Environment-friendly hydrofluoric acid waste liquid conversion and recovery treatment system

Abstract

The invention discloses an environment-friendly hydrofluoric acid waste liquid conversion and recovery treatment system, and relates to the technical field of resource recovery. The method comprises the following steps: deploying a sampling sensor array in a hydrofluoric acid waste liquid storage pool, acquiring particle and fluorine ion concentration data by an acquisition module according to a preset frequency, and constructing a time sequence data matrix; the distribution field construction module generates a concentration distribution field by using a UK-EVF spatial interpolation algorithm, and extracts characteristic parameters such as average concentration, concentration gradient and variable coefficient; the type determination module inputs the time sequence data matrix and the characteristic parameters into a classification identification model, and accurately judges the type of the waste liquid; and the recovery processing module automatically calls a corresponding environment-friendly recovery scheme accordingly. The defect that the concentration distribution in the pool cannot be comprehensively mastered through traditional single-point detection is overcome, precise classification and targeted recovery are achieved, and the treatment efficiency and the resource recovery rate of the hydrofluoric acid waste liquid are improved.

Description

An environmentally friendly hydrofluoric acid waste liquid conversion and recycling system Technical Field

[0001] This invention belongs to the field of resource recycling technology, specifically relating to an environmentally friendly hydrofluoric acid waste liquid conversion and recycling system. Background Technology

[0002] Hydrofluoric acid is widely used in various industrial fields such as integrated circuits, photovoltaics, electronic electroplating, and fluorochemicals, and is one of the core raw materials in industrial production. In the process of using hydrofluoric acid, a large amount of hydrofluoric acid waste liquid is inevitably generated. This waste liquid is highly corrosive and toxic, and the concentration of fluoride ions and particles in the waste liquid fluctuates greatly. If directly discharged, it will seriously pollute the soil, groundwater, and surface water, damage the ecological environment, and threaten human health. On the other hand, hydrofluoric acid waste liquid contains recyclable fluorine resources. If it is treated by simply using harmless disposal methods, it will result in a serious waste of fluorine resources, which is not in line with the current industrial development trend of green, low-carbon, and resource recycling. Therefore, the environmentally friendly conversion and recycling treatment of hydrofluoric acid waste liquid has become an important research direction in this field.

[0003] Currently, the core technical problem facing existing hydrofluoric acid waste liquid conversion and recycling technologies is the inability to scientifically classify the waste liquid based on accurate concentration monitoring data. This hinders targeted recycling and treatment, resulting in low recycling efficiency and low resource recovery rates, making it difficult to achieve environmentally friendly and resource-efficient waste liquid recovery. Existing technologies often rely on manual or semi-automated detection at single or limited sampling points for concentration monitoring of hydrofluoric acid waste liquid. This lack of a systematic sampling layout prevents comprehensive and accurate acquisition of the concentration distribution throughout the waste liquid storage tank, leading to an incomplete and inaccurate understanding of the concentration distribution patterns and ultimately resulting in low conversion and recycling efficiency for hydrofluoric acid waste liquid. Summary of the Invention

[0004] The purpose of this invention is to solve the problem of low conversion and recovery efficiency of hydrofluoric acid waste liquid due to the inability to fully and accurately obtain the concentration distribution of the entire waste liquid storage tank, and to propose an environmentally friendly hydrofluoric acid waste liquid conversion and recovery treatment system.

[0005] This invention proposes an environmentally friendly hydrofluoric acid waste liquid conversion and recycling system. The system includes: an array of sampling sensors formed by installing multiple online monitoring points in a hydrofluoric acid waste liquid storage tank; a data acquisition module for acquiring particle concentration and fluoride ion concentration data recorded by each sampling sensor at a preset frequency to obtain a concentration dataset, and constructing a time-series data matrix based on the concentration dataset; a distribution field construction module for constructing a full-tank concentration distribution field based on the concentration dataset using a UK-EVF spatial interpolation algorithm, and extracting feature parameters of the full-tank concentration distribution field; the feature parameters include average concentration, maximum concentration, minimum concentration, concentration gradient, and coefficient of variation; a type determination module for substituting the time-series data matrix and the feature parameters into a preset waste liquid classification and identification model to obtain the target waste liquid type; and a recycling and processing module for automatically calling the corresponding environmentally friendly recycling scheme according to the target waste liquid type, thereby performing recycling and processing.

