Automatic dosing control method and device for removing hardness of industrial high-salinity wastewater

Abstract

The invention discloses an automatic dosing control method and device for removing hardness of industrial high-salinity wastewater. Comprising the following steps: collecting raw water quality parameters before medicament feeding, effluent water quality parameters after medicament feeding and medicament feeding amount as historical operation data; performing data cleaning and data standardization processing on the historical operation data to obtain a standard data set; training and testing the supervised learning model by adopting a standard data set to obtain an effluent quality prediction model; on the basis of the effluent quality prediction model, taking the raw water quality parameter and the medicament dosage as input quantities to obtain an effluent quality parameter prediction value; and based on the raw water quality parameter, the current agent dosage and the effluent quality parameter predicted value, dynamically adjusting the agent dosage by adopting a reinforcement learning model. According to the invention, full-automatic control from water quality monitoring to medicament adding is realized, manual intervention is not needed, the dosage can be automatically adjusted according to real-time water quality data, and the treatment efficiency and precision are remarkably improved.

Description

Technical Field

[0001] The present invention relates to an automatic dosing method for industrial chemicals, and particularly to an automatic dosing control method and device for removing the hardness of industrial high-salt wastewater. Background Art

[0002] When the hardness components in industrial wastewater are too high, it will not only cause equipment scaling and pipeline blockage, but also reduce the utilization efficiency of water resources, and even threaten the environment and human health in severe cases. Traditional industrial wastewater hardness treatment methods include chemical precipitation method, ion exchange softening method, membrane separation method, etc. However, these methods have certain limitations in practical applications. For example, the chemical precipitation method requires precise control of the dosing amount, otherwise it may lead to poor treatment effect or generate a large amount of sludge; although the ion exchange softening method has a simple process and less sludge volume, the treatment cost is relatively high, and the ion exchange resin needs to be regenerated or replaced regularly; the membrane separation method faces problems such as membrane fouling and high energy consumption.

[0003] With the development of technology, some emerging water treatment technologies and equipment have been gradually applied to the treatment of industrial wastewater hardness. Among them, the combined use of high-efficiency sedimentation tank and circulating clarifier has become an effective treatment solution. This combined equipment can remove most of the hardness components in the wastewater through steps such as coagulation, flocculation, and sedimentation. However, in actual operation, this combined equipment still has problems such as difficult control of the dosing amount and the treatment efficiency being affected by the hardness load. In view of the above problems, the automatic dosing system has become a new solution for the treatment of industrial wastewater hardness. The automatic dosing system can automatically adjust the dosing amount according to the real-time change of the wastewater hardness to ensure stable and reliable treatment effect. This device not only improves the treatment efficiency, but also reduces the operation cost, and has important application value.

[0004] At present, some automatic dosing devices have been applied to the treatment of industrial wastewater hardness, but these devices still have some significant problems. First of all, most of the existing automatic dosing devices adopt simple sensors and control algorithms, which are difficult to accurately control the dosing amount, resulting in unstable treatment effect. Secondly, these systems often lack intelligence and adaptability, and cannot automatically adjust the dosing strategy according to different water qualities and treatment requirements. In addition, the existing devices also have problems such as high energy consumption and high maintenance cost, which limit their wide application in the treatment of industrial wastewater hardness. Summary of the Invention

[0005] Object of the Invention: The object of the present invention is to provide an automatic dosing control method for removing the hardness of industrial high-salt wastewater, and solve the problems that the existing automatic dosing system has inaccurate dosing amount control and cannot automatically adjust the dosing strategy according to different water qualities and treatment requirements. Another object of the present invention is to propose an automatic dosing control device for removing the hardness of industrial high-salt wastewater, and solve the problem of how to execute the above automatic dosing control method.

[0006] Technical solution: An automatic dosing control method for removing the hardness of industrial high-salt wastewater according to the present invention includes the following steps:

[0007] Collect the raw water quality parameters before dosing, the effluent water quality parameters after dosing, and the chemical dosage as historical operation data;

[0008] Perform data cleaning and data standardization on the historical operation data to obtain a standard data set;

[0009] Use the standard data set to train and test a supervised learning model to obtain an effluent water quality prediction model;

[0010] Based on the effluent water quality prediction model, using the raw water quality parameters and the chemical dosage as input quantities, obtain the predicted values of the effluent water quality parameters;

[0011] Based on the raw water quality parameters, the current chemical dosage, and the predicted values of the effluent water quality parameters, use a reinforcement learning model to dynamically adjust the chemical dosage.

