (such as heavy metals and soluble salts) to reduce their impact on the environment and improve the reuse value of fly ash. Patent CN118558722A discloses a waste incineration fly ash washing system and a washing method thereof, which first mixes the raw ash with the washing liquid in the secondary washing liquid storage tank, stirs and settles in the fly ash washing tank, and then discharges the supernatant into the water treatment system. The second and third washings use the washing liquid in the third and fourth washing liquid storage tanks to repeat the above process, and the supernatant is refluxed to the previous washing liquid storage tank in turn. Finally, the fourth washing is carried out through a water replenishment device, and the solid separated by washing is used as fly ash that can be recycled.

[0003] In the fly ash washing process, the water-to-ash ratio is a key operating parameter that directly affects the removal efficiency of harmful components (such as chlorine, heavy metals, etc.) in fly ash. An appropriately high water-to-ash ratio can improve the washing quality. In addition, a reasonable setting of the water-to-ash ratio helps to optimize the use of the washing liquid. If the water-to-ash ratio is too high, although it may increase the removal rate of harmful components, it will also lead to excessive consumption of washing liquid, increasing the treatment cost; on the contrary, a too low water-to-ash ratio may lead to incomplete washing. If the washing is not clean, the salt content exceeds 1%. In the subsequent disposal process, mainly when fly ash and other hazardous wastes are melted and harmlessly disposed, the disposal equipment will be corroded, and the flue gas will be easily crystallized and clogged after the flue gas is cooled. The flue gas disposal system.

[0004] The water-cement ratio in this patent is manually set based on actual operating experience and the requirements for treatment effects. However, the manual setting method in the existing patent relies on predetermined empirical values ​​or fixed ratios, which makes it slow to respond to real-time changes in fly ash characteristics or water washing liquid status. For example, when faced with different batches of fly ash with large fluctuations in composition, the static water-cement ratio cannot be flexibly adjusted, which may cause unstable chlorine and heavy metal removal efficiency and affect the final treatment effect. In addition, manual adjustment not only increases the workload of operators, but also easily introduces human errors.

[0005] Therefore, an optimized control scheme for fly ash water washing is desired. Summary of the invention

[0006] In view of the shortcomings of the prior art, the present application provides an adaptive control system and method for fly ash washing.

[0007] According to one aspect of the present application, an adaptive control method for fly ash washing is provided, and the fly ash washing process comprises: using the washing liquid in the secondary washing liquid storage tank to wash the raw fly ash sent to the primary washing tank once and filter it to obtain the primary washed fly ash and the primary washing liquid; treating the primary washing liquid with wastewater; using the washing liquid in the tertiary washing liquid storage tank to wash the primary washed fly ash sent to the secondary washing tank twice and filter it to obtain the secondary washed fly ash and the secondary washing liquid; discharging the secondary washing liquid into the secondary washing liquid storage tank; using the washing liquid in the clean water washing liquid storage tank to wash the secondary washed fly ash sent to the tertiary washing tank three times and filter it to obtain the final fly ash and the tertiary washing liquid; discharging the tertiary washing liquid into the tertiary washing liquid storage tank; wherein, the adaptive control method for fly ash washing comprises: in each process of washing the fly ash , the water-cement ratio is adjusted in real time based on the state of the washing liquid, including: collecting a time series set of washing liquid state data in a washing liquid storage tank through a sensor group, the washing liquid state data including a pH value of the washing liquid and a conductivity of the washing liquid; grouping and mapping the time series set of the washing liquid state data to obtain a Gram angular field image of the pH value of the washing liquid and a Gram angular field image of the conductivity of the washing liquid; extracting time series pattern features of the Gram angular field image of the pH value of the washing liquid and the Gram angular field image of the conductivity of the washing liquid respectively to obtain a time series pattern feature of the pH value of the washing liquid and a time series pattern feature of the conductivity of the washing liquid; performing dynamic memory mechanism attention fusion on the time series pattern feature of the pH value of the washing liquid and the time series pattern feature of the conductivity of the washing liquid to obtain a multi-dimensional time series representation of the washing liquid state, and obtaining a recommended water-cement ratio based on the multi-dimensional time series representation of the washing liquid state.

[0008] According to another aspect of the present application, an adaptive control system for fly ash washing is provided, which includes: a water washing liquid state data acquisition module, which is used to collect a time series set of water washing liquid state data in a water washing liquid storage tank through a sensor group, wherein the water washing liquid state data includes a water washing liquid pH value and a water washing liquid conductivity; a water washing liquid state data grouping and mapping module, which is used to group and map the time series set of water washing liquid state data to obtain a water washing liquid pH value mapping Gram angular field image and a water washing liquid conductivity mapping Gram angular field image; a water washing liquid pH value conductivity time series A pattern feature extraction module is used to extract time-series pattern features from the Gram angular field image mapped by the pH value of the washing liquid and the Gram angular field image mapped by the conductivity of the washing liquid, respectively, to obtain time-series pattern features of the pH value of the washing liquid and time-series pattern features of the conductivity of the washing liquid; a water-cement ratio recommendation result generation module is used to perform dynamic memory mechanism attention fusion on the time-series pattern features of the pH value of the washing liquid and the time-series pattern features of the conductivity of the washing liquid to obtain a multi-dimensional time-series representation of the state of the washing liquid, and obtain a recommended water-cement ratio based on the multi-dimensional time-series representation of the state of the washing liquid.

[0009] The present application has significant technical effects due to the adoption of the above technical solutions: the adaptive control system and method for fly ash washing provided by the present application adopts AI-based data processing and coding technology to group the washing liquid status data in the washing liquid storage tank, and then maps each grouped parameter data set to the Gram angle field and performs characteristic analysis of the time series pattern, so as to adaptively adjust the water-cement ratio based on the dynamic memory attention fusion between the time series pattern characteristics of the washing liquid pH value and the time series pattern characteristics of the washing liquid conductivity. In this way, the changes in the washing process can be understood in real time, and the water-cement ratio can be adaptively adjusted according to the state of the washing liquid after the current washing, so as to ensure that stable chlorine and heavy metal removal can be maintained when facing fly ash with large composition fluctuations. At the same time, the need for manual adjustment is reduced, and errors are reduced. BRIEF DESCRIPTION OF THE DRAWINGS

[0010] By describing the embodiments of the present application in more detail in conjunction with the accompanying drawings, the above and other purposes, features and advantages of the present application will become more apparent. The accompanying drawings are used to provide a further understanding of the embodiments of the present application and constitute a part of the specification. Together with the embodiments of the present application, they are used to explain the present application and do not constitute a limitation of the present application. In the accompanying drawings, the same reference numerals generally represent the same components or steps.

[0011] Figure 1 The figure is a flow chart of an adaptive control method for fly ash washing according to an embodiment of the present application.

[0012] Figure 2 Schematic diagram of data flow of an adaptive control method for fly ash washing according to an embodiment of the present application.

[0013] Figure 3 Flow chart of step S140 in the adaptive control method for fly ash washing according to an embodiment of the present application.

[0014] Figure 4 Flow chart of step S141 in the adaptive control method for fly ash washing according to an embodiment of the present application.

[0015] Figure 5 The system block diagram of an adaptive control system for fly ash washing according to an embodiment of the present application. DETAILED DESCRIPTION

[0016] Below, the exemplary embodiments according to the present application will be described in detail with reference to the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all the embodiments of the present application, and it should be understood that the present application is not limited to the exemplary embodiments described here.

[0017] Fly ash washing technology is used to treat fly ash generated during the incineration of domestic waste and hazardous waste, with the purpose of removing harmful substances to reduce environmental impact and improve the reuse value of fly ash. Patent CN118558722A introduces a waste incineration fly ash washing system and a washing method thereof, which purifies fly ash by multiple washings and reflux of washing liquid. In the fly ash washing process, the water-ash ratio is a key operating parameter, which affects the removal efficiency of harmful components and the amount of washing liquid used. An appropriately high water-ash ratio can improve the washing quality, but too high or too low water-ash ratios have their disadvantages. The setting of the water-ash ratio in the existing patents depends on empirical values ​​or fixed ratios, which makes the water-ash ratio unable to flexibly adapt to changes in fly ash characteristics or the state of washing liquid, resulting in unstable removal efficiency and affected treatment effects. In addition, manually adjusting the water-ash ratio not only increases the workload, but may also introduce errors. Based on this, the present application proposes an optimized adaptive control method for fly ash washing.