[0006] By utilizing the UK-EVF spatial interpolation algorithm to construct the concentration distribution field of the entire pool and extracting multi-dimensional feature parameters such as average concentration, concentration gradient, and coefficient of variation, the limitations of traditional methods in fully and accurately grasping the concentration distribution pattern within the waste liquid pool are overcome. Furthermore, by inputting the time-series data matrix and feature parameters into the waste liquid classification and identification model, the target waste liquid type can be accurately determined, thereby automatically calling the corresponding environmental protection recycling scheme and improving the conversion and recycling efficiency of hydrofluoric acid waste liquid.

[0007] Optionally, each sampling point in the sampling sensor array includes a fluoride ion concentration sensor; the fluoride ion concentration sensor is an immersion-type online fluoride ion selective electrode, which has a built-in temperature sensor and transmitter, and directly outputs a temperature-compensated digital fluoride ion concentration value; the front end of the fluoride ion selective electrode is provided with a microporous permeable membrane and a built-in buffer tank, which are used to maintain the stability of the ion strength at the electrode measurement interface and eliminate the interference of metal ions and complexed ions in the waste liquid.

[0008] Optionally, the distribution field construction module includes: a preprocessing module for preprocessing the concentration dataset to obtain an effective concentration dataset; a model construction module for initializing the UK-EVF spatial interpolation algorithm and constructing a linear observation model and a state transition model based on the effective concentration dataset and corresponding sampling spatial coordinates; the linear observation model is used to associate sampling point coordinates with corresponding concentrations; the state transition model is used to associate concentration relationships between different sampling point coordinates; a concentration estimation module is used to determine the optimal concentration estimate for each predicted point in the hydrofluoric acid waste storage pool based on the linear observation model and the state transition model; and a distribution field generation module is used to substitute the optimal concentration estimate for each predicted point into a preset feature vector field model for spatial smoothing, and perform point-by-point interpolation calculations on the entire spatial region of the hydrofluoric acid waste storage pool using a preset interpolation step size to generate a full-pool concentration distribution field; the full-pool concentration distribution field includes particle concentration values ​​and fluoride ion concentration values ​​corresponding to all spatial coordinates within the waste pool.

[0009] By combining data preprocessing and UK-EVF spatial interpolation, an observation and state transition model can be accurately constructed to determine the optimal concentration estimate for each point in the entire storage pool. Then, spatial smoothing and point-by-point interpolation are used to generate a complete concentration distribution field. Compared with traditional single-point detection, this significantly improves the spatial coverage and numerical accuracy of hydrofluoric acid waste liquid concentration monitoring.

[0010] Optionally, the concentration estimation module includes: a predicted sample point generation module, used to generate 2n+1 Sigma points through unscented transformation, substitute the Sigma points into the state transition model for spatial propagation to obtain predicted Sigma points, and calculate the state prior mean and prior covariance matrix based on the predicted Sigma points and preset weights; a sample point observation module, used to substitute the predicted Sigma points into a linear observation model for propagation to the observation space, and calculate the predicted observation mean, observation covariance matrix, and cross-covariance matrix; and a Kalman gain determination module, used to calculate the Kalman gain based on the cross-covariance matrix and the observation covariance matrix, update the state prior mean and covariance matrix based on the predicted observation mean and the Kalman gain, and obtain the optimal concentration estimate for each predicted point based on the updated state prior mean and covariance matrix.

[0011] By accurately generating Sigma points through unscented transformation to complete spatial propagation and observation mapping, and then iteratively updating the prior mean, covariance matrix and Kalman gain, adaptive optimal filtering and correction of concentration state are achieved, effectively avoiding nonlinear interpolation errors, and significantly improving the accuracy, stability and robustness of concentration estimation at each prediction point in the hydrofluoric acid waste storage pool, thus providing a foundation for the subsequent generation of a high-precision full-pool concentration distribution field.

[0012] Optionally, the system further includes a purification module, comprising: the purification module being used for closed-loop control of the purification filtration device through a differential pressure feedback control model; the purification filtration device being a PTFE membrane filtration device installed at the inlet of the hydrofluoric acid waste storage tank; pressure sensors being installed at the inlet and outlet of the filtration device for collecting differential pressure signals; the closed-loop control being performed by running an incremental PID algorithm through a PLC controller to dynamically adjust the output frequency of the feed pump inverter according to the deviation between the differential pressure and a preset pressure threshold; when the differential pressure exceeds a preset dangerous differential pressure value, the feed flow rate is reduced and pulse backwashing is initiated until the differential pressure returns to below the preset safe differential pressure value.