[0012] Preferably, the raw water quality parameters include the flow rate, total hardness, and pH of the raw water before dosing, the effluent water quality parameters include the flow rate, total hardness, and pH of the effluent after dosing, and the chemicals include sodium hydroxide, sodium carbonate, and hydrochloric acid.

[0013] Preferably, the data cleaning includes: removing outliers, duplicate values, and missing values in the historical operation data, using the box plot method for outlier detection, determining the reasonable range of the data, and identifying the data beyond the range;

[0014] Data standardization includes: performing normalization on the historical operation data, converting the data to the [0, 1] interval, and the standardization formula is as follows:

[0015]

[0016] where x is the original data, x max and x min are the maximum and minimum values of this parameter respectively, and x normal is the standardized data.

[0017] Preferably, the training and testing of the supervised learning model using the standard data set includes:

[0018] Divide the standard data set into a training set and a testing set, use the training set to train the supervised learning model, and evaluate the performance of the supervised learning model on the testing set to ensure the prediction accuracy.

[0019] Preferably, the supervised learning model includes:

[0020] The effluent water quality prediction model includes an input layer, a hidden layer, and an output layer, where:

[0021] Input layer: Receives the raw water quality parameters and the chemical dosage.

[0022] Hidden layer: A multi-layer fully connected layer with the ReLU activation function.

[0023] Output layer: Predicts the effluent water quality parameters, using the mean squared error (MSE) as the loss function. The formula is as follows:

[0024]

[0025] where y i is the true value, and

[0026] is the predicted value.

[0027] Preferably, the dynamic adjustment of the chemical dosage based on the raw water quality parameters, the current chemical dosage, and the predicted effluent water quality parameters using a reinforcement learning model includes:

[0028] State space S: Includes the raw water quality parameters, the current chemical dosage, and the predicted effluent water quality. NaOH 、 and Δu HCl ;

[0029] Reward function R: Defines the reward by measuring the difference between the effluent water quality and the target water quality through the mean squared error (MSE). A penalty term for the chemical dosage can be added to the reward function to control the rationality of the chemical dosage. The reward is defined as:

[0030]

[0031] where y out and y target are the actual effluent water quality and the expected effluent water quality parameters respectively, and λ is a coefficient used to balance the water quality error and the economy of the chemical dosage;

[0032] Policy optimization and training: Use a deep Q-network (DQN) to approximate the Q-value function through a neural network. The Q-value function Q(s,a) represents the maximum expected reward that can be obtained after taking action a in state s. The update formula for the Q-value function is:

[0033] Q(s t ,a t )=Q(s t ,a t )+α(r t +γmax a′ Q(s t+1, a′) - Q(s t , a t ))

[0034] where α is the learning rate, r t is the reward, γ is the discount factor used to control the weight of future rewards, and max a′ Q(s t+1 , a′) is the maximum Q value in the next state. Through training, the Q - value function gradually converges to select the optimal chemical dosing adjustment strategy;

[0035] When the value of the reward function reaches the target or the dosing adjustment strategy is stable, the training stops.

[0036] Based on the above - mentioned method, on the other hand, the present invention discloses an automatic dosing control device for removing the hardness of industrial high - salt wastewater, including:

[0037] A data acquisition module, which is used to collect the raw water quality parameters before dosing, the effluent water quality parameters after dosing, and the chemical dosing amount, and clean and standardize the collected data;

[0038] A central processing module, which is built - in with a supervised learning model and a reinforcement learning model. Based on historical data and real - time input data, it can quickly analyze the water quality change trend, predict the best chemical dosing amount, and has the ability of self - learning. It can continuously optimize the algorithm parameters according to the treatment effect feedback to improve the treatment efficiency and accuracy;

[0039] An automatic chemical dosing unit, which receives the chemical dosing amount output by the central processing module and dynamically adjusts the chemical dosing amount to ensure that the effluent hardness continuously meets the preset standard.