[0018] It should be understood that in the adaptive control method for fly ash washing of the present application, the fly ash washing process includes: using the washing liquid in the secondary washing liquid storage tank to wash the original fly ash sent to the primary washing tank once and filter it to obtain primary washed fly ash and primary washing liquid; treating the primary washing liquid as wastewater; using the washing liquid in the tertiary washing liquid storage tank to wash the primary washed fly ash sent to the secondary washing tank twice and filter it to obtain secondary washed fly ash and secondary washing liquid; discharging the secondary washing liquid into the secondary washing liquid storage tank; using the washing liquid in the clean water washing liquid storage tank to wash the secondary washed fly ash sent to the tertiary washing tank thrice and filter it to obtain final fly ash and tertiary washing liquid; discharging the tertiary washing liquid into the tertiary washing liquid storage tank.

[0019] Specifically, first, the washing liquid in the secondary washing liquid storage tank is used to wash the original fly ash sent to the primary washing tank. The original fly ash has a complex composition and contains high levels of salt (such as chloride), heavy metals, dioxins and other harmful substances. After one washing, the chlorine content in the fly ash is significantly reduced, and some heavy metals are also transferred to the wastewater, reducing the chlorine content in the fly ash, laying the foundation for the subsequent washing steps. Then, the primary washing liquid needs to be treated with wastewater to avoid direct discharge of the washing liquid containing harmful substances, which will cause environmental pollution. Then, the washing liquid in the tertiary washing liquid storage tank is used to wash the primary washing fly ash sent to the secondary washing tank for a secondary washing, which can further reduce the chlorine content in the fly ash and further reduce the heavy metal content, which is conducive to improving the quality of the fly ash and preparing for the high-temperature melting treatment of the subsequent process. The washing liquid obtained by the secondary washing needs to be discharged into the secondary washing liquid storage tank to recycle the washing liquid and reduce the use of fresh water. By recycling the washing liquid, water consumption can be reduced, and water resources can be saved and recycled. Then, the washing liquid in the clean water washing liquid storage tank is used to wash the secondary washed fly ash sent to the tertiary washing tank for a third time to reduce the chlorine content in the fly ash to a minimum and ensure the quality of the fly ash. After the third washing, the chlorine content in the fly ash is reduced to about 1%, which greatly reduces the generation of acidic gas in the subsequent process, contributes to the long-term stable operation of the high-temperature melting equipment, and reduces the pollution to the environment. At the same time, the third washing liquid is discharged into the third washing liquid storage tank. In this way, by recycling and recycling the third washing liquid, the consumption of water resources can be further reduced, the discharge of wastewater can be reduced, and a win-win situation of environmental protection and economic benefits can be achieved. In general, the entire fly ash washing process effectively reduces the chlorine content in the fly ash and the leaching of heavy metals through a multi-stage countercurrent method, while also realizing the recycling of water resources.

[0020] In particular, in each process of washing fly ash, the water-cement ratio can be adjusted in real time based on the state of the washing liquid. Accordingly, in each process of washing fly ash, the water-cement ratio is adjusted in real time based on the state of the washing liquid. The technical concept of the present application is to collect the time series set of washing liquid state data (washing liquid pH value and washing liquid conductivity) in the washing liquid storage tank through the sensor group, and use AI-based data processing and encoding technology to group the washing liquid state data, and then map the grouped parameter data sets to the Gram angle field and perform characteristic analysis of the time series pattern, so as to adaptively adjust the water-cement ratio based on the dynamic memory attention fusion between the washing liquid pH value time series pattern characteristics and the washing liquid conductivity time series pattern characteristics. In this way, the changes in the washing process can be understood in real time, and the water-cement ratio can be adaptively adjusted according to the state of the washing liquid after the current washing, so as to ensure that the removal of chlorine and heavy metals can be maintained in a stable state when facing fly ash with large composition fluctuations. At the same time, the need for manual adjustment is reduced, and the error is reduced.

[0021] Figure 1 The figure is a flow chart of an adaptive control method for fly ash washing according to an embodiment of the present application. Figure 2 FIG. 1 is a data flow diagram of an adaptive control method for fly ash washing according to an embodiment of the present application. Figure 1 and Figure 2 As shown, the adaptive control method for fly ash washing according to the embodiment of the present application includes: S110, collecting a time series set of water washing liquid status data in a water washing liquid storage tank through a sensor group, the water washing liquid status data including the water washing liquid pH value and the water washing liquid conductivity; S120, grouping and mapping the time series set of the water washing liquid status data to obtain a water washing liquid pH value mapping Gram angular field image and a water washing liquid conductivity mapping Gram angular field image; S130, respectively performing time series pattern feature extraction on the water washing liquid pH value mapping Gram angular field image and the water washing liquid conductivity mapping Gram angular field image to obtain a water washing liquid pH value time series pattern feature and a water washing liquid conductivity time series pattern feature; S140, performing dynamic memory mechanism attention fusion on the water washing liquid pH value time series pattern feature and the water washing liquid conductivity time series pattern feature to obtain a multi-dimensional time series representation of the water washing liquid state, and based on the multi-dimensional time series representation of the water washing liquid state, obtain a recommended water-ash ratio.

[0022] In step S110, a time series set of water washing liquid state data in the water washing liquid storage tank is collected by the sensor group, and the water washing liquid state data includes the water washing liquid pH value and the water washing liquid conductivity. It should be understood that the time series set of water washing liquid state data in the water washing liquid storage tank specifically includes the pH value and conductivity information of the water washing liquid at different time points. Specifically, the pH value of the water washing liquid reflects its acidity and alkalinity. In the fly ash washing process, different pH values ​​may imply changes in the content of acidic or alkaline substances dissolved in the water. For example, when the acidic or alkaline components in the fly ash (such as certain heavy metal oxides or salts) are dissolved in the water washing liquid, the pH value of the water washing liquid will change. The conductivity of the water washing liquid is related to the concentration of ions dissolved therein. The dissolution of soluble salts in the fly ash (such as sodium chloride, potassium chloride, etc.) into the water washing liquid will increase the ion concentration, thereby causing the conductivity to increase. The time series change of conductivity can reflect the dissolution rate and degree of soluble salts in the fly ash. For example, at the beginning of water washing, if the conductivity rises rapidly, it means that a large amount of soluble salts in the fly ash are rapidly dissolving into the water washing liquid. In general, when the time series changes in the pH value and conductivity of the water washing liquid show that certain components (such as heavy metals and soluble salts) are rapidly dissolving, the water-cement ratio may need to be appropriately increased. Because a higher water-cement ratio can provide more solvent (water washing liquid) to dissolve and dilute these harmful components, ensuring that they can be more fully removed from the fly ash. In other words, by real-time monitoring of the water washing liquid status data, the water-cement ratio can be dynamically adjusted according to the actual situation during the water washing process to achieve the best water washing effect, that is, effectively remove harmful substances in the fly ash while avoiding waste of water washing liquid.

[0023] The time series collection of water washing liquid status data including water washing liquid pH value and water washing liquid conductivity in the water washing liquid storage tank through the sensor group is a complex process involving multiple links. The following is a detailed description of its implementation process: First, the selection and installation of sensors. For pH sensors, the selection should be based on the characteristics of the water washing liquid. If the water washing liquid is corrosive, glass electrodes or special polymer electrode sensors are more suitable because they can resist corrosion and accurately measure the hydrogen ion concentration in the solution to determine the pH value. The installation location is crucial. It is necessary to ensure that the sensor probe can fully contact the water washing liquid. It is feasible to install it on the side wall or bottom of the water washing liquid storage tank, but avoid interference from stirring devices or other equipment to ensure the accuracy of the measurement.