[0013] Optionally, the type determination module includes: a branch input determination module, used for the preset waste liquid classification and recognition model including a first input branch and a second input branch; the first input branch is used to receive the time-series data matrix; the second input branch receives the feature parameters; a branch output determination module, used for the first input branch to sequentially substitute the time-series data matrix into a convolutional layer and a pooling layer to obtain a time-series feature vector; the second input branch to input the feature parameters into a fully connected layer to obtain a spatial feature vector; and a type generation module, used to concatenate and fuse the time-series feature vector and the spatial feature vector, and then sequentially substitute them into a fully connected layer and a Softmax layer to obtain a waste liquid type probability value; the waste liquid types include low-impurity type, mixed acid type, and high-metal salt type; and the waste liquid type with the highest probability value is taken as the target waste liquid type.

[0014] By extracting deep temporal features from the time-series data matrix in the first branch and spatial features from the feature parameters in the second branch, a comprehensive and accurate representation of waste liquid information is achieved. After feature splicing and classification output, it can efficiently and accurately distinguish between three types of waste liquid: low-impurity type, mixed acid type, and high metal salt type. This effectively avoids the bias of single feature identification and greatly improves the accuracy and robustness of waste liquid type determination, providing a precise classification basis for subsequent precise treatment and resource utilization of waste liquid.

[0015] Optionally, the recovery and treatment module includes: a low-impurity treatment module, used to automatically switch to distillation purification process and initialize distillation column parameters if the target waste liquid is of low impurity type; a mixed acid treatment module, used to automatically switch to azeotropic distillation process, add extractant tributyl phosphate, and initialize distillation column parameters if the target waste liquid is of mixed acid type; and a high-metal salt treatment module, used to automatically switch to fluoride salt conversion process and initialize reactor parameters if the target waste liquid is of high-metal salt type.

[0016] The beneficial effects of this invention are as follows: This invention proposes an environmentally friendly hydrofluoric acid waste liquid conversion and recycling system. By using the UK-EVF spatial interpolation algorithm to construct the concentration distribution field of the entire pool and extracting multi-dimensional feature parameters such as average concentration, concentration gradient, and coefficient of variation, it overcomes the limitation of traditional methods in that they cannot fully and accurately grasp the concentration distribution law in the waste liquid pool. Furthermore, by inputting the time series data matrix and feature parameters into the waste liquid classification and identification model, the target waste liquid type can be accurately determined, thereby automatically calling the corresponding environmentally friendly recycling scheme and improving the conversion and recycling efficiency of hydrofluoric acid waste liquid. Attached Figure Description

[0017] The invention will now be further described with reference to the accompanying drawings.

[0018] Figure 1 is a flowchart of a method for implementing an environmentally friendly hydrofluoric acid waste liquid conversion and recycling system according to an embodiment of the present invention. Detailed Implementation

[0019] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments.

[0020] Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0021] This invention provides an environmentally friendly hydrofluoric acid waste liquid conversion and recycling system. Referring to Figure 1, Figure 1 is a flowchart illustrating a method for implementing an environmentally friendly hydrofluoric acid waste liquid conversion and recycling system according to an embodiment of this invention. A sampling sensor array is formed by installing multiple online monitoring points in a hydrofluoric acid waste liquid storage tank. The system includes a data acquisition module, a distribution field construction module, a type determination module, and a recycling and treatment module. The data acquisition module is used to implement step S101, the distribution field construction module is used to implement step S102, the type determination module is used to implement step S103, and the recycling and treatment module is used to implement step S104. In step S101, the particle concentration and fluoride ion concentration data recorded by each sampling sensor are acquired at a preset frequency to obtain a concentration dataset, and a time-series data matrix is ​​constructed based on the concentration dataset. In step S102, based on the concentration dataset, the concentration distribution field of the entire tank is constructed using the UK-EVF spatial interpolation algorithm, and the feature parameters of the concentration distribution field of the entire tank are extracted. In step S103, the time-series data matrix and feature parameters are substituted into a preset waste liquid classification and identification model to obtain the target waste liquid type. In step S104, the corresponding environmental protection recycling scheme is automatically called according to the target waste liquid type to carry out recycling and treatment.