[0040] Beneficial effects: Compared with the prior art, the present invention has the following remarkable advantages:

[0041] 1. The present invention integrates on - line detection instruments, a central processing unit device, supervised learning and reinforcement learning algorithms, realizes full - automatic control from water quality monitoring to chemical dosing, and can automatically adjust the chemical dosing amount according to real - time water quality data without manual intervention, significantly improving the treatment efficiency and accuracy;

[0042] 2. Through the accurate prediction of the model, the device can dynamically adjust the chemical dosing amount according to the actual water quality conditions, avoiding resource waste and environmental pollution caused by excessive dosing in traditional methods, and at the same time can also reduce the treatment cost and improve economic benefits;

[0043] 3. The present invention fully considers the water quality characteristics and treatment requirements of different industrial wastewaters, and can quickly adapt to water quality changes by adjusting the parameters of the neural network algorithm to ensure stable and reliable treatment effects;

[0044] 4. The present invention can collect and analyze a large amount of water quality data. With the continuous accumulation and analysis of data, the neural network algorithm will continuously learn and improve, further enhancing the processing efficiency and accuracy, and realizing the continuous optimization and upgrade of the system. BRIEF DESCRIPTION OF THE DRAWINGS

[0045] Figure 1 is the working flow chart of the present invention;

[0046] Figure 2 is the working flow chart of the central processing module;

[0047] Figure 3 is Figure 2 the working flow chart of the machine learning model in DETAILED DESCRIPTION OF THE EMBODIMENTS

[0048] The technical solution of the present invention will be further described below in conjunction with the drawings.

[0049] An automatic dosing control method for removing the hardness of industrial high-salt wastewater includes the following steps:

[0050] (1) Collect the raw water quality parameters before dosing, the effluent water quality parameters after dosing, and the chemical dosage as historical operation data; as Figure 1 shown, the raw water enters the wastewater regulation tank, mechanical clarifier, and variable pore filter in sequence and then discharges water. Among them, the mechanical clarifier is connected to the chemical dosing metering pump device. A total of two sets of on-line detection instruments are provided. One set of on-line detection instruments is arranged before the inlet of the mechanical clarifier to detect the raw water quality parameters (including flow rate, total hardness, pH) before dosing; the other set of on-line detection instruments is arranged between the mechanical clarifier and the variable pore filter to detect the effluent water quality parameters (including flow rate, total hardness, pH) after dosing. The monitoring and statistical data frequency of the on-line detection instruments is once every 2 hours. The chemicals used in this embodiment include sodium hydroxide, sodium carbonate, and hydrochloric acid.

[0051] (2) Perform data cleaning and data standardization processing on the historical operation data to obtain a standard data set; the raw water and effluent water quality parameters are stored in the database after preprocessing, and the data is transmitted to the central processing module. The data cleaning includes: removing outliers, duplicate values, and missing values in the historical operation data. The outlier detection uses the box plot method to determine the reasonable range of the data and identify the data beyond the range;

[0052] Data standardization includes: performing normalization processing on the historical operation data to convert the data to the [0,1] interval. The standardization formula is as follows:

[0053]

[0054] where x is the original data, x max and xmin are the maximum and minimum values of the parameter, respectively, and x normal is the standardized data.

[0055] (3) The present invention designs a model by combining reinforcement learning with supervised learning, as Figure 3 shown. Among them, supervised learning is used to construct a water quality prediction model, and the reinforcement learning model is used to optimize the chemical dosage to ensure that the final effluent water quality meets the expectations. The supervised learning model is used to predict the effluent water quality under a given chemical dosage, and the network structure includes an input layer, a hidden layer, and an output layer:

[0056] Input layer: Accepts the raw water quality parameters and the chemical dosage;

[0057] Hidden layer: A multi-layer fully connected layer with the activation function ReLU;

[0058] Output layer: Predicts the effluent water quality parameters, and uses the mean squared error MSE as the loss function. The formula is as follows:

[0059]

[0060] where y i is the true value, is the predicted value.

[0061] The standard data set is divided into a training set and a test set according to a ratio of 7:3. The supervised learning model is trained using the training set, and the performance of the supervised learning model on the test set is evaluated to ensure the prediction accuracy. After training and testing, a water quality prediction model for the effluent is obtained.

[0062] (4) Based on the water quality prediction model for the effluent, taking the raw water quality parameters and the chemical dosage as input quantities, the predicted values of the effluent water quality parameters are obtained;

[0063] (5) Based on the raw water quality parameters, the current chemical dosage, and the predicted values of the effluent water quality parameters, the chemical dosage is dynamically adjusted using the reinforcement learning model. The reinforcement learning model dynamically adjusts the dosages of sodium hydroxide, sodium carbonate, and hydrochloric acid based on the current water quality state to reduce the deviation from the target water quality. It mainly consists of the following parts:

[0064] State space S: Includes the raw water quality parameters, the current chemical dosage, and the predicted effluent water quality;

[0065] Action space A: The adjustment value of the chemical dosage, Δu NaOH 、 and Δu HCl ;

[0066] Reward function R: The reward is defined by measuring the difference between the water quality of the effluent and the target water quality through the mean squared error (MSE). A penalty term for the chemical dosage can be added to the reward function to control the rationality of the chemical dosage. The reward is defined as:

[0067]

[0068] where y out and y target are the actual effluent water quality and the expected effluent water quality parameters respectively, and λ is a coefficient used to balance the water quality error and the economy of the chemical dosage;

[0069] Policy optimization and training: Use a deep Q-network (DQN) to approximate the Q-value function through a neural network. The Q-value function Q(s,a) represents the maximum expected reward that can be obtained after taking action a in state s. The update formula for the Q-value function is:

[0070] Q(s t ,a t ) = Q(s t ,a t ) + α(r t + γmax a′ Q(s t+1 ,a′) - Q(s t ,a t ))

[0071] where α is the learning rate, r t is the reward, γ is the discount factor used to control the weight of future rewards, max a′ Q(s t+1 ,a′) is the maximum Q-value in the next state. Through training, the Q-value function gradually converges to select the optimal chemical dosage adjustment strategy;

[0072] Sample the initial state s0, execute action A t , adjust the dosage, use the supervised learning model to predict the effluent water quality Y pred after adjustment, calculate R t according to the reward function, and update the policy network. Stop training when the reward function value reaches the target or the dosage adjustment strategy is stable.

[0073] Based on the above method, the present invention further provides an automatic chemical dosing control device for removing the hardness of industrial high-salt wastewater, including:

[0074] A data acquisition module for collecting the raw water quality parameters before chemical dosing, the effluent water quality parameters after chemical dosing, and the chemical dosage, and cleaning and standardizing the collected data;

[0075] The raw water and effluent data are stored in the database after pre - processing, and the data is transmitted to the central processing module;

[0076] The central processing module, which has a supervised learning model and a reinforcement learning model built - in, quickly analyzes the water quality change trend based on historical data and real - time input data, predicts the optimal chemical dosage, and has the ability of self - learning. It can continuously optimize the algorithm parameters according to the treatment effect feedback, improving the treatment efficiency and accuracy; at the same time, it is responsible for generating control signals according to the prediction results and driving the chemical dosing device.

[0077] The machine - learning model built - in the central processing module quickly analyzes the water quality change trend. After determining the optimal dosages of soda ash, liquid caustic soda, and hydrochloric acid, it directly controls the automatic chemical dosing unit (chemical dosing metering pump device), dynamically adjusts the chemical dosage, and ensures that the hardness and pH value of the effluent from the mechanical clarifier meet the preset standards. This mechanism can quickly adjust the chemical dosage according to the slight changes in water quality data, thus effectively avoiding the problems of poor treatment effect or unnecessary waste of chemicals caused by improper chemical dosage (too much or too little). This optimization not only improves the treatment efficiency but also significantly reduces the operating cost.

[0078] The automatic chemical dosing unit receives the chemical dosage output by the central processing module, dynamically adjusts the chemical dosage, and ensures that the effluent hardness continuously meets the preset standards.

[0079] Model Deployment and Application

[0080] Online prediction: Input the current raw water quality and initial dosage, predict the effluent quality. Then adjust the dosage according to the prediction results.

[0081] Closed - loop control: Combine the reinforcement learning optimization strategy with the automatic chemical dosing device, convert the optimal dosage into a control signal, and drive the chemical dosing device to automatically add chemicals through the control system. The control signals include the on / off signals of the dosing device, dosing rate, etc., to achieve real - time adjustment and feedback control.

[0082] System verification: Conduct system verification in the actual water treatment process. First, connect the system to the water treatment process and set reasonable initial parameters. Then, through a period of operation and monitoring, evaluate the system's performance indicators (such as dosing accuracy, water quality stability, operating cost, etc.). If the system performs well, it can continue to be used; if there are problems, further optimization and improvement are required.

[0083] Such as Figure 2As shown, by real-time detecting different influent water qualities and effluent water quality requirements, calculating them and storing them in the database, and transmitting the data information to the central processing module, a suitable chemical dosing scheme is recommended according to the expected effluent water quality. Subsequently, the effluent water quality data is fed back to the central processor in real time. If the effluent water quality meets the expectation, it becomes a data point under this working condition and is stored in the database. If there is a deviation from the expectation, it is re-calculated by the intelligent chemical dosing control device and then output to the chemical dosing device, finally achieving precise chemical dosing.

Claims

1. An automatic dosing control method for removing the hardness of industrial high-salt wastewater, characterized in that: The steps include: Collect the raw water quality parameters before adding the agent, the effluent water quality parameters after adding the agent, and the agent dosage as historical operation data; Perform data cleaning and data standardization on historical operation data to obtain a standard data set; The effluent water quality prediction model was obtained by training and testing the supervised learning model using a standard data set; Based on the effluent water quality prediction model, the raw water quality parameters and the dosage of the reagent are used as input to obtain the predicted values ​​of the effluent water quality parameters; Based on the raw water quality parameters, the current dosage of the reagent and the predicted values ​​of the effluent water quality parameters, the reinforcement learning model is used to dynamically adjust the dosage of the reagent.

2. The automatic dosing control method for removing hardness of industrial high-salt wastewater according to claim 1 is characterized in that: The raw water quality parameters include the flow rate, total hardness and pH of the raw water before adding the reagent, and the effluent water quality parameters include the flow rate, total hardness and pH of the effluent after adding the reagent. The reagents include sodium hydroxide, sodium carbonate and hydrochloric acid.

3. The automatic dosing control method for removing hardness of industrial high-salt wastewater according to claim 1 is characterized in that: The data cleaning includes: removing outliers, duplicate values ​​and missing values ​​in historical operation data, and the outlier detection uses a box plot method to determine the reasonable range of data and identify data that exceeds the range; Data standardization includes: normalizing historical operation data and converting the data to the [0,1] interval. The standardization formula is as follows: Among them, x is the original data, x max and x min are the maximum and minimum values ​​of the parameter, respectively, normal is the standardized data.

4. The automatic dosing control method for removing hardness of industrial high-salt wastewater according to claim 1 is characterized in that: The use of standard data sets to train and test supervised learning models includes: The standard data set is divided into a training set and a test set. The supervised learning model is trained using the training set, and the performance of the supervised learning model on the test set is evaluated to ensure prediction accuracy.

5. The automatic dosing control method for removing hardness of industrial high-salt wastewater according to claim 1 is characterized in that: The supervised learning model includes: The effluent water quality prediction model includes input layer, hidden layer and output layer, among which: Input layer: accepts raw water quality parameters and dosage of reagents; Hidden layer: multi-layer fully connected layer, activation function is ReLU; Output layer: predict the water quality parameters of the outlet water, using mean square error MSE as the loss function, the formula is as follows: Among them, y i is the true value, is the predicted value.

6. The automatic dosing control method for removing hardness of industrial high-salt wastewater according to claim 1 is characterized in that: The method of dynamically adjusting the dosage of the reagent based on the raw water quality parameters, the current dosage of the reagent and the predicted value of the effluent water quality parameters by using the reinforcement learning model includes: State space S: includes raw water quality parameters, current reagent dosage and predicted effluent quality; Action space A: Adjustment value of the dosage of the agent, Δu NaOH , and Δu HCl ; Reward function R: The reward is defined by measuring the difference between the effluent quality and the target water quality through the mean square error MSE. A penalty term for the dosage of the reagent can be added to the reward function to control the rationality of the dosage of the reagent. The reward is defined as: Among them, y out and target are the actual effluent water quality and the expected effluent water quality parameters, respectively, and λ is the coefficient used to balance the water quality error and the economy of the dosage of the reagent; Strategy optimization and training: Use a deep Q network (DQN) to approximate the Q value function through a neural network. The Q value function Q(s,a) represents the maximum expected reward that can be obtained after taking action a in state s. The update formula of the Q value function is: Q(s t ,a t )=Q(s t ,a t )+α(r t +γmax a′ Q(s t+1 ,a′)-Q(s t ,a t )) Among them, α is the learning rate, r t is the reward, γ is the discount factor used to control the weight of future rewards, max a′ Q(s t+1 ,a′) is the maximum Q value in the next state. Through training, the Q value function gradually converges to select the optimal dosage adjