Intelligent impurity removal control method and system

Abstract

The invention discloses an intelligent impurity removal control method and system, and relates to the technical field of intelligent impurity removal, and the method comprises the steps: employing a genetic algorithm to optimize the specific configuration of cleaning parameters in a cleaning mode, transmitting the optimized cleaning parameters to an execution mechanism to start a cleaning program, continuously monitoring the operation state in the cleaning process, and carrying out the cleaning operation. And obtaining a cleaning effect evaluation report. According to the method, the multi-target fitness function of the impurity removal efficiency, the equipment utilization rate and the environmental influence degree is defined, and the multi-target conflict of the cleaning parameters is solved by utilizing the global search capability of the genetic algorithm. On the basis that the impurity removal effect is guaranteed, comprehensive optimization of reduction of agent consumption and loss reduction of equipment is achieved, meanwhile, the stability of the cleaning process is maintained through a dynamic fine adjustment mechanism, and the cleaning and impurity removal technology is promoted to develop towards the environment-friendly and efficient direction.

Description

Technical Field

[0001] The present invention relates to the technical field of intelligent impurity removal, and in particular to a control method and system for intelligent impurity removal. Background Art

[0002] With the development of industrial and intelligent manufacturing technologies, the demand for refined impurity treatment has increased significantly. Ordinary impurity removal control methods mainly rely on manual experience to set fixed thresholds, and their technical routes can be traced back to the mechanical adjustment architecture based on PID controllers in the 1980s. In complex industrial scenarios, the non-linear coupling relationship of production parameters and random disturbance factors severely restrict the flexibility of intelligent impurity removal control systems, and it is difficult to achieve full-process adaptive optimization.

[0003] The technical bottlenecks of existing intelligent impurity removal control methods are reflected in two aspects: the pre-trained reinforcement learning model cannot respond to sudden changes in working conditions in real time, resulting in a delay in the switching of cleaning modes; there is also a lack of a dynamic balance mechanism among cleaning efficiency, energy consumption constraints, and equipment losses. In the optimization stage of cleaning parameters, grid search or gradient descent methods are mostly used, which are prone to falling into local optimal solutions. Summary of the Invention

[0004] In view of the above existing problems, the present invention is proposed.

[0005] Therefore, the present invention provides a control method for intelligent impurity removal, which solves the problems of failure in multi-objective collaborative optimization of cleaning parameters and poor adaptability to dynamic environments in intelligent impurity removal control methods.

[0006] To solve the above technical problems, the present invention provides the following technical solutions:

[0007] In a first aspect, the present invention provides a control method for intelligent impurity removal, which includes collecting operation data and performing preprocessing to obtain a standardized data set;

[0008] Using time series analysis methods and deep learning models to mine the parameter change laws in the standardized data set, and combining with the Q-learning algorithm to construct an adaptive learning model to output an optimal parameter combination;

[0009] According to the optimal parameter combination, using multi-modal sensors to detect the impurity characteristics, and using a deep learning model to classify and identify the impurity characteristics to generate specific impurity type labels. According to the impurity type, select the most suitable cleaning mode from a multi-modal cleaning mode library;

[0010] Using a genetic algorithm to optimize the specific configuration of the cleaning parameters in the cleaning mode, and transmitting the optimized cleaning parameters to the actuator to start the cleaning program, and continuously monitoring the operation status during the cleaning process to obtain a cleaning effect evaluation report.

[0011] As a preferred solution of the control method for intelligent impurity removal described in the present invention, wherein: the preprocessing includes detecting outliers and removing outliers;

[0012] Perform normalization processing on the preprocessed operation data to generate a normalized data set.

[0013] As a preferred solution of the control method for intelligent impurity removal described in the present invention, wherein: using time series analysis method and deep learning model to mine the parameter change rules in the normalized data set, including the following steps,

[0014] Select fluid pressure, water flow velocity, and the pressure difference across the membrane as input variables, and set the index of the time stamp to construct a time series feature data matrix;

[0015] Input the time series feature data matrix into the LSTM model to generate predicted values of future operating states;

[0016] Compare the predicted values of future operating states with the time series feature data matrix to identify the trend rules of parameter changes in the normalized data set.

[0017] As a preferred solution of the control method for intelligent impurity removal described in the present invention, wherein: combining the Q-learning algorithm to construct an adaptive learning model and output the optimal parameter combination, including the following steps,

[0018] Use the Q-learning algorithm to define the state space, action space, and reward function, and create a Q-table;

[0019] Input the predicted values of future operating states into the state space, and select the optimal action from the action space of the Q-table according to the current actual operating state using the greedy strategy;

[0020] Record the adjusted operating state and reward value according to the optimal action combined with the reward function, update the Q value in the Q-table and perform iterative loops until the Q-table converges, and then obtain the optimal parameter combination.

[0021] As a preferred solution of the control method for intelligent impurity removal described in the present invention, wherein: according to the optimal parameter combination, use a multi-modal sensor to detect the impurity characteristics, and use a deep learning model to classify and identify the impurity characteristics to generate specific impurity type labels, including the following steps,

[0022] Use a mitochondrial diameter analyzer combined with the principle of laser scattering to record the distribution of the scattered light intensity of particles in the water flow on the laser and statistically analyze the size range and distribution of the particles to generate a particle size distribution map;

[0023] Use ultrasonic sensors to emit high-frequency sound waves to the surface of the water filtration membrane, receive reflected echo signals, calculate the thickness of the deposits on the surface of the water filtration membrane, and generate a distribution map of the deposit thickness;

[0024] Use pH sensors and ORP sensors to monitor changes in pH and ORP values ​​in water, determine whether the water contains organic matter and biofilm, and generate a water quality chemical properties report based on the judgment results;

[0025] The particle size distribution map, the attachment thickness distribution map and the water quality chemical characteristics report are combined into a multi-dimensional impurity feature set and converted into an image form to be input into the CNN model;

[0026] The CNN model generates a feature map by sliding the convolution kernel over a multi-dimensional impurity feature set.