[0022] In one implementation, the system is deployed in a hydrofluoric acid waste liquid storage tank and includes multiple online monitoring points. Each monitoring point is equipped with a particle concentration sensor and a fluoride ion concentration sensor. The particle concentration sensor is used to output the number concentration of insoluble solid particles in the waste liquid at that point in real time. The fluoride ion concentration sensor is used to output the mass concentration of fluoride ions in the waste liquid at that point in real time. The particle concentration sensor model can be RIONKS-20F, KS-42AF, etc., and the fluoride ion concentration sensor can be HORIBAHF-960EM, HORIBAHF-960M, etc.

[0023] In one implementation, the preset frequency is determined by technicians; the system continuously collects particle concentration and fluoride ion concentration data from each sampling sensor within a fixed time window according to the preset sampling frequency; the concentration data of all sampling points at the same moment are arranged in a fixed order as a single row of data with time sequence as the row dimension and the concentration data of each sensor as the column dimension, and then spliced ​​together sequentially point by point to form a two-dimensional time-series data matrix of dimension T×N; where T is the total number of samples within the time window and N is the concentration data dimension of all sensors, thereby transforming discrete concentration data into standardized time-series features.

[0024] In one implementation, the characteristic parameters include average concentration, maximum concentration, minimum concentration, concentration gradient, and coefficient of variation; a pre-set environmental recycling scheme corresponding to each of the three types of waste liquid is provided: low impurity type corresponds to distillation purification, mixed acid type corresponds to azeotropic distillation, and high metal salt type corresponds to fluoride salt conversion; after receiving the output of the type determination module, the initialization operation of the corresponding process package is automatically triggered, the pre-set process parameters are called, and the waste liquid recycling treatment is completed automatically throughout the entire process without manual intervention in process switching.

[0025] In one implementation, each sampling point in the sampling sensor array includes a fluoride ion concentration sensor; the fluoride ion concentration sensor is an immersion-type online fluoride ion selective electrode, which has a built-in temperature sensor and transmitter, and directly outputs a temperature-compensated digital fluoride ion concentration value; the front end of the fluoride ion selective electrode is equipped with a microporous permeable membrane and a built-in buffer tank to maintain the stability of the ion strength at the electrode measurement interface and eliminate Al in the waste liquid. 3+ Fe 3+ Interference from metal ions and their complexes.

[0026] In one embodiment, the distribution field construction module includes: a preprocessing module for preprocessing the concentration dataset to obtain an effective concentration dataset; a model construction module for initializing the UK-EVF spatial interpolation algorithm and constructing a linear observation model and a state transition model based on the effective concentration dataset and corresponding sampling spatial coordinates; the linear observation model is used to associate the sampling point coordinates with the corresponding concentration; the state transition model is used to associate the concentration relationship between different sampling point coordinates; a concentration estimation module is used to determine the optimal concentration estimate of each prediction point in the hydrofluoric acid waste storage pool based on the linear observation model and the state transition model; and a distribution field generation module is used to substitute the optimal concentration estimate of each prediction point into a preset feature vector field model for spatial smoothing, and perform point-by-point interpolation calculation on the entire spatial region of the hydrofluoric acid waste storage pool through a preset interpolation step size to generate a full pool concentration distribution field; the full pool concentration distribution field includes particle concentration values ​​and fluoride ion concentration values ​​corresponding to all spatial coordinates in the waste pool.

[0027] In one implementation, preprocessing of the concentration dataset includes existing processing steps such as outlier removal, missing value imputation, and data calibration to ensure the accuracy and usability of the concentration data.

[0028] In one implementation, the concentration observations in the effective concentration dataset are bound one-to-one with the three-dimensional spatial coordinates of the corresponding sampling points to construct a linear observation model Z=H·M+V; where Z is the sensor concentration observation, H is the observation matrix constructed based on the spatial coordinates, M is the spatial state vector, and V is the observation noise constant, which is determined by the technicians; this model realizes the linear mapping between the spatial coordinates of the sampling points and the concentration observations.

[0029] In one implementation, a linear spatial state transition model X is constructed based on the spatial relationship of sampling points within the waste liquid pool. k+1 =A·X k +W; where X k Let X be the current state vector. k+1 Let A be the state vector of adjacent points, A be the state transition matrix constructed based on coordinate difference and spatial distance, and W be the process noise. This model characterizes the continuous transmission and change of concentration between different spatial points.