[0024] The selection of conductivity sensors also needs to consider the composition of the washing liquid. For washing liquids containing a certain amount of salt, electrode conductivity sensors are a common choice, which can accurately measure the conductivity of ions in them. The installation position must also be able to fully contact the washing liquid, and attention should be paid to reducing the interference of temperature changes on the measurement results. It can be installed close to the pH sensor, but mutual interference should be avoided.

[0025] Next is the setup of the data acquisition system. The sensor needs to be connected to the data acquisition device through a stable cable or signal line to ensure the accuracy and stability of signal transmission and prevent signal loss or interference. The acquisition frequency should be determined according to the characteristics of the water washing process and the requirements for data accuracy. If the water washing process is slow and the changes are not large, a lower acquisition frequency, such as collecting data every few minutes, is sufficient; if the water washing process is fast and the composition may change greatly, a higher acquisition frequency is required, such as collecting data every second or every few seconds. At the same time, when collecting data, the timestamp of each data point should be recorded, which is crucial for constructing an accurate time series set to facilitate the subsequent accurate analysis of the changes in the state of the water washing liquid over time.

[0026] Data transmission and storage are also key links. The collected data must be transmitted to the central control system or data processing unit. There are two types of transmission methods: wired and wireless. Wired transmission can use Ethernet cables or industrial field buses such as Profibus, Modbus, etc.; wireless transmission can use technologies such as ZigBee, Wi-Fi or Bluetooth. The choice of which method should be based on the actual application scenario and device layout. For data storage, you can use relational databases such as MySQL, Oracle, etc., or you can use non-relational databases such as MongoDB, InfluxDB, etc. For time series data, specialized time series databases such as InfluxDB may have more advantages, which can easily query, analyze and visualize data.

[0027] Finally, there is calibration and maintenance. Sensors need to be calibrated regularly. pH sensors can be calibrated with standard buffer solutions, and conductivity sensors can be calibrated with standard solutions of known conductivity. The calibration cycle should be determined based on the frequency of use and environmental conditions of the sensor. It is generally recommended to calibrate once a week or month. At the same time, the installation of the sensor should be checked regularly to ensure that it is not loose, damaged or contaminated. The working status of the data acquisition equipment and transmission lines should also be checked to promptly discover and solve possible problems, such as line failures, equipment crashes, etc., to ensure the normal operation of the data acquisition system.

[0028] Only through the above series of rigorous operating procedures can we ensure that the time series set of water washing liquid status data in the water washing liquid storage tank is accurately collected through the sensor group, which can provide a reliable data basis for subsequent data analysis and processing, and thus realize the effective monitoring and optimization of the water washing process.

[0029] In step S120, the time series set of the water washing liquid state data is grouped and mapped to obtain a water washing liquid pH value mapping Gram angular field image and a water washing liquid conductivity mapping Gram angular field image. Specifically, in an embodiment of the present application, the step S120 includes: grouping the time series set of the water washing liquid state data based on the water washing liquid state parameter sample dimension to obtain a water washing liquid pH value time series set and a water washing liquid conductivity time series set; performing a Gram angular field transformation on the water washing liquid pH value time series set and the water washing liquid conductivity time series set respectively to obtain the water washing liquid pH value mapping Gram angular field image and the water washing liquid conductivity mapping Gram angular field image.

[0030] It should be understood that each parameter in the water washing liquid state reflects different physical characteristic information. The pH value focuses on reflecting the changes in pH during the water washing process, and the conductivity is used to track the changes in the concentration of electrolytes dissolved in the water washing liquid. Therefore, in order to be able to understand and analyze the characteristics of each water washing liquid parameter more finely, in the technical solution of the present application, the time series set of the water washing liquid state data is grouped based on the water washing liquid state parameter sample dimension to obtain the water washing liquid pH value time series set and the water washing liquid conductivity time series set. In this way, the separated time series sets can be pattern recognized separately to better understand the trend of each parameter evolving over time. For example, pH value and conductivity may each have different periodic fluctuations or other significant patterns, which are very important for predicting future water washing conditions.

[0031] Then, considering that the water washing liquid pH value time series set and the water washing liquid conductivity time series set both contain patterns, trends and some abnormal information in the time dimension, this information is crucial for the adjustment of the water-cement ratio. Therefore, in order to be able to more intuitively observe the time series change characteristic information in the water washing liquid state parameters, in the technical solution of the application of the present application, the water washing liquid pH value time series set and the water washing liquid conductivity time series set are respectively subjected to Gram angular field transformation to obtain the water washing liquid pH value mapping Gram angular field image and the water washing liquid conductivity mapping Gram angular field image. It is worth mentioning that the Gram angular field transformation can convert one-dimensional time series data into a two-dimensional image form, which helps to capture the long-term dependencies and periodic patterns in the time series. In particular, the Gram angular field image retains the time dependency in the original time series, and can better capture the time series dynamic characteristics of the pH value and conductivity data of the water washing liquid state, and observe the changing trends in different time periods and the correlation between them.

[0032] The specific implementation process of performing Gram angular field transformation on the water washing liquid pH value time series set and the water washing liquid conductivity time series set to obtain the water washing liquid pH value mapping Gram angular field image and the water washing liquid conductivity mapping Gram angular field image is as follows: first, the data preparation stage. At this stage, for the water washing liquid pH value and conductivity time series data obtained by grouping, detailed data cleaning and preprocessing are required. Data cleaning is a crucial first step, and its purpose is to remove those abnormal values ​​caused by sensor failure or external interference. This requires setting a reasonable data range, through which those obviously unreasonable data points are identified and excluded to ensure the data quality of subsequent processing. At the same time, since the Gram angular field requires a relatively fixed input data range, it is necessary to normalize the water washing liquid pH value and conductivity time series data. Usually, the values ​​are mapped to the interval [-1,1] or [0,1]. Doing so not only helps to maintain the consistency of the values, but also lays the foundation for the subsequent angle encoding steps.

[0033] After normalization, the next step is angle encoding. The goal here is to convert the normalized values ​​into angle values ​​in the polar coordinate system. Specifically, for each time point t, the corresponding normalized value x(t) is converted into an angle θ(t) by the formula θ(t)=arccos(x(t)) or θ(t)=arcsin(x(t)). The choice of sine or cosine depends on whether you want to maintain the positive or negative characteristics in the time series. The key to this step is that it provides a way to represent the positional relationship between each time point in the time series, thereby preparing for the further construction of the Gram angle field.

[0034] When all time points have been successfully converted to angle values, we can start to construct the Gram angular field matrix. This matrix is ​​an N×N matrix, where N represents the length of the time series. Each element M(i,j) in the matrix is ​​calculated by the angle between two time points i and j, that is, M(i,j)=cos(θ(i)-θ(j)) or M(i,j)=sin(θ(i)-θ(j)). This construction method allows the dynamic connection between different time points to be captured, forming a complete "field" that reflects the changing law of the water washing liquid state during the entire time period.

[0035] Once the Gram angular field matrix is ​​obtained, the last step is to convert it into an image form. This can be achieved through a given color mapping scheme, such as using a heat map color mode so that different values ​​correspond to different color intensities. In this way, the originally complex and difficult to understand one-dimensional time series is converted into an intuitive and easy-to-understand two-dimensional image. For the pH value time series set of the water washing liquid, it will become an image reflecting the pH value change pattern; and for the conductivity time series set of the water washing liquid, another image showing the conductivity change characteristics will be generated. These two images together constitute the state representation of the water washing liquid at different time points, which can provide a reliable data source for subsequent feature extraction.

[0036] It is worth noting that in practical applications, the above steps must take into account the real-time and accuracy of the system. Because the water-cement ratio is adjusted in real time during the fly ash washing process, the speed from data acquisition to image generation must be fast enough to support instant decision-making. At the same time, the sensor measurement accuracy and data transmission must be ensured to avoid data quality issues affecting the final image representation and subsequent analysis. In addition, considering the effective use of computing resources, the calculation process should be simplified as much as possible and implemented with efficient algorithms, especially in embedded systems or edge computing environments.

[0037] In summary, through the above complete process, the Gramm angular field transformation of the water washing liquid pH value time series set and the water washing liquid conductivity time series set can be accurately performed respectively to obtain high-quality water washing liquid pH value mapping Gramm angular field image and water washing liquid conductivity mapping Gramm angular field image, which provide an effective data form for subsequent analysis and processing.

[0038] In step S130, the Graham angular field image of the pH value of the washing liquid and the Graham angular field image of the conductivity of the washing liquid are respectively subjected to time series pattern feature extraction to obtain the time series pattern feature of the pH value of the washing liquid and the time series pattern feature of the conductivity of the washing liquid. Specifically, in an embodiment of the present application, the step S130 includes: respectively passing the Graham angular field image of the pH value of the washing liquid and the Graham angular field image of the conductivity of the washing liquid through a water washing liquid state time series pattern feature extractor based on a micro convolutional neural network to obtain a water washing liquid pH value time series pattern feature vector as the water washing liquid pH value time series pattern feature and a water washing liquid conductivity time series pattern feature vector as the water washing liquid conductivity time series pattern feature. It should be understood that in order to further capture and refine the temporal feature information and inherent temporal patterns in the Gram angular field image of each parameter, in the technical solution of the present application, the Gram angular field image of the pH value of the washing liquid and the Gram angular field image of the conductivity of the washing liquid are respectively subjected to a water washing liquid state temporal pattern feature extractor based on a micro-convolutional neural network to utilize the smaller convolution kernel of the micro-convolutional neural network to extract local temporal features and time dependencies, thereby obtaining a water washing liquid pH value temporal pattern feature vector and a water washing liquid conductivity temporal pattern feature vector. In particular, the micro-convolutional neural network significantly reduces the number of parameters by simplifying the network structure and introducing global average pooling instead of the traditional fully connected layer, thereby extracting a representative temporal feature pattern.

[0039] The Gram angular field image mapped to the pH value of the washing liquid and the Gram angular field image mapped to the conductivity of the washing liquid are respectively passed through the washing liquid state time series pattern feature extractor based on the micro convolutional neural network to obtain the washing liquid pH value time series pattern feature vector and the washing liquid conductivity time series pattern feature vector. This process can convert the Gram angular field image into a representative feature vector so as to better capture the long-term dependencies and periodic patterns in the original washing liquid pH value and washing liquid conductivity time series data. The following is a detailed description of its implementation process: First, the construction of a micro convolutional neural network. The network structure design is crucial, which determines the network's ability to extract image features. When determining the number of network layers, considering the characteristics of the micro convolutional neural network, the number of layers should not be too many, and it is generally more appropriate to design 3-5 convolutional layers. This can not only reduce the amount of calculation, but also prevent the occurrence of overfitting. The number of neurons in each convolutional layer needs to be determined based on the size of the input image and the complexity of the features to be extracted. If the input Gram angular field image size is small, then a relatively small number of neurons can be set, such as 32-64; if the image size is large or more complex features need to be extracted, the number of neurons should be appropriately increased.

[0040] The choice of convolution kernel size is also one of the key factors in building the network. For processing Gram angular field images, smaller convolution kernels are often more conducive to capturing local features. Common convolution kernel sizes are 3×3 or 5×5. At the same time, the convolution step size must be determined, which is generally set to 1 or 2, which can control the speed and coverage of the convolution operation on the image.

[0041] The choice of activation function also has an important impact on network performance. In micro convolutional neural networks, commonly used activation functions are ReLU and Leaky ReLU. The ReLU function outputs the input value when the input is greater than 0, and outputs 0 when the input is less than 0. It has the advantages of effectively solving the gradient vanishing problem and speeding up network training. Leaky ReLU outputs a smaller slope value multiplied by the input value when the input is less than 0. In some cases, such as when there are many negative numbers in the input data, it may be more advantageous than ReLU.

[0042] Next is the training phase of the feature extractor. This process first requires data preparation. A large number of Grammar angle field images of the pH value mapping of the washing liquid and the Grammar angle field images of the conductivity mapping of the washing liquid should be collected as training data. These images must cover different states and changes of the washing liquid. Only in this way can we ensure that the network can learn various feature patterns. At the same time, these images must be annotated. The annotation information can be the corresponding actual state characteristics of the washing liquid, such as the content range of a certain component. In this way, the network can learn based on the annotated information.

[0043] After preparing the training data, it needs to be divided into training set, validation set and test set. It is usually more appropriate to divide it in a ratio of 7:2:1 or 8:1:1. The training set is used for network training, the validation set is used to adjust the network parameters and evaluate the network performance during the training process, and the test set is used to evaluate the final performance of the network after the network training is completed.

[0044] During the training process, choosing a suitable optimization algorithm is the key. Common optimization algorithms include stochastic gradient descent (SGD) and its variants, such as Adagrad, Adadelta, Adam, etc. These optimization algorithms minimize the loss function by adjusting the weights of the network. The definition of the loss function is based on the annotation information and the output of the network. For example, the mean square error (MSE) can be used as the loss function, which calculates the average of the squared difference between the network output and the annotation information.

[0045] During the training process, the setting of the learning rate should not be ignored. The learning rate determines the step size of the network weight adjustment. If the learning rate is too large, the network may not converge; if the learning rate is too small, the network training speed will be very slow. Therefore, it is necessary to find a suitable value by experimenting with different learning rate values. Generally, the initial learning rate can be set between 0.001-0.01. At the same time, the appropriate number of training rounds should be set according to the complexity of the network and the size of the training data. Usually, the number of training rounds can be set between 10-100 rounds.

[0046] Finally, it is the feature extraction stage. The Gram angular field image of the pH value of the washing liquid and the Gram angular field image of the conductivity of the washing liquid are respectively input into the water washing liquid state time series pattern feature extractor of the trained micro convolutional neural network. When inputting, make sure that the format and size of the image meet the requirements of the network. If the image size does not meet the requirements, appropriate scaling or cropping operations can be performed. The network processes the input image through a series of operations such as convolutional layers and activation functions, and finally outputs the water washing liquid pH value time series pattern feature vector and the water washing liquid conductivity time series pattern feature vector.

[0047] In step S140, the time series pattern characteristics of the pH value of the washing liquid and the time series pattern characteristics of the conductivity of the washing liquid are fused by a dynamic memory mechanism to obtain a multi-dimensional time series representation of the state of the washing liquid, and based on the multi-dimensional time series representation of the state of the washing liquid, a recommended water-cement ratio is obtained. Specifically, Figure 3 FIG. 1 is a flow chart of step S140 in the adaptive control method for fly ash washing according to an embodiment of the present application. Figure 3 As shown, the step S140 includes: S141, performing dynamic memory mechanism attention fusion on the time series pattern feature vector of the pH value of the washing liquid and the time series pattern feature vector of the conductivity of the washing liquid to obtain a fine-grained interaction representation vector between the multi-dimensional time series features of the washing liquid state as the multi-dimensional time series representation of the washing liquid state; S142, obtaining the recommended water-cement ratio value based on the fine-grained interaction representation vector between the multi-dimensional time series features of the washing liquid state.

[0048] In step S141, the water washing liquid pH value time series pattern feature vector and the water washing liquid conductivity time series pattern feature vector are subjected to dynamic memory mechanism attention fusion to obtain a fine-grained interaction representation vector between the water washing liquid state multi-dimensional time series features as the water washing liquid state multi-dimensional time series representation. Specifically, Figure 4 FIG. 1 is a flow chart of step S141 in the adaptive control method for fly ash washing according to an embodiment of the present application. Figure 4As shown, the step S141 includes: S1411, performing collaborative implicit feature extraction on the water washing liquid pH value time series pattern feature vector and the water washing liquid conductivity time series pattern feature vector to obtain an implicit collaborative coding vector between the water washing liquid pH value-conductivity time series features; S1412, based on the implicit collaborative coding vector between the water washing liquid pH value-conductivity time series features, performing fine-grained interaction between water washing liquid state features on the water washing liquid pH value time series pattern feature vector and the water washing liquid conductivity time series pattern feature vector to obtain a fine-grained interaction representation vector between the water washing liquid state multi-dimensional time series features.