[0027] Use the ReLU activation function to remove negative values ​​of the feature map and retain positive values. At the same time, perform the maximum pooling operation to reduce the resolution of the feature map to obtain high-level abstract features.

[0028] The high-level abstract features are passed to the fully connected layer, and the weight matrix is ​​used to map the high-level abstract features to different impurity types and convert them into probability distribution using the Softmax function;

[0029] According to the category corresponding to the highest probability, the specific impurity type label is output.

[0030] As a preferred solution of the intelligent impurity removal control method of the present invention, wherein: according to the impurity type, the most suitable cleaning mode is selected from the multi-cleaning mode library, including the following steps:

[0031] Construct a multivariate cleaning pattern library based on specific impurity type labels;

[0032] Match the most suitable cleaning mode according to the different types of impurities.

[0033] As a preferred solution of the control method of intelligent impurity removal of the present invention, the genetic algorithm is used to optimize the specific configuration of the cleaning parameters in the cleaning mode, the optimized cleaning parameters are transmitted to the actuator to start the cleaning program, the operating status is continuously monitored during the cleaning process, and a cleaning effect evaluation report is obtained, including the following steps:

[0034] Initialize the cleaning parameters of the cleaning mode and combine them into a cleaning parameter set;

[0035] Based on the cleaning parameter set, a genetic algorithm is used to generate M individuals as the initial population;

[0036] Statistically calculate the impurity removal efficiency of the impurity removal equipment, equipment utilization rate, and the impact of the environment on the impurity removal equipment, define the fitness function, and calculate the fitness value of each individual;

[0037] According to the fitness value of each individual, the roulette wheel selection method is used to assign a selection probability to each individual, and a fitness threshold is set. Individuals exceeding the fitness threshold are regarded as individuals with higher fitness and enter the next-generation population by random sampling;

[0038] In the next-generation population, individuals with higher fitness are paired and a crossover point is randomly selected. The parts of the parameters after the crossover point are exchanged and mutation operations are performed simultaneously until convergence is achieved for each individual, and the individual with the highest fitness value is obtained;

[0039] The individual with the highest fitness value is used as the optimized cleaning parameter;

[0040] The optimized cleaning parameter is transmitted to the actuator and the corresponding cleaning mode is started for cleaning. During the cleaning process, the core indicators during the operation of the cleaning equipment are continuously monitored and fine-tuned according to the actual situation;

[0041] After the cleaning is completed, cleaning samples are collected and the cleaning effect is evaluated, and the cleaning effect is integrated into a cleaning effect evaluation report.

[0042] In a second aspect, the present invention provides an intelligent impurity removal control system, including a data acquisition module that acquires operation data and performs preprocessing to obtain a standardized data set;

[0043] A parameter output module that uses time series analysis and a deep learning model to mine the parameter change rules in the standardized data set, and constructs an adaptive learning model in combination with the Q-learning algorithm to output an optimal parameter combination;

[0044] A cleaning selection module that, according to the optimal parameter combination, uses a multi-modal sensor to detect the impurity characteristics, classifies and identifies the impurity characteristics using a deep learning model, generates specific impurity type labels, and selects the most suitable cleaning mode from a multi-source cleaning mode library according to the impurity type;

[0045] A report generation module that uses a genetic algorithm to optimize the specific configuration of the cleaning parameters in the cleaning mode, transmits the optimized cleaning parameters to the actuator to start the cleaning program, and continuously monitors the operation status during the cleaning process to obtain a cleaning effect evaluation report.

[0046] In a third aspect, the present invention provides a computer device, including a memory and a processor, where the memory stores a computer program, and: when the computer program is executed by the processor, any step of the intelligent impurity removal control method described in the first aspect of the present invention is implemented.

[0047] Fourthly, the present invention provides a computer-readable storage medium, on which a computer program is stored, wherein: when the computer program is executed by a processor, any step of the intelligent impurity removal control method described in the first aspect of the present invention is implemented.

[0048] The beneficial effects of the present invention are as follows: The genetic algorithm is used to optimize the specific configuration of the cleaning parameters in the cleaning mode, and the optimized cleaning parameters are transmitted to the actuator to start the cleaning program. During the cleaning process, the operating state is continuously monitored to obtain a cleaning effect evaluation report. By defining a multi-objective fitness function for impurity removal efficiency, equipment utilization rate, and environmental impact degree, the global search ability of the genetic algorithm is utilized to solve the multi-objective conflict of cleaning parameters. On the basis of ensuring the impurity removal effect, the comprehensive optimization of reducing the chemical agent consumption and reducing the loss of the equipment is realized. At the same time, the stability of the cleaning process is maintained through a dynamic fine-tuning mechanism, promoting the development of the cleaning and impurity removal technology towards the direction of green and high efficiency. BRIEF DESCRIPTION OF THE DRAWINGS

[0049] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following will briefly introduce the drawings required for the description of the embodiments. Obviously, the following drawings are only some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other drawings can be obtained based on these drawings.