[0030] In one implementation, the preset feature vector field model is a continuous smooth model based on spatial gradient, which is a general smooth model in the field of spatial data processing. Its construction method is as follows: based on the concentration of each prediction point and its spatial gradient, a spatial smooth relationship is constructed by using gradient consistency constraints and Gaussian kernel weighting functions; the optimal concentration estimate is smoothed through this model, so that the concentration change of adjacent points is continuous and the gradient is gentle, eliminating local abnormal fluctuations caused by interpolation, and ensuring that the concentration distribution of the whole pool conforms to the actual spatial distribution law of fluid.

[0031] In one implementation, based on the actual size of the waste liquid tank and the required monitoring accuracy, technicians pre-set a fixed interpolation step size. Using the three-dimensional spatial boundary of the storage tank as the range, the entire spatial location within the tank is traversed in a grid according to the preset interpolation step size to generate regular grid points covering the entire area. For each grid point, a spatial weighted interpolation algorithm is used, combined with the smoothed predicted point concentration value, to calculate the particle concentration and fluoride ion concentration corresponding to that grid point. The concentration calculation of all grid points is completed point by point, forming a fully covered, equally spaced, and high-precision spatial concentration data set.

[0032] In one implementation, the spatial coordinates (X, Y, Z) of all gridded traversal points are bound one-to-one with the corresponding calculated particle concentration and fluoride ion concentration values. According to the spatial position relationship, all point data are integrated to construct a three-dimensional spatial concentration distribution field. This distribution field completely contains the particle concentration and fluoride ion concentration values ​​corresponding to any spatial coordinate in the waste liquid storage tank, which can intuitively reflect the high and low concentration distribution, concentration gradient and uniformity of the entire tank range. Finally, standardized whole tank concentration distribution field data that can be used for feature extraction and waste liquid classification are output.

[0033] In one embodiment, the concentration estimation module includes: a predicted sample point generation module, used to generate 2n+1 Sigma points through unscented transformation, substitute the Sigma points into a state transition model for spatial propagation to obtain predicted Sigma points, and calculate the state prior mean and prior covariance matrix based on the predicted Sigma points and preset weights; a sample point observation module, used to substitute the predicted Sigma points into a linear observation model for propagation to the observation space, and calculate the predicted observation mean, observation covariance matrix, and cross-covariance matrix; and a Kalman gain determination module, used to calculate the Kalman gain based on the cross-covariance matrix and the observation covariance matrix, update the state prior mean and covariance matrix based on the predicted observation mean and the Kalman gain, and obtain the optimal concentration estimate for each predicted point based on the updated state prior mean and covariance matrix.

[0034] In one implementation, the prediction sample point generation module first performs an unscented transformation based on the system state variable dimension n to generate sampled Sigma points. Based on the current state vector dimension n, 2n+1 Sigma points are calculated and generated using the unscented transformation rules, and corresponding mean weights and covariance weights are configured. Specifically, the mean weights and covariance weights are determined as follows: historical concentration data of hydrofluoric acid waste liquid under different operating conditions, concentrations, and spatial locations are pre-collected, including the concentration values ​​of the sampling points, corresponding spatial coordinates, and the true values ​​of the actual monitored concentrations, forming a historical training dataset. With the optimization objective of minimizing the error between the estimated concentration value and the actual concentration value, a weight fitness function is constructed: minimizing the mean square error between the predicted concentration value and the measured concentration value. Within a reasonable range of unscented transformation weight values, the mean weights and covariance weights are iterated and optimized. Different weight combinations are substituted into the UK-EVF model for concentration estimation, the estimation error is calculated, and the set of weights that minimizes the concentration estimation error is selected. The weighted values, serving as the optimal mean weight and optimal covariance weight for this system, are used to substitute the generated Sigma points into the pre-built state transition model to complete spatial propagation calculations, resulting in a set of Sigma points after state prediction. Based on the propagated Sigma points and preset weights, a weighted summation operation is performed to calculate the prior mean of the state. Based on the deviation between the Sigma points and the prior mean, and the preset covariance weights, a weighted calculation is performed to obtain the prior covariance matrix of the state. The prior mean and prior covariance matrix of the state are output, providing basic prediction data for subsequent observation and update steps.