[0049] It should be understood that there is a complex nonlinear interactive relationship between the pH value of the washing liquid and the conductivity of the washing liquid, that is, the time series dynamic pattern characteristics of the pH value of the washing liquid and the conductivity of the washing liquid influence each other on the time axis and show the characteristics of synergistic implicit interaction. In order to accurately capture these nonlinear time series associations and their subtle changes, the scheme of the present application introduces a dynamic memory mechanism attention fusion method to process the time series pattern feature vector of the pH value of the washing liquid and the time series pattern feature vector of the conductivity of the washing liquid, and obtain a fine-grained interaction representation vector between the multi-dimensional time series features of the washing liquid state. Specifically, the dynamic memory mechanism can identify and memorize the long-term trends and short-term fluctuations in the state parameters of the washing liquid, allowing the model to "remember" the interaction pattern between historical features and flexibly adjust the memory content according to the current input. At the same time, the attention mechanism enables the model to focus on the most relevant time series features, eliminate the interference of irrelevant information, and achieve a detailed understanding of the dynamic changes in the time series of the pH value and the conductivity of the washing liquid.

[0050] Specifically, firstly, the PH value time series pattern feature vector of the water washing liquid and the conductivity time series pattern feature vector of the water washing liquid are subjected to collaborative implicit feature extraction to obtain an implicit collaborative coding vector between the PH value and conductivity time series features of the water washing liquid. The above process can be expressed as: ;in, and represent the time series pattern feature vector of the pH value of the washing liquid and the time series pattern feature vector of the conductivity of the washing liquid respectively, , and They are point-by-point addition, point-by-point subtraction, and point-by-point multiplication. For cascade processing, is the point convolutional encoding operation, is the Leaky ReLU activation function, It is the implicit collaborative coding vector between the pH value and conductivity time series features of the washing liquid.

[0051] It should be understood that by performing collaborative implicit feature extraction on the water washing liquid pH value time series pattern feature vector and the water washing liquid conductivity time series pattern feature vector, it is possible to capture the deep-level correlation between the two feature vectors, which is not limited to a simple linear combination of features, but involves a complex nonlinear feature interaction pattern. Specifically, through the convolution operation, the local correlation pattern between the water washing liquid pH value time series pattern feature vector and the water washing liquid conductivity time series pattern feature vector can be identified; and by using a specific The activation function can learn nonlinear feature representation. Through this data processing method, the complex correlation between the original water washing liquid pH value time series data and the water washing liquid conductivity time series data can be accurately captured.

[0052] Specifically, in an embodiment of the present application, the step S1412 includes: writing the implicit collaborative coding vector between the water washing liquid pH value-conductivity time series features into a dynamic memory unit to obtain a water washing liquid pH value-conductivity time series dynamic key vector; using the water washing liquid pH value-conductivity time series dynamic key vector as a guide, dynamically modulating the water washing liquid pH value time series pattern feature vector and the water washing liquid conductivity time series pattern feature vector to obtain an optimized water washing liquid pH value time series pattern feature vector and an optimized water washing liquid conductivity time series pattern feature vector; weightedly fusing the optimized water washing liquid pH value time series pattern feature vector and the optimized water washing liquid conductivity time series pattern feature vector to obtain a fine-grained interaction representation vector between the multi-dimensional time series features of the water washing liquid state.

[0053] Next, the implicit collaborative coding vector between the pH value and conductivity time series features of the washing liquid is written into the dynamic memory unit to obtain the pH value and conductivity time series dynamic key vector of the washing liquid. The above process can be expressed as: ;in, is the implicit collaborative coding vector between the pH value and conductivity time series features of the water washing liquid, is a 1×1 convolution operation, It is the time series dynamic bond vector of pH value and conductivity of water washing liquid.

[0054] It should be understood that the dynamic memory unit is a special type of memory structure, and its working principle is similar to the memory unit in the recursive neural network, but it has greater flexibility and adaptability. Here, by writing the implicit collaborative coding vector between the pH value and conductivity time series characteristics of the water washing liquid into the dynamic memory unit, the dynamic adjustment capability of the dynamic memory unit can be used to make the storage and subsequent use of the pH value and conductivity time series characteristic information of the water washing liquid flexible according to the continuously newly input data (such as the pH value and conductivity data newly collected later), so as to better fit the changing state of the water washing liquid in actual production, and always keep the effective recording and processing of the latest and most relevant information for the current scene.

[0055] More specifically, in an embodiment of the present application, the water washing liquid pH value-conductivity time series dynamic key vector is used as a guide, and the water washing liquid pH value time series pattern feature vector and the water washing liquid conductivity time series pattern feature vector are respectively subjected to dynamic attention modulation to obtain the optimized water washing liquid pH value time series pattern feature vector and the optimized water washing liquid conductivity time series pattern feature vector, including: extracting the water washing liquid pH value-conductivity time series dynamic key vector from the dynamic memory unit, and performing attention modulation of the first converter structure on the water washing liquid pH value time series pattern feature vector and the water washing liquid pH value-conductivity time series dynamic key vector to obtain the optimized water washing liquid pH value time series pattern feature vector. The process can be expressed as: ;in, represents the time series pattern feature vector of the pH value of the water washing liquid, is the pH value query embedding matrix of the water washing liquid, is the first water wash pH value bias vector, is matrix multiplication, is the pH value query vector of the washing liquid, The pH value of the water wash is embedded in the matrix. is the second water wash pH value bias vector, is the pH value vector of the washing liquid, yes The transposed vector of yes Length, yes function, It is the characteristic vector of the time series pattern of pH value of water washing liquid after optimization.

[0056] The water washing liquid pH value-conductivity time series dynamic key vector is extracted from the dynamic memory unit, and the water washing liquid conductivity time series pattern feature vector and the water washing liquid pH value-conductivity time series dynamic key vector are subjected to attention modulation of the second converter structure to obtain the optimized water washing liquid conductivity time series pattern feature vector. This process can be expressed as: ;in, represents the characteristic vector of the time series pattern of the conductivity of the water washing liquid, is the conductivity query embedding matrix of the water wash solution, is the first water wash conductivity bias vector, is matrix multiplication, is the conductivity query vector of the water wash solution, is the conductivity value of the water wash embedded in the matrix, is the conductivity bias vector of the second wash solution, is the conductivity value vector of the water wash solution, yes The transposed vector of yes Length, yes function, is the characteristic vector of the time series pattern of water washing liquid conductivity after optimization.

[0057] It should be understood that the water washing liquid pH value-conductivity time series dynamic key vector is extracted from the dynamic memory unit, and the water washing liquid pH value time series pattern feature vector and the water washing liquid pH value-conductivity time series dynamic key vector are subjected to attention modulation of the first converter structure to obtain the optimized water washing liquid pH value time series pattern feature vector, and at the same time, the water washing liquid pH value-conductivity time series dynamic key vector is extracted from the dynamic memory unit, and the water washing liquid conductivity time series pattern feature vector and the water washing liquid pH value-conductivity time series dynamic key vector are subjected to attention modulation of the second converter structure to obtain the optimized water washing liquid conductivity time series pattern feature vector. Here, the core of the first converter structure and the second converter structure are both self-attention mechanisms, which allow the model to consider the influence of all other elements when processing each input element, not just adjacent elements. This global perspective helps the model capture long-distance dependencies and better understand complex data. In the technical solution of the present application, the attention modulation of the first converter structure and the attention modulation of the second converter structure can perform query attention-based feature modulation on the water washing liquid pH value time series pattern feature vector and the water washing liquid conductivity time series pattern feature vector based on the water washing liquid pH value-conductivity time series dynamic key vector, which is equivalent to using the water washing liquid pH value-conductivity time series dynamic key vector as a common attention space to enhance the expression of key information in the water washing liquid pH value time series pattern feature vector and the water washing liquid conductivity time series pattern feature vector. This not only reduces the influence of noise and irrelevant information in the original vector, but also enables the optimized water washing liquid pH value time series pattern feature vector and the optimized water washing liquid conductivity time series pattern feature vector to align the semantic space, thereby facilitating the interaction and fusion between subsequent features.

[0058] Finally, the optimized water washing liquid pH value time series pattern feature vector and the optimized water washing liquid conductivity time series pattern feature vector are weighted fused to obtain the fine-grained interaction representation vector between the multi-dimensional time series features of the water washing liquid state. The above process can be expressed as: ;in, is the time series pattern feature vector of pH value of water washing liquid after optimization, is the characteristic vector of the conductivity time series pattern of the water washing liquid after optimization, is the weighting coefficient, It is a fine-grained interaction representation vector between the multi-dimensional time series features of the water washing liquid state.

[0059] It should be understood that by weighted fusion of the optimized water washing liquid pH value time series pattern feature vector and the optimized water washing liquid conductivity time series pattern feature vector, these two optimized feature vectors can be further integrated through linear transformation to generate a fine-grained interaction representation vector between the final multi-dimensional time series features of the water washing liquid state.

[0060] In step S142, based on the fine-grained interactive characterization vector between the multi-dimensional time series features of the water washing liquid state, the recommended water-cement ratio is obtained. Specifically, in the embodiment of the present application, the step S142 includes: passing the fine-grained interactive characterization vector between the multi-dimensional time series features of the water washing liquid state through a water-cement ratio adjuster based on a decoder to obtain the recommended water-cement ratio. It should be understood that the fine-grained interactive characterization vector between the multi-dimensional time series features of the water washing liquid state contains rich and complex information, which is obtained by a series of operations such as grouping, mapping, feature extraction and fusion of data such as the pH value and conductivity of the water washing liquid. This information reflects the state changes of the water washing liquid at different times and the relationship between various states. In order to convert this complex, high-dimensional characterization vector into a specific water-cement ratio value that has practical guiding significance for the fly ash washing process, it is necessary to input the fine-grained interactive characterization vector between the multi-dimensional time series features of the water washing liquid state into the water-cement ratio adjuster based on a decoder. The decoder specifically refers to a neural network model, which is trained by a large amount of water washing liquid state data and the corresponding optimal water-cement ratio data, and can learn the mapping relationship from the input characterization vector to the appropriate water-cement ratio. When receiving a new characterization vector, it can output the recommended water-cement ratio according to the learned relationship. The recommended water-cement ratio can guide the actual operation in the fly ash washing process. By using this recommended value, the water-cement ratio can be adjusted according to the real-time state of the water washing liquid, thereby optimizing the washing process. In particular, in an embodiment of the present application, a possible implementation method of passing the fine-grained interactive characterization vector between the multi-dimensional time series features of the water washing liquid state through a water-cement ratio adjuster based on a decoder to obtain the recommended water-cement ratio value can be: multiplying the decoding weight matrix of the decoder with the fine-grained interactive characterization vector between the multi-dimensional time series features of the water washing liquid state to obtain the decoded fine-grained interactive characterization vector between the multi-dimensional time series features of the water washing liquid state, and accumulating and summing all the eigenvalues ​​of the fine-grained interactive characterization vector between the multi-dimensional time series features of the water washing liquid state to obtain the recommended water-cement ratio value.

[0061] Here, considering that the water washing liquid pH value time series pattern feature vector and the water washing liquid conductivity time series pattern feature vector respectively represent the image semantic features of the Grammar angular field image corresponding to the water washing liquid pH value and the water washing liquid conductivity, when performing attention interaction between features based on the dynamic memory mechanism, due to the differences in the dynamic memory mechanism caused by the differences in the semantic features of the Grammar angular field image corresponding to the time series distribution pattern of the source data, the fine-grained interaction representation vector between the multi-dimensional time series features of the water washing liquid state will have a probabilistic convergence and divergence based on different attention interactions, thereby affecting the accuracy of the recommended water-to-cement ratio value obtained by the decoder-based water-to-cement ratio adjuster.

[0062] Preferably, the fine-grained interaction characterization vector between the multi-dimensional time series features of the water washing liquid state is passed through a decoder-based water-cement ratio adjuster to obtain a recommended water-cement ratio value, including: determining the median eigenvalue, the maximum eigenvalue and the minimum eigenvalue in the fine-grained interaction characterization vector between the multi-dimensional time series features of the water washing liquid state, and dividing the median eigenvalue by the difference between the maximum eigenvalue and the minimum eigenvalue to obtain a fine-grained interaction distribution probability value between the multi-dimensional time series features of the water washing liquid state. The process can be expressed as: ;in, , and represent the median eigenvalue, the maximum eigenvalue and the minimum eigenvalue in the fine-grained interaction representation vector between the multi-dimensional time series features of the water washing liquid state, respectively. Represents the fine-grained interaction distribution probability value between the multi-dimensional time series features of the water washing liquid state.

[0063] The fine-grained interaction representation vector between the multi-dimensional time series features of the water washing liquid state is normalized to the maximum value to obtain the fine-grained interaction probability feature vector between the multi-dimensional time series features of the water washing liquid state .

[0064] The power function of each eigenvalue of the fine-grained interaction probability eigenvector between the multi-dimensional time series features of the water washing liquid state is calculated with one minus the fine-grained interaction distribution probability value between the multi-dimensional time series features of the water washing liquid state as the exponent to obtain the fine-grained interaction convergence eigenvector between the multi-dimensional time series features of the first water washing liquid state. The process can be expressed as: ;in, represents the fine-grained interaction probability feature vector between the multi-dimensional time series features of the water washing liquid state, represents the fine-grained interactive distribution probability value between the multi-dimensional time series features of the water washing liquid state, It means point multiplication by position. Indicates one minus Calculate for index The power function of each eigenvalue of A fine-grained interactive convergence feature vector representing the multi-dimensional time series features of the first water washing liquid state.

[0065] Taking the fine-grained interaction distribution probability value between the multi-dimensional time series features of the water washing liquid state as the exponent, the power function of each eigenvalue of the point difference feature vector between the unit feature vector and the fine-grained interaction probability feature vector between the multi-dimensional time series features of the water washing liquid state is calculated to obtain the fine-grained interaction convergence feature vector between the multi-dimensional time series features of the second water washing liquid state. The process can be expressed as: ;in, represents the unit eigenvector, represents the fine-grained interaction probability feature vector between the multi-dimensional time series features of the water washing liquid state, represents the fine-grained interactive distribution probability value between the multi-dimensional time series features of the water washing liquid state, and They represent point-by-point subtraction and point-by-point multiplication, respectively. Indicates Calculate for index Point reduction The power function of each eigenvalue in the obtained eigenvector is A fine-grained interactive convergence feature vector representing the multi-dimensional time series features of the second water washing liquid state.

[0066] The fine-grained interaction probability feature vector between the multi-dimensional time series features of the water washing liquid state is multiplied by a difference between one minus the fine-grained interaction distribution probability value between the multi-dimensional time series features of the water washing liquid state to obtain a first fine-grained interaction limiting feature vector between the multi-dimensional time series features of the water washing liquid state. This process can be expressed as: ;in, represents the fine-grained interaction probability feature vector between the multi-dimensional time series features of the water washing liquid state, represents the fine-grained interactive distribution probability value between the multi-dimensional time series features of the water washing liquid state, It means point multiplication by position. A fine-grained interactively defined feature vector representing the multi-dimensional time series features of the first water washing liquid state.

[0067] The point difference feature vector is point-multiplied with the fine-grained interaction distribution probability value between the multi-dimensional time series features of the water washing liquid state to obtain the fine-grained interaction limiting feature vector between the multi-dimensional time series features of the second water washing liquid state. This process can be expressed as: ;in, represents the unit eigenvector, represents the fine-grained interaction probability feature vector between the multi-dimensional time series features of the water washing liquid state, represents the fine-grained interactive distribution probability value between the multi-dimensional time series features of the water washing liquid state, and They represent point-by-point subtraction and point-by-point multiplication, respectively. A fine-grained interactive definition feature vector between multi-dimensional time series features representing the state of the second washing liquid.

[0068] After dot-multiplying the fine-grained interactive convergence feature vector between the first water-washing liquid state multidimensional time series features and the fine-grained interactive convergence feature vector between the second water-washing liquid state multidimensional time series features, dot-multiplying the fine-grained interactive limiting feature vector between the first water-washing liquid state multidimensional time series features and the fine-grained interactive limiting feature vector between the second water-washing liquid state multidimensional time series features to obtain an optimized fine-grained interactive representation vector between the water-washing liquid state multidimensional time series features. This process can be expressed as: ;in, represents the fine-grained interactive convergence feature vector between the multi-dimensional time series features of the first water washing liquid state, Indicates the fine-grained interactive convergence feature vector between the multi-dimensional time series features of the second water washing liquid state, A fine-grained interactive feature vector representing the multi-dimensional time series features of the first water washing liquid state, A fine-grained interactive feature vector representing the multi-dimensional time series features of the second water washing liquid state, and They represent point-by-point multiplication and point-by-point addition, respectively. A fine-grained interaction representation vector between multi-dimensional time series features of the optimized water washing liquid state is represented.

[0069] The fine-grained interaction representation vector between the optimized multi-dimensional time series features of the water washing liquid state is passed through the decoder-based water-cement ratio adjuster to obtain the recommended water-cement ratio value.

[0070] That is, by using the cross entropy form power series of the distribution analytical probability of the fine-grained interaction characterization vector between the multi-dimensional time series features of the water washing liquid state as the probability distribution convergence limit, the regression probability convergence approximation is performed on the basis of the combination of the feature set distribution and the probability density distribution of the fine-grained interaction characterization vector between the multi-dimensional time series features of the water washing liquid state, so as to avoid the fatal potential divergence of the fine-grained interaction characterization vector between the multi-dimensional time series features of the water washing liquid state. Under the premise of conservative convergence control, the probability distribution cross entropy of the fine-grained interaction characterization vector between the multi-dimensional time series features of the water washing liquid state is further used as the objective function to guide the limited recovery strategy, so as to achieve the common agility of the convergence of the probability convergence and divergence of the fine-grained interaction characterization vector between the multi-dimensional time series features of the water washing liquid state to the probability density distribution space, and improve the accuracy of the recommended water-to-cement ratio value obtained by the fine-grained interaction characterization vector between the multi-dimensional time series features of the water washing liquid state through the decoder-based water-to-cement ratio adjuster.

[0071] In summary, an adaptive control method for fly ash washing based on an embodiment of the present application is explained, which uses AI-based data processing and encoding technology to group the washing liquid status data in the washing liquid storage tank, and then maps each grouped parameter data set to the Gram angle field and performs characteristic analysis of the timing pattern, thereby adaptively adjusting the water-cement ratio based on the dynamic memory attention fusion representation between the timing pattern characteristics of the washing liquid pH value and the timing pattern characteristics of the washing liquid conductivity. In this way, the changes in the washing process can be understood in real time, and the water-cement ratio can be adaptively adjusted according to the state of the washing liquid after the current washing, so as to ensure that when facing fly ash with large composition fluctuations, stable chlorine and heavy metal removal can be maintained. At the same time, the need for manual adjustment is reduced, and errors are reduced.

[0072] Figure 5FIG. 1 is a system block diagram of an adaptive control system for fly ash washing according to an embodiment of the present application. Figure 5 As shown, according to an embodiment of the present application, the adaptive control system 100 for fly ash washing includes: a washing liquid state data acquisition module 110, which is used to collect a time series set of washing liquid state data in a washing liquid storage tank through a sensor group, wherein the washing liquid state data includes a pH value of the washing liquid and a conductivity of the washing liquid; a washing liquid state data grouping and mapping module 120, which is used to group and map the time series set of the washing liquid state data to obtain a Gram angular field image of a washing liquid pH value mapping and a Gram angular field image of a washing liquid conductivity mapping; a washing liquid state data time series pattern feature The feature extraction module 130 is used to perform time series pattern feature extraction on the Gram angular field image mapped by the pH value of the washing liquid and the Gram angular field image mapped by the electrical conductivity of the washing liquid to obtain the time series pattern features of the pH value of the washing liquid and the time series pattern features of the electrical conductivity of the washing liquid; the water-cement ratio recommendation result generation module 140 is used to perform dynamic memory mechanism attention fusion on the time series pattern features of the pH value of the washing liquid and the time series pattern features of the electrical conductivity of the washing liquid to obtain a multi-dimensional time series representation of the state of the washing liquid, and obtain a recommended water-cement ratio based on the multi-dimensional time series representation of the state of the washing liquid.

[0073] Here, those skilled in the art will appreciate that the specific functions and operations of the various units and modules in the adaptive control system 100 for fly ash washing have been described above with reference to Figures 1 to 4 The adaptive control method for fly ash washing has been described in detail in the description of the adaptive control method for fly ash washing, and therefore, its repeated description will be omitted.

[0074] In summary, an adaptive control system 100 for fly ash washing based on an embodiment of the present application is explained, which uses AI-based data processing and encoding technology to group the water washing liquid status data in the water washing liquid storage tank, and then maps each grouped parameter data set to the Gram angle field and performs characteristic analysis of the timing pattern, thereby adaptively adjusting the water-cement ratio based on the dynamic memory attention fusion representation between the timing pattern characteristics of the water washing liquid pH value and the timing pattern characteristics of the water washing liquid conductivity. In this way, the changes in the water washing process can be understood in real time, and the water-cement ratio can be adaptively adjusted according to the state of the water washing liquid after the current water washing, so as to ensure that when facing fly ash with large composition fluctuations, stable chlorine and heavy metal removal can be maintained. At the same time, the need for manual adjustment is reduced, and errors are reduced.

Claims

1. An adaptive control method for fly ash washing, a fly ash washing process, comprising: The original fly ash sent to the primary washing tank is washed once with washing liquid from the secondary washing liquid storage tank and filtered to obtain primary washed fly ash and primary washing liquid; The primary washing liquid is subjected to wastewater treatment; the primary washing fly ash sent to the secondary washing tank is subjected to secondary washing and filtering using the washing liquid of the tertiary washing liquid storage tank to obtain secondary washing fly ash and secondary washing liquid; discharging the secondary water wash liquid into the secondary water wash liquid storage tank; The secondary washed fly ash sent to the tertiary washing tank is washed three times with water using the washing liquid in the clean water washing liquid storage tank and filtered to obtain final fly ash and tertiary washing liquid; The tertiary washing liquid is discharged into the tertiary washing liquid storage tank; characterized in that the adaptive control method for fly ash washing comprises: in each process of fly ash washing, the water-cement ratio is adjusted in real time based on the state of the washing liquid, including: collecting a time series set of washing liquid state data in the washing liquid storage tank through a sensor group, the washing liquid state data including the pH value of the washing liquid and the conductivity of the washing liquid; grouping and mapping the time series set of the washing liquid state data to obtain a washing liquid pH value mapping Gram angle field image and a water The conductivity of the washing liquid is mapped to a Gramm angular field image; the Gramm angular field image mapped to the pH value of the washing liquid and the Gramm angular field image mapped to the conductivity of the washing liquid are respectively subjected to time series pattern feature extraction to obtain the time series pattern feature of the pH value of the washing liquid and the time series pattern feature of the conductivity of the washing liquid; the time series pattern feature of the pH value of the washing liquid and the time series pattern feature of the conductivity of the washing liquid are subjected to attention fusion using a dynamic memory mechanism to obtain a multi-dimensional time series representation of the state of the washing liquid, and based on the multi-dimensional time series representation of the state of the washing liquid, a recommended water-cement ratio is obtained.

2. The adaptive control method for fly ash washing according to claim 1, characterized in that: The time series set of the washing liquid state data is grouped and mapped to obtain a Gram angular field image of the washing liquid pH value mapping and a Gram angular field image of the washing liquid conductivity mapping, including: grouping the time series set of the washing liquid state data based on the washing liquid state parameter sample dimension to obtain a washing liquid pH value time series set and a washing liquid conductivity time series set; performing Gram angular field transformation on the washing liquid pH value time series set and the washing liquid conductivity time series set respectively to obtain the washing liquid pH value mapping Gram angular field image and the washing liquid conductivity mapping Gram angular field image.

3. The adaptive control method for fly ash washing according to claim 2, characterized in that: The Graham angular field image mapping the pH value of the washing liquid and the Graham angular field image mapping the electrical conductivity of the washing liquid are respectively subjected to time series pattern feature extraction to obtain the time series pattern feature of the pH value of the washing liquid and the time series pattern feature of the electrical conductivity of the washing liquid, including: the Graham angular field image mapping the pH value of the washing liquid and the Graham angular field image mapping the electrical conductivity of the washing liquid are respectively subjected to a water washing liquid state time series pattern feature extractor based on a micro convolutional neural network to obtain a time series pattern feature vector of the pH value of the washing liquid as the time series pattern feature of the pH value of the washing liquid and a time series pattern feature vector of the electrical conductivity of the washing liquid as the time series pattern feature of the electrical conductivity of the washing liquid.

4. The adaptive control method for fly ash washing according to claim 3, characterized in that: A dynamic memory mechanism attention fusion is performed on the time series pattern characteristics of the pH value of the washing liquid and the time series pattern characteristics of the electrical conductivity of the washing liquid to obtain a multi-dimensional time series representation of the water washing liquid state, and a recommended water-cement ratio is obtained based on the multi-dimensional time series representation of the water washing liquid state, including: a dynamic memory mechanism attention fusion is performed on the time series pattern characteristic vector of the pH value of the washing liquid and the time series pattern characteristic vector of the electrical conductivity of the washing liquid to obtain a fine-grained interaction representation vector between the multi-dimensional time series characteristics of the water washing liquid state as the multi-dimensional time series representation of the water washing liquid state; based on the fine-grained interaction representation vector between the multi-dimensional time series characteristics of the water washing liquid state, the recommended water-cement ratio is obtained.

5. The adaptive control method for fly ash washing according to claim 4, characterized in that: The dynamic memory mechanism attention fusion is performed on the water washing liquid pH value time series pattern feature vector and the water washing liquid conductivity time series pattern feature vector to obtain a fine-grained interaction representation vector between the multi-dimensional time series features of the water washing liquid state, including: collaborative implicit feature extraction is performed on the water washing liquid pH value time series pattern feature vector and the water washing liquid conductivity time series pattern feature vector to obtain an implicit collaborative coding vector between the water washing liquid pH value-conductivity time series features; based on the implicit collaborative coding vector between the water washing liquid pH value-conductivity time series features, the water washing liquid pH value time series pattern feature vector and the water washing liquid conductivity time series pattern feature vector are subjected to fine-grained interaction between water washing liquid state features to obtain a fine-grained interaction representation vector between the multi-dimensional time series features of the water washing liquid state.

6. The adaptive control method for fly ash washing according to claim 5, characterized in that: According to the implicit synergistic coding vector between the water washing liquid pH value and the conductivity time series features, the water washing liquid pH value time series pattern feature vector and the water washing liquid conductivity time series pattern feature vector are subjected to fine-grained interaction between water washing liquid state features to obtain a fine-grained interaction representation vector between the multi-dimensional time series features of the water washing liquid state, including: writing the implicit synergistic coding vector between the water washing liquid pH value and the conductivity time series features into a dynamic memory unit to obtain a water washing liquid pH value and conductivity time series dynamic key vector; using the water washing liquid pH value and the conductivity time series pattern feature vector to obtain a fine-grained interaction representation vector between the multi-dimensional time series features of the water washing liquid state; and Taking the conductivity time series dynamic key vector as a guide, dynamic attention modulation is performed on the water washing liquid pH value time series pattern feature vector and the water washing liquid conductivity time series pattern feature vector to obtain the optimized water washing liquid pH value time series pattern feature vector and the optimized water washing liquid conductivity time series pattern feature vector; the optimized water washing liquid pH value time series pattern feature vector and the optimized water washing liquid conductivity time series pattern feature vector are weightedly fused to obtain the fine-grained interaction representation vector between the multi-dimensional time series features of the water washing liquid state.

7. The adaptive control method for fly ash washing according to claim 6, characterized in that: Taking the water washing liquid pH value-conductivity time series dynamic key vector as a guide, the water washing liquid pH value time series pattern feature vector and the water washing liquid conductivity time series pattern feature vector are respectively subjected to dynamic attention modulation to obtain the optimized water washing liquid pH value time series pattern feature vector and the optimized water washing liquid conductivity time series pattern feature vector, including: extracting the water washing liquid pH value-conductivity time series dynamic key vector from the dynamic memory unit, and performing attention modulation of the first converter structure on the water washing liquid pH value time series pattern feature vector and the water washing liquid pH value-conductivity time series dynamic key vector to obtain the optimized water washing liquid pH value time series pattern feature vector; extracting the water washing liquid pH value-conductivity time series dynamic key vector from the dynamic memory unit, and performing attention modulation of the second converter structure on the water washing liquid conductivity time series pattern feature vector and the water washing liquid pH value-conductivity time series dynamic key vector to obtain the optimized water washing liquid conductivity time series pattern feature vector.

8. The adaptive control method for fly ash washing according to claim 7, characterized in that: Based on the fine-grained interaction characterization vector between the multi-dimensional time series features of the washing liquid state, the recommended water-cement ratio value is obtained, including: passing the fine-grained interaction characterization vector between the multi-dimensional time series features of the washing liquid state through a decoder-based water-cement ratio adjuster to obtain the recommended water-cement ratio value.

9. An adaptive control system for fly ash washing, characterized in that: include: A water washing liquid state data acquisition module, used for collecting a time series set of water washing liquid state data in a water washing liquid storage tank through a sensor group, wherein the water washing liquid state data includes a water washing liquid pH value and a water washing liquid conductivity; a water washing liquid state data grouping and mapping module, used for grouping and mapping the time series set of water washing liquid state data to obtain a water washing liquid pH value mapping Gram angular field image and a water washing liquid conductivity mapping Gram angular field image; a water washing liquid state data time series pattern feature extraction module, used for respectively performing time series pattern feature extraction on the water washing liquid pH value mapping Gram angular field image and the water washing liquid conductivity mapping Gram angular field image to obtain a water washing liquid pH value time series pattern feature and a water washing liquid conductivity time series pattern feature; The water-cement ratio recommendation result generation module is used to perform dynamic memory mechanism attention fusion on the water washing liquid pH value time series pattern characteristics and the water washing liquid conductivity time series pattern characteristics to obtain a multi-dimensional time series representation of the water washing liquid state, and obtain a recommended water-cement ratio based on the multi-dimensional time series representation of the water washing liquid state.

10. The adaptive control system for fly ash washing according to claim 9, characterized in that: The water washing liquid state data grouping mapping module includes: a water washing liquid state time series data grouping unit, which is used to group the time series set of the water washing liquid state data based on the water washing liquid state parameter sample dimension to obtain a water washing liquid pH value time series set and a water washing liquid conductivity time series set; a water washing liquid state data mapping unit, which is used to perform Gram angular field transformation on the water washing liquid pH value time series set and the water washing liquid conductivity time series set respectively to obtain the water washing liquid pH value mapping Gram angular field image and the water washing liquid conductivity mapping Gram angular field image.

Images