[0050] Figure 1 It is the multi-stage closed-loop control flow chart in Embodiment 1;

[0051] Figure 2 It is the multi-objective parameter optimization genetic algorithm flow chart in Embodiment 1;

[0052] Figure 3 It is the impurity classification flow chart of multi-modal feature fusion in Embodiment 1. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0053] In order to make the above objects, features, and advantages of the present invention more obvious and understandable, the following will make a detailed description of the specific embodiments of the present invention in conjunction with the drawings of the specification.

[0054] Many specific details are set forth in the following description in order to fully understand the present invention. However, the present invention can also be implemented in other ways different from those described herein. Those skilled in the art can make similar generalizations without departing from the connotation of the present invention. Therefore, the present invention is not limited by the specific embodiments disclosed below.

[0055] Second, the "one embodiment" or "embodiment" referred to herein means a specific feature, structure, or characteristic that may be included in at least one implementation manner of the present invention. The "in one embodiment" that appears in different places in this specification does not necessarily refer to the same embodiment, nor is it an individual or alternative embodiment that is mutually exclusive with other embodiments.

[0056] Embodiment 1, referring to Figures 1 to 3 , which is the first embodiment of the present invention. This embodiment provides a control method for intelligent impurity removal, including the following steps:

[0057] S1. The operating data includes fluid pressure, water flow velocity, water flow temperature, and the pressure difference across the membrane (i.e., the difference in fluid pressure across the water quality filtration membrane);

[0058] Performing preprocessing includes detecting and removing outliers;

[0059] Calculating the mean and standard deviation of the operating data, and using the 3σ principle to consider any data point that deviates from the mean by more than 3 times the standard deviation as an outlier and deleting it;

[0060] Normalizing the preprocessed operating data (using Z-score normalization to subtract the mean of each operating data from the corresponding operating data, and then dividing by the standard deviation of the operating data) to generate a normalized data set.

[0061] S2. Using time series analysis method and deep learning model to mine the parameter change rules in the normalized data set.

[0062] Including the following steps,

[0063] Selecting fluid pressure, water flow velocity, and the pressure difference across the membrane as input variables, and setting the index of the time stamp (converting the input variables to datetime format) to construct a time series feature data matrix;

[0064] Selecting the time window length (such as 5 minutes), segmenting the input variables according to the time window and combining them into feature vectors, and then creating an empty two-dimensional matrix (the number of rows is equal to the number of time windows, and the number of columns is equal to the number of features within each time window), and filling the feature vectors into the corresponding rows of the two-dimensional matrix. For example, the feature vector of the first time window is filled into the first row, and so on;

[0065] Inputting the time series feature data matrix into the LSTM model to generate predicted values of future operating states;

[0066] Divide the time series feature data matrix into a training set (80%) and a test set (20%). Select the mean squared error as the loss function and Adam as the optimizer. Divide the training set into several small batches (such as 32, 64, or 128). Each time, update only once using the data of one small batch. Perform forward propagation on the data after the updated batch to obtain the predicted value of the operating state and compare it with the time series feature data matrix to obtain the specific value of the loss function. According to the specific value of the loss function, use the chain rule to calculate the gradients of the weights and biases of each layer layer by layer from the output layer of the LSTM, and use the Adam optimizer to update the weights and bias terms of the LSTM according to the calculated gradients. Repeat the steps of the loss function, backpropagation, and gradient update until all batches are processed. Use the test set to evaluate the performance of the LSTM by calculating the accuracy, recall, and F1 score of the LSTM. When the performance on the test set no longer improves, obtain the trained LSTM model and generate the predicted value of the future operating state based on the memory pattern learned during the training process;

[0067] Compare the predicted value of the future operating state with the time series feature data matrix (for example, if the predicted value of the future operating state is 1.25 bar and the actual value of the time series feature data matrix is 1.30 bar, then the difference is -0.05 bar) to identify the trend law of parameter changes in the standardized dataset;

[0068] Evaluate the change trend of the parameter over time by fitting a straight line. If the slope is positive, it means the parameter is on an upward trend. If the slope is negative, it means the parameter is on a downward trend (take a practical example. If the predicted value of the future operating state shows that the pressure difference across the membrane continues to rise, and the time series feature data matrix also shows a similar upward trend, then it can be inferred that there is a long-term growth trend).

[0069] S3. Combine the Q-learning algorithm to construct an adaptive learning model and output the optimal parameter combination.

[0070] It includes the following steps,

[0071] Use the Q-learning algorithm to define the state space, action space, and reward function, and create a Q-table (that is, a two-dimensional table. The rows represent states, the columns represent actions, and each cell stores the Q value corresponding to the state-action pair. The Q value is set to zero or a random decimal);

[0072] State space: The fluid pressure, water flow velocity, and pressure difference across the membrane are used as variables. These three variables are discretized into a finite number of states (for example, the water flow velocity is divided into three levels: slow, medium, and fast). Finally, the discretized states of the three variables are combined to form a state space (for example, if the fluid pressure has 3 levels and the water flow velocity has 3 levels, there are a total of 9 possible states).

[0073] Action space: Determine the executable actions (adjust the fluid pressure, change the water flow velocity, start the cleaning program). Similarly, the three actions are discretized into a finite number of actions (for example, the action of adjusting the fluid pressure is divided into increasing, decreasing, or remaining unchanged). All actions are combined to form an action space.

[0074] Input the predicted value of the future operating state into the state space, and based on the current actual operating state, use the ε-greedy strategy to select the optimal action from the action space of the Q-table.

[0075] Use the ε-greedy strategy to set a positive number ε less than 1 to represent the probability of choosing a random action (such as 0.1). The remaining probability of 1 - ε is used to choose the currently known best action. A random number between 0 and 1 is randomly generated. If the random number is less than ε, a random action is selected. Otherwise, according to the current actual operating state, query the Q-table to find the action corresponding to the maximum Q value as the optimal action (if there are multiple actions with the same maximum Q value, randomly select one from them).

[0076] Based on the optimal action, combine the reward function to record the adjusted operating state and reward value, update the Q value in the Q-table, and perform iterative loops until the Q-table converges to obtain the optimal parameter combination.

[0077] Execute the selected optimal action (assuming the selected action is to increase the fluid pressure, then adjust the fluid pressure setting accordingly), and record the adjusted operating state, such as (high fluid pressure, low water flow velocity, small pressure difference across the membrane), and the immediate reward value obtained after performing the action, such as an immediate reward of +10 points.

[0078] After execution, find the corresponding Q value in the Q-table. For example, assume the current state is (medium fluid pressure, medium water flow velocity, small pressure difference across the membrane), and the action performed is to increase the fluid pressure. Then find the Q value of the corresponding state-action pair in the Q-table to update the current state.

[0079] According to the new state, find the maximum Q value of the next state in the Q table, and update the Q value of the current state-action pair based on the learning rate (that is, the weight of controlling new and old information) and the discount factor (that is, the importance of future rewards) (assuming the learning rate is 0.1 and the discount factor is 0.9, the new Q value will be adjusted to a value between the original Q value and the immediate reward plus the future reward).

[0080] S4. Based on the optimal parameter combination, the impurity characteristics are detected using a multimodal sensor, and the impurity characteristics are classified and identified using a deep learning model to generate a specific impurity type label.

[0081] The following steps are included:

[0082] Multimodal sensors include linear particle size analyzers, ultrasonic sensors, pH sensors, and ORP sensors;

[0083] Use a linear particle size analyzer combined with the principle of laser scattering to record the intensity distribution of scattered light from particles in the water flow to the laser, and count the size range and distribution of the particles to generate a particle size distribution map;

[0084] Prepare the water sample to be tested and introduce it into the sample pool of the linear particle size analyzer. Start the linear particle size analyzer to make the laser beam pass through the water sample in the sample pool. The particles in the water sample will scatter the laser to generate scattered light signals. The detector of the linear particle size analyzer will automatically detect the scattered light intensity at different angles and record it.

[0085] According to Mie scattering theory (applicable to spherical particles) or Rayleigh scattering theory (applicable to small particles), the size of the particles is inferred (larger particles will produce stronger scattered light at smaller angles, while smaller particles will produce stronger scattered light at larger angles). According to the inference results, the particles are classified according to size. For example, the particles can be divided into several size intervals (such as 0.1-1 microns, 1-10 microns, 10-100 microns, etc.). For each size interval, the number of particles falling in each interval is counted (i.e., frequency distribution). Finally, Matplotlib is used to plot the frequency distribution and size interval into a bar graph to obtain a particle size distribution graph;

[0086] Use ultrasonic sensors to emit high-frequency sound waves to the surface of the water filtration membrane, receive reflected echo signals, calculate the thickness of the deposits on the surface of the water filtration membrane, and generate a distribution map of the deposit thickness;

[0087] When receiving the reflected echo signal, record the time from the emission of the acoustic wave to each reflection point (the time to the first reflection point (the filter membrane - attachment interface) is a, and the time to the second reflection point (the attachment - air interface) is b), and use the sound speed value (i.e., the known sound speed of the acoustic wave in water) combined with the time difference to determine the propagation speed of the acoustic wave between the water quality filter membrane surface and the attachment layer. Use b / 2 minus a / 2, estimate the thickness of the attachment on the water quality filter membrane surface and multiply it by the propagation speed of the acoustic wave between the water quality filter membrane surface and the attachment layer to obtain the actual thickness of the attachment on the water quality filter membrane surface;

[0088] Use a pH sensor and an ORP sensor to monitor the changes in the pH value and ORP value of the water body, judge whether there are organic substances and biofilms in the water body, and generate a water quality chemical property report according to the judgment result;

[0089] When the pH value remains within the range of 6.5 - 8.5, it indicates that the acidity and alkalinity of the water body are relatively stable (in some scenarios, there are more specific pH value requirements, such as the pH value of a swimming pool usually remains between 7.2 and 7.8). When the pH value shows significant fluctuations, it indicates that there are certain chemical reactions or the influence of pollutants in the water body;

[0090] A relatively high ORP value (200 - 400 mV) usually indicates that the water body has a strong oxidation ability and a small number of organic substances and microorganisms, while a relatively low ORP value (below 200 mV) indicates that there are more organic substances or biofilms in the water body, which will consume oxygen and reduce the oxidation ability of the water body;

[0091] Combine the particle size distribution map, the attachment thickness distribution map, and the water quality chemical property report into a multi - dimensional impurity feature set and convert it into an image form to input into the CNN model;

[0092] In this step, combining multiple data sources and using a deep learning model for analysis can provide higher detection accuracy. Compared with relying only on one type of sensor data or simple statistical analysis methods, it can more accurately identify impurities and pollutants in the water body;