[0035] In one implementation, the predicted Sigma points after state propagation are substituted into a pre-constructed linear observation model to map the Sigma points from the state space to the observation space, obtaining the observation domain Sigma points. Based on the observation domain Sigma points and preset weights, the predicted observation mean is calculated using weighted averages. Based on the deviation between the observation domain Sigma points and the predicted observation mean, the observation covariance matrix is ​​calculated using preset weights. Based on the deviation between the state space Sigma points and the state prior mean, and the deviation between the observation domain Sigma points and the predicted observation mean, the cross-covariance matrix between the state variables and the observed variables is calculated, and the predicted observation mean, the observation covariance matrix, and the cross-covariance matrix are output.

[0036] In one implementation, the Kalman gain is calculated based on the inverse matrix product of the cross-covariance matrix and the observation covariance matrix. The observation residual is obtained by calculating the difference between the actual observed value and the predicted observation mean. The state prior mean is corrected and updated using the Kalman gain and the observation residual to obtain the state posterior mean. The state covariance matrix is ​​updated by combining the Kalman gain, the observation covariance matrix, and the prior covariance matrix to obtain the state posterior covariance matrix. The updated state posterior mean is output as the optimal concentration estimate for the current spatial prediction point, thus completing the optimal concentration estimate for that point.

[0037] In one embodiment, the system further includes a purification module, comprising: a purification module for closed-loop control of the purification filtration device via a differential pressure feedback control model; the purification filtration device is a PTFE membrane filtration device installed at the inlet of the hydrofluoric acid waste storage tank; pressure sensors are installed at the inlet and outlet of the filtration device for collecting differential pressure signals; the closed-loop control is performed by running an incremental PID algorithm through a PLC controller to dynamically adjust the output frequency of the feed pump inverter according to the deviation between the differential pressure and a preset pressure threshold; when the differential pressure exceeds a preset dangerous differential pressure value, the feed flow rate is reduced and pulse backwashing is initiated until the differential pressure returns to below the preset safe differential pressure value.

[0038] In one implementation, the preset dangerous differential pressure value is 0.25 MPa; the preset safe differential pressure value is 0.12 MPa.

[0039] In one embodiment, the type determination module includes: a branch input determination module, used to preset a waste liquid classification and recognition model including a first input branch and a second input branch; the first input branch is used to receive a time-series data matrix; the second input branch receives feature parameters; a branch output determination module, used to have the first input branch sequentially substitute the time-series data matrix into a convolutional layer and a pooling layer to obtain a time-series feature vector; the second input branch inputs the feature parameters into a fully connected layer to obtain a spatial feature vector; and a type generation module, used to concatenate and fuse the time-series feature vector and the spatial feature vector, and then sequentially substitute them into a fully connected layer and a Softmax layer to obtain a waste liquid type probability value; the waste liquid types include low-impurity type, mixed acid type, and high-metal salt type; and the waste liquid type with the highest probability value is taken as the target waste liquid type.

[0040] In one implementation, the waste liquid classification and identification model includes two input branches: the first branch receives the particle concentration time-series data matrix and extracts time-series features through a 1D convolutional layer and a pooling layer; the second branch receives the distribution field feature parameters and extracts spatial distribution features through a fully connected layer; the features from the two branches are fused and output as waste liquid type probabilities through a fully connected layer and a Softmax layer. The specific implementation steps include: Step 1: Data preparation and adaptation. The time-series data matrix output by the acquisition module and the feature parameters output by the distribution field construction module are standardized to eliminate the dimensional differences between different dimensions of the data. Time series data of concentration and fluoride ion concentration were normalized and scaled to the [0,1] interval; feature parameters (average concentration, maximum concentration, etc.) were standardized to have a mean of 0 and a variance of 1, ensuring that the two types of data could be adapted to the input requirements of the waste liquid classification and identification model; Step 2: The model is input simultaneously into two branches. The standardized time series data matrix is ​​input into the first branch of the waste liquid classification and identification model, and the standardized feature parameters are input into the second branch of the model to achieve simultaneous extraction of time series features and spatial distribution features: The first branch is used for time series feature extraction: The time series data matrix is ​​first input into a 1D convolution The first layer extracts features of concentration variation patterns over time using pre-defined convolutional kernels (3×1, 32 kernels), filtering redundant data. This is then fed into a pooling layer (using max pooling, 2×1 kernel size) to reduce the dimensionality of the extracted temporal features, preserving core temporal information (such as concentration fluctuation frequency and peak variation patterns) to obtain a temporal feature vector. The second branch extracts spatial distribution features: distribution field feature parameters (average concentration, maximum concentration, minimum concentration, concentration gradient, and coefficient of variation) are directly input into the fully connected layer, and through linear transformation and an activation function (using ReLU), the features are extracted... Step 3: Feature fusion and classification operation. The temporal feature vector output by the first branch is concatenated and fused with the spatial feature vector output by the second branch to obtain the fused feature vector. The fused feature vector is input into the subsequent fully connected layer, and the deep correlation information of the fused features is further extracted through multi-layer linear transformation. Finally, it is input into the Softmax layer to output the probability values ​​of various waste liquid types (low impurity type, mixed acid type, high metal salt type). Step 4: The waste liquid type with the highest probability value is taken as the target waste liquid type.

[0041] In one embodiment, the recycling and processing module includes: a low-impurity processing module, used to automatically switch to a distillation purification process and initialize the distillation column parameters if the target waste liquid is of low impurity type; a mixed acid processing module, used to automatically switch to an azeotropic distillation process, add the extractant tributyl phosphate, and initialize the distillation column parameters if the target waste liquid is of mixed acid type; and a high-metal salt processing module, used to automatically switch to a fluoride salt conversion process and initialize the reactor parameters if the target waste liquid is of high-metal salt type.

[0042] In one implementation, if the target waste liquid is of low impurity type, the initial distillation column parameters are: top temperature 45±0.5℃, bottom temperature 105±2℃, reflux ratio 3.5, and top pressure -50Pa; if the target waste liquid is of mixed acid type, the initial distillation column parameters are: top temperature 52±0.5℃, bottom temperature 110±2℃, and reflux ratio 4.0, with hydrofluoric acid collected from the side stream; if the target waste liquid is of high metal salt type, the initial reactor parameters are: stirring speed 120rpm, reaction temperature 75±2℃, the theoretical dosage of sodium hydroxide is calculated based on the average fluoride ion concentration, the final pH of the reaction is controlled at 8.5±0.3, and sodium fluoride product is obtained after solid-liquid separation after the reaction.

[0043] In one implementation method, the distillation purification process, the azeotropic distillation process, and the fluoride salt conversion process are all commonly used existing technologies on the market, and will not be described in detail here.

[0044] The foregoing has described one embodiment of the present invention in detail, but this content is merely a preferred embodiment and should not be considered as limiting the scope of the present invention. All equivalent variations and modifications made within the scope of the claims of this invention should still fall within the scope of the claims of this invention.

Claims

1. An environmentally friendly hydrofluoric acid waste liquid conversion and recycling system, characterized in that, A sampling sensor array consisting of multiple online monitoring points is installed in a hydrofluoric acid waste liquid storage tank. The system includes: a data acquisition module, used to acquire particle concentration and fluoride ion concentration data recorded by each sampling sensor at a preset frequency to obtain a concentration dataset, and construct a time-series data matrix based on the concentration dataset; a distribution field construction module, used to construct a full-tank concentration distribution field based on the concentration dataset using the UK-EVF spatial interpolation algorithm, and extract feature parameters of the full-tank concentration distribution field; the feature parameters include average concentration, maximum concentration, minimum concentration, concentration gradient, and coefficient of variation; a type determination module, used to substitute the time-series data matrix and the feature parameters into a preset waste liquid classification and identification model to obtain the target waste liquid type; and a recycling and processing module, used to automatically call the corresponding environmental recycling scheme according to the target waste liquid type, thereby performing recycling and processing.

2. The environmentally friendly hydrofluoric acid waste liquid conversion and recycling treatment system according to claim 1, characterized in that, Each sampling point in the sampling sensor array includes a fluoride ion concentration sensor; the fluoride ion concentration sensor is an immersion-type online fluoride ion selective electrode, which has a built-in temperature sensor and transmitter, and directly outputs a temperature-compensated digital fluoride ion concentration value; the front end of the fluoride ion selective electrode is provided with a microporous permeable membrane and a built-in buffer tank, which are used to maintain the stability of the ion strength at the electrode measurement interface and eliminate the interference of metal ions and complexed ions in the waste liquid.

3. The environmentally friendly hydrofluoric acid waste liquid conversion and recycling treatment system according to claim 1, characterized in that, The distribution field construction module includes: a preprocessing module for preprocessing the concentration dataset to obtain an effective concentration dataset; a model construction module for initializing the UK-EVF spatial interpolation algorithm and constructing a linear observation model and a state transition model based on the effective concentration dataset and corresponding sampling spatial coordinates; the linear observation model is used to associate sampling point coordinates with corresponding concentrations; the state transition model is used to associate concentration relationships between different sampling point coordinates; a concentration estimation module is used to determine the optimal concentration estimate for each prediction point in the hydrofluoric acid waste storage pool based on the linear observation model and the state transition model; and a distribution field generation module is used to substitute the optimal concentration estimate for each prediction point into a preset feature vector field model for spatial smoothing, and perform point-by-point interpolation calculations on the entire spatial region of the hydrofluoric acid waste storage pool using a preset interpolation step size to generate a full-pool concentration distribution field; the full-pool concentration distribution field includes particle concentration values ​​and fluoride ion concentration values ​​corresponding to all spatial coordinates within the waste pool.

4. The environmentally friendly hydrofluoric acid waste liquid conversion and recycling treatment system according to claim 3, characterized in that, The concentration estimation module includes: a predicted sample point generation module, used to generate 2n+1 Sigma points through unscented transformation, substitute the Sigma points into the state transition model for spatial propagation to obtain predicted Sigma points, and calculate the state prior mean and prior covariance matrix based on the predicted Sigma points and preset weights; a sample point observation module, used to substitute the predicted Sigma points into a linear observation model for propagation to the observation space, and calculate the predicted observation mean, observation covariance matrix, and cross-covariance matrix; and a Kalman gain determination module, used to calculate the Kalman gain based on the cross-covariance matrix and the observation covariance matrix, update the state prior mean and covariance matrix based on the predicted observation mean and the Kalman gain, and obtain the optimal concentration estimate for each predicted point based on the updated state prior mean and covariance matrix.

5. The environmentally friendly hydrofluoric acid waste liquid conversion and recycling treatment system according to claim 1, characterized in that, The system also includes a purification module, comprising: the purification module, used for closed-loop control of the purification filtration device through a differential pressure feedback control model; the purification filtration device is a PTFE membrane filtration device, installed at the inlet of the hydrofluoric acid waste storage tank; pressure sensors are installed at the inlet and outlet of the filtration device to collect differential pressure signals; the closed-loop control is to dynamically adjust the output frequency of the feed pump inverter according to the deviation between the differential pressure and a preset pressure threshold by running an incremental PID algorithm through a PLC controller; when the differential pressure exceeds a preset dangerous differential pressure value, the feed flow rate is reduced and pulse backwashing is initiated until the differential pressure returns to below the preset safe differential pressure value.

6. The environmentally friendly hydrofluoric acid waste liquid conversion and recycling treatment system according to claim 1, characterized in that, The type determination module includes: a branch input determination module, used for the preset waste liquid classification and recognition model including a first input branch and a second input branch; the first input branch is used to receive the time-series data matrix; the second input branch receives the feature parameters; a branch output determination module, used for the first input branch to sequentially substitute the time-series data matrix into a convolutional layer and a pooling layer to obtain a time-series feature vector; the second input branch to input the feature parameters into a fully connected layer to obtain a spatial feature vector; and a type generation module, used to concatenate and fuse the time-series feature vector and the spatial feature vector, and then sequentially substitute them into a fully connected layer and a Softmax layer to obtain a waste liquid type probability value; the waste liquid types include low-impurity type, mixed acid type, and high-metal salt type; the waste liquid type with the highest probability value is taken as the target waste liquid type.

7. The environmentally friendly hydrofluoric acid waste liquid conversion and recycling treatment system according to claim 6, characterized in that, The recycling and processing module includes: a low-impurity processing module, which automatically switches to a distillation purification process and initializes the distillation column parameters if the target waste liquid is of low impurity type; a mixed acid processing module, which automatically switches to an azeotropic distillation process, adds the extractant tributyl phosphate, and initializes the distillation column parameters if the target waste liquid is of mixed acid type; and a high-metal salt processing module, which automatically switches to a fluoride salt conversion process and initializes the reactor parameters if the target waste liquid is of high-metal salt type.

Images