[0093] The CNN model slides and scans on the multi - dimensional impurity feature set through the convolution kernel to generate a feature map;

[0094] Set the moving step size of the convolution kernel, and place the convolution kernel at the upper left corner of the multi-dimensional impurity feature set to start covering a small area (the same size as the convolution kernel). For each value in the small area covered by the convolution kernel (i.e., the particle size or the thickness of the attachment at a specific position in the multi-dimensional impurity feature set), find the corresponding convolution kernel value and pair them, then multiply by the weight at the corresponding position in the convolution kernel (assuming the convolution kernel covers a 3x3 area, the area has 9 values, which are 1, 2, 3, 4, 5, 6, 7, 8, 9 respectively. At the same time, the convolution kernel also has 9 weights, which are 0.1, 0.2, 0.3, 0.4, 0.5, 0.6, 0.7, 0.8, 0.9 respectively. Then multiply each value in the small area covered by the convolution kernel with the convolution kernel weights one by one, and then add them together). According to the set moving step size, move the convolution kernel one small grid to the right or down and repeat the weighted summation process until the convolution kernel traverses the entire multi-dimensional impurity feature set. Each convolution kernel will generate a new value and combine them into a feature map;

[0095] Use the ReLU activation function to remove the negative values of the feature map and retain the positive values, and at the same time perform the max pooling operation to reduce the resolution of the feature map to obtain high-level abstract features;

[0096] Select the size of the pooling window (2x2). Starting from the upper left corner of the feature map, cover a small area of the feature map with the pooling window, and then gradually move the window to the right and down until it covers the entire feature map and find the maximum value of the covered features and record it to form a new and reduced-size feature map (this feature map has a lower spatial resolution but retains important feature information);

[0097] Repeat the max pooling operation to make the feature map more abstract to form high-level abstract features (low-level features (such as edges, textures) will be converted into high-level features (such as shapes, object parts));

[0098] Transfer the high-level abstract features to the fully connected layer, use the weight matrix to map the high-level abstract features to different impurity types and use the Softmax function to convert them into probability distributions;

[0099] Suppose there are three types of impurities, namely impurity 1, impurity 2, and impurity 3. Combine the three impurities into an impurity vector, perform a dot product operation between the impurity vector and the weight matrix in the fully connected layer to obtain the matching scores for the impurity types. Subsequently, use the Softmax function to normalize the matching scores of the three impurity types, calculate the exponential values of each matching score and sum them up. Then divide the exponential value of each score by the sum to obtain the final probability distribution (the probability of impurity 1 is exp(S1)÷(exp(S1)+exp(S2)+exp(S3)) = 0.8, the probability of impurity 2 is exp(S2)÷(exp(S1)+exp(S2)+exp(S3)) = 0.15, and the probability of impurity 3 is exp(S3)÷(exp(S1)+exp(S2)+exp(S3)) = 0.05, where S1 represents the probability distribution of impurity 1, S2 represents the probability distribution of impurity 2, and S3 represents the probability distribution of impurity 3);

[0100] Output the specific impurity type label according to the category corresponding to the highest probability;

[0101] The specific impurity type labels are suspended matter, biofilm, and organic matter;

[0102] The impurity type label obtained in this step does not require manual intervention to interpret the probability distribution output by the model. Instead, it automatically selects the category with the highest probability and outputs the corresponding label, significantly reducing human error and workload.

[0103] S5. Select the most suitable cleaning mode from the multi - element cleaning mode library according to the impurity type.

[0104] It includes the following steps,

[0105] Construct a multi - element cleaning mode library according to the specific impurity type label;

[0106] The multi - element cleaning mode library includes a backwashing mode (reverse - flush the water quality filtration membrane with high - pressure water flow to remove the particulate matter deposited on the membrane surface), a chemical cleaning mode (use chemical reagents such as acids, alkalis, and oxidants to dissolve or remove organic matter or biofilm), and an ultrasonic cleaning mode (utilize the cavitation effect generated by high - frequency vibration to destroy impurities such as inorganic salt scale, colloidal particles, and oil and grease attached to the membrane surface);

[0107] Match the suitable cleaning mode according to the different types of impurities (if there are multiple possible matches, the most suitable one can be selected according to the specific situation);

[0108] For example, for biofilm impurities, the chemical cleaning mode or the ultrasonic cleaning mode can be selected. If the biofilm is relatively stubborn, the ultrasonic cleaning mode is preferred. If the biofilm is relatively light, the chemical cleaning mode can be selected.

[0109] S6. Use a genetic algorithm to optimize the specific configuration of the cleaning parameters in the cleaning mode, transfer the optimized cleaning parameters to the actuator to start the cleaning program, continuously monitor the operating status during the cleaning process, and obtain a cleaning effect evaluation report.

[0110] It includes the following steps:

[0111] Initialize the cleaning parameters of the cleaning mode and combine them into a set of cleaning parameters;

[0112] The cleaning parameters for the backwashing mode are cleaning time (30 minutes), cleaning frequency (24 hours), and water flow pressure (0.5 MPa). The cleaning parameters for the chemical cleaning mode are cleaning agent type (sodium hypochlorite), chemical agent concentration (1%), chemical reaction time (40 - 60 minutes), and temperature (about 30°C). The cleaning parameters for the ultrasonic cleaning mode are ultrasonic frequency (40 kHz), action time (35 - 45 minutes), and temperature (22 - 25°C);

[0113] Based on the set of cleaning parameters, use a genetic algorithm to generate M individuals as the initial population;

[0114] Here, M is determined according to the specific problem requirements and the required population size, and the common value range is between 10 and 100;

[0115] Count the impurity removal efficiency of the impurity removal equipment, equipment utilization rate, and the impact of the environment on the impurity removal equipment, define the fitness function and calculate the fitness value of each individual (allocate different weights to the three parameters of impurity removal efficiency, equipment utilization rate, and the impact of the environment on the impurity removal equipment according to actual needs, multiply and add them to obtain the fitness value of each individual);

[0116] Collect the amount of impurities to be removed by the impurity removal equipment, cleaning duration, actual working time, total available time, wastewater treatment volume, and chemical reagent usage;

[0117] Divide the amount of impurities to be removed by the total amount of impurities and multiply by the cleaning duration to obtain the impurity removal efficiency of the impurity removal equipment; divide the actual working time of the impurity removal equipment by the total available time to obtain the equipment utilization rate; divide the sum of the wastewater treatment volume and chemical reagent usage by the amount of impurities to be removed to obtain the impact of the environment on the impurity removal equipment;

[0118] According to the fitness value of each individual, the roulette wheel selection method is used to assign a selection probability to each individual (assuming there are five individuals, namely individual 1, individual 2, individual 3, individual 4, and individual 5, with respective fitness values of 67.73, 55.40, 89.21, 72.15, and 60.32. The total fitness value is 344.81. Divide the fitness value of each individual by the total fitness value to obtain the selection probability of each individual. For example, the selection probability of individual 1 is 0.1964), and a fitness threshold is set. Individuals exceeding the fitness threshold are regarded as individuals with higher fitness and enter the next-generation population by random sampling;

[0119] The numerical range of the fitness threshold needs to be set according to the problem requirements. A percentile is selected, that is, the top 20% or top 30% of the individuals are regarded as individuals with higher fitness;

[0120] Sort the fitness values of the five individuals from largest to smallest as individual 3 (89.21), individual 4 (72.15), individual 1 (67.73), individual 5 (60.32), individual 2 (55.40). If the fitness threshold is the top 30% of the individuals, then individual 3 and individual 4 are regarded as individuals with high fitness values;

[0121] In the next-generation population, pair the individuals with higher fitness (2 individuals as a pair, called parental individuals) and randomly select a crossover point. Exchange the parts of the parameters after the crossover point and perform mutation operations simultaneously until each individual converges to obtain the individual with the highest fitness value;

[0122] The cleaning parameters of the individual are cleaning time, cleaning frequency, cleaning agent type, cleaning agent concentration, water flow pressure, ultrasonic frequency, and temperature. Randomly select the 3rd parameter (i.e., cleaning agent type) as the crossover point, and exchange the parts of the parameters of the two parental individuals after the crossover point to generate two new offspring individuals (for example, the parameters of parental individual A are [45 minutes, once every 48 hours, sodium hypochlorite, 2%, 5 m / s, 35 kHz, 28°C], and the parameters of parental individual B are [50 minutes, once every 24 hours, sulfuric acid, 1.5%, 6 m / s, 40 kHz, 30°C]. Select the cleaning agent type as the crossover point, then the new offspring individuals after the exchange are offspring individual 1 [45 minutes, once every 48 hours, sulfuric acid, 1.5%, 6 m / s, 40 kHz, 30°C], offspring individual 2 [50 minutes, once every 24 hours, sodium hypochlorite, 1.2%, 5 m / s, 35 kHz, 28°C]);

[0123] Set a relatively small mutation probability (e.g., 0.01). For each newly generated offspring individual, randomly select one or more parameters as mutation points (such as selecting the cleaning time), and correct the parameter values corresponding to the mutation points (for example, assuming the cleaning time of offspring individual 1 is 45 minutes, the mutation operation can adjust it to 46 minutes or 44 minutes).

[0124] Take the individual with the highest fitness value as the optimized cleaning parameters (among the five individuals listed, the fitness value of individual 3 is the highest, which is 89.21. Select individual 3 as the optimal cleaning parameters [60 minutes, once every 72 hours, sodium hypochlorite, 1%, 4 m / s, 30 kHz, 25 °C]);

[0125] Transfer the optimized cleaning parameters to the actuator (i.e., the cleaning equipment control system) and start the corresponding cleaning mode for cleaning. During the cleaning process, continuously monitor the core indicators during the operation of the cleaning equipment and make fine-tuning according to the actual situation;

[0126] The core indicators during the operation of the cleaning equipment include the impurity removal efficiency and water quality parameters (pH value, conductivity, turbidity);

[0127] When the impurity removal efficiency is low, gradually increase the cleaning agent concentration (1.05% or 1.1%) and extend the cleaning time according to the actual situation (extend to 63 minutes, that is, increase by 5% or 66 minutes, increase by 10%); when the water quality parameters exceed the standard range (pH value is 6.5 - 8.5, conductivity is 200 μS / cm, turbidity is below 1 NTU), if the pH value is lower than 6.5, add sodium hydroxide to increase the pH value, if it is higher than 8.5, add sulfuric acid to lower the pH value. When the conductivity exceeds 200 μS / cm, add sodium chloride solution to increase the conductivity to the required range. When the turbidity exceeds 1 NTU, replace the more efficient filter element and increase the filtration times to reduce the turbidity below 1 NTU;

[0128] After cleaning, collect the cleaning samples and evaluate the cleaning effect (such as measuring the impurity content and water quality parameters before and after cleaning. Assuming the turbidity of the water sample before cleaning is 5 NTU and the turbidity of the water sample after cleaning is 1 NTU, then the proportion of impurity removal can be calculated as 5 - 1 divided by 5, which is 80%). Integrate the cleaning effect into a cleaning effect evaluation report;

[0129] The cleaning effect evaluation report includes cleaning parameters (such as water flow pressure, cleaning agent concentration, etc.), changes in water quality parameters, impurity removal ratio, and fine-tuning results;

[0130] The cleaning effect evaluation report in this step, compared with the existing cleaning impurity technologies that only focus on a single index (such as the impurity removal ratio), this report covers multiple aspects such as the impurity removal ratio and water quality parameters, provides a more comprehensive and detailed evaluation, and can help managers make more scientific and reasonable decisions, optimize the cleaning plan, and thus greatly improve the cleaning effect.

[0131] This embodiment also provides a control system for intelligent impurity removal, including: a data acquisition module that acquires operation data and performs preprocessing to obtain a standardized data set;

[0132] a parameter output module that uses time series analysis and deep learning models to mine the parameter change rules in the standardized data set, and combines the Q-learning algorithm to construct an adaptive learning model to output the optimal parameter combination;

[0133] a cleaning selection module that, according to the optimal parameter combination, uses a multi-modal sensor to detect the impurity characteristics, and uses a deep learning model to classify and identify the impurity characteristics, generates specific impurity type labels, and selects the most suitable cleaning mode from the multi-source cleaning mode library according to the impurity type;

[0134] a report generation module that uses a genetic algorithm to optimize the specific configuration of the cleaning parameters in the cleaning mode, transfers the optimized cleaning parameters to the actuator to start the cleaning program, continuously monitors the operation status during the cleaning process, and obtains a cleaning effect evaluation report.

[0135] This embodiment also provides a computer device applicable to the intelligent impurity removal control method, including: a memory and a processor; the memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions to implement the intelligent impurity removal control method proposed in the above embodiment.

[0136] This computer device can be a terminal. This computer device includes a processor, a memory, a communication interface, a display screen, and an input device connected through a system bus. Among them, the processor of this computer device is used to provide computing and control capabilities. The memory of this computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The communication interface of this computer device is used to communicate with an external terminal in a wired or wireless manner, and the wireless manner can be achieved through WIFI, a carrier network, NFC (Near Field Communication), or other technologies. The display screen of this computer device can be a liquid crystal display screen or an electronic ink display screen, and the input device of this computer device can be a touch layer covered on the display screen, or a button, a trackball, or a touchpad set on the computer device housing, or an external keyboard, touchpad, or mouse, etc.

[0137] This embodiment also provides a storage medium, on which a computer program is stored. When the program is executed by a processor, it implements the control method for intelligent impurity removal proposed in the above embodiment; the storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (abbreviated as SRAM), electrically erasable programmable read-only memory (abbreviated as EEPROM), erasable programmable read-only memory (abbreviated as EPROM), programmable read-only memory (abbreviated as PROM), read-only memory (abbreviated as ROM), magnetic memory, flash memory, magnetic disk or optical disc.

[0138] In summary, the present invention: uses a genetic algorithm to optimize the specific configuration of cleaning parameters in the cleaning mode, transfers the optimized cleaning parameters to the actuator to start the cleaning program, continuously monitors the operating state during the cleaning process, and obtains a cleaning effect evaluation report. By defining a multi-objective fitness function for impurity removal efficiency, equipment utilization rate, and environmental impact degree, the global search ability of the genetic algorithm is used to solve the multi-objective conflict of cleaning parameters. On the basis of ensuring the impurity removal effect, the comprehensive optimization of reducing the chemical consumption and reducing the loss of the equipment is realized. At the same time, the stability of the cleaning process is maintained through a dynamic fine-tuning mechanism, promoting the development of the clean impurity removal technology towards the direction of green and high efficiency.

[0139] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit them. Although the present invention has been described in detail with reference to the preferred embodiments, those of ordinary skill in the art should understand that the technical solutions of the present invention can be modified or equivalently replaced without departing from the spirit and scope of the technical solutions of the present invention, and they should all be covered by the scope of the claims of the present invention.

Claims

1. A control method for intelligent impurity removal, characterized in that: include, Collect and preprocess the operating data to obtain a standardized data set; Use time series analysis and deep learning models to mine parameter change patterns in standardized data sets, and build an adaptive learning model in combination with the Q-learning algorithm to output the optimal parameter combination; According to the optimal parameter combination, the impurity characteristics are detected by multimodal sensors, and the impurity characteristics are classified and identified using a deep learning model to generate specific impurity type labels. According to the impurity type, the most suitable cleaning mode is selected from the multi-element cleaning mode library; Genetic algorithms are used to optimize the specific configuration of cleaning parameters in the cleaning mode, and the optimized cleaning parameters are transmitted to the actuator to start the cleaning program. During the cleaning process, the operating status is continuously monitored to obtain a cleaning effect evaluation report.

2. The control method for intelligent impurity removal according to claim 1 is characterized in that: The preprocessing includes detecting outliers and removing outliers; The preprocessed running data is standardized to generate a standardized data set.

3. The control method for intelligent impurity removal according to claim 2 is characterized in that: Using time series analysis and deep learning models to mine parameter change patterns in standardized data sets includes the following steps: Select fluid pressure, water flow velocity, and pressure difference on both sides of the membrane as input variables, set the index of timestamp, and construct the time series feature data matrix; Input the time series feature data matrix into the LSTM model to generate the predicted value of future operating status; The predicted values ​​of future operating status are compared with the time series feature data matrix to identify the trend of parameter changes in the standardized data set.

4. The control method for intelligent impurity removal according to claim 3 is characterized in that: Combine the Q-learning algorithm to build an adaptive learning model and output the optimal parameter combination, including the following steps: Use the Q-learning algorithm to define the state space, action space, and reward function, and create a Q table; The predicted value of the future running state is input into the state space, and according to the current actual running state, the greedy strategy is used to select the optimal action from the action space of the Q table; According to the optimal action combined with the reward function, the adjusted running status and reward value are recorded, the Q value in the Q table is updated and iterated repeatedly until the Q table converges to obtain the optimal parameter combination.

5. The control method for intelligent impurity removal according to claim 4 is characterized in that: According to the optimal parameter combination, the impurity characteristics are detected by multimodal sensors, and the impurity characteristics are classified and identified using a deep learning model to generate specific impurity type labels, including the following steps: Use a linear particle size analyzer combined with the principle of laser scattering to record the intensity distribution of scattered light from particles in the water flow to the laser, and count the size range and distribution of the particles to generate a particle size distribution map; Use ultrasonic sensors to emit high-frequency sound waves to the surface of the water filtration membrane, receive reflected echo signals, calculate the thickness of the deposits on the surface of the water filtration membrane, and generate a distribution map of the deposit thickness; Use pH sensors and ORP sensors to monitor changes in pH and ORP values ​​in water, determine whether the water contains organic matter and biofilm, and generate a water quality chemical properties report based on the judgment results; The particle size distribution map, the attachment thickness distribution map and the water quality chemical characteristics report are combined into a multi-dimensional impurity feature set and converted into an image form to be input into the CNN model; The CNN model generates a feature map by sliding the convolution kernel over a multi-dimensional impurity feature set. Use the ReLU activation function to remove negative values ​​of the feature map and retain positive values. At the same time, perform the maximum pooling operation to reduce the resolution of the feature map to obtain high-level abstract features. The high-level abstract features are passed to the fully connected layer, and the weight matrix is ​​used to map the high-level abstract features to different impurity types and convert them into probability distribution using the Softmax function; According to the category corresponding to the highest probability, the specific impurity type label is output.

6. The intelligent impurity removal control method according to claim 5 is characterized in that: According to the impurity type, the most suitable cleaning mode is selected from the multi-cleaning mode library, including the following steps: Construct a multivariate cleaning pattern library based on specific impurity type labels; Match the appropriate cleaning mode according to the different types of impurities.

7. The control method for intelligent impurity removal according to claim 6 is characterized in that: The specific configuration of cleaning parameters in the cleaning mode is optimized using a genetic algorithm, the optimized cleaning parameters are transmitted to the actuator to start the cleaning program, the operating status is continuously monitored during the cleaning process, and a cleaning effect evaluation report is obtained, including the following steps: Initialize the cleaning parameters of the cleaning mode and combine them into a cleaning parameter set; Based on the cleaning parameter set, a genetic algorithm is used to generate M individuals as the initial population; Statistically calculate the impurity removal efficiency of the impurity removal equipment, equipment utilization rate, and the impact of the environment on the impurity removal equipment, define the fitness function, and calculate the fitness value of each individual; According to the fitness value of each individual, a selection probability is assigned to each individual using the roulette wheel selection method, and a fitness threshold is set. Individuals exceeding the fitness threshold are regarded as individuals with higher fitness and enter the next generation population by random sampling; In the next generation population, individuals with higher fitness are paired and a crossover point is randomly selected. Some parameters after the crossover point are exchanged and mutation operations are performed at the same time until each individual reaches convergence, and the individual with the highest fitness value is obtained; The individual with the highest fitness value is used as the optimized cleaning parameter; The optimized cleaning parameters are transmitted to the actuator and the corresponding cleaning mode is started for cleaning. During the cleaning process, the core indicators of the cleaning equipment are continuously monitored and fine-tuned according to the actual situation. After cleaning is completed, cleaning samples are collected and the cleaning effect is evaluated, and the cleaning effect is integrated into a cleaning effect evaluation report.

8. A control system for intelligent impurity removal, based on the control method for intelligent impurity removal according to any one of claims 1 to 7, characterized in that: include, Data acquisition module, which collects and preprocesses the operation data to obtain a standardized data set; The parameter output module uses time series analysis and deep learning models to mine the parameter change patterns in the standardized data set, and combines the Q-learning algorithm to build an adaptive learning model to output the optimal parameter combination; The cleaning selection module uses a multimodal sensor to detect impurity characteristics based on the optimal parameter combination, and uses a deep learning model to classify and identify the impurity characteristics, generate specific impurity type labels, and select the most suitable cleaning mode from the multi-cleaning mode library based on the impurity type; The report generation module uses genetic algorithms to optimize the specific configuration of cleaning parameters in the cleaning mode, transmits the optimized cleaning parameters to the actuator to start the cleaning program, continuously monitors the operating status during the cleaning process, and obtains a cleaning effect evaluation report.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the intelligent impurity removal control method according to any one of claims 1 to 7 are implemented.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the intelligent impurity removal control method according to any one of claims 1 to 7 are implemented.

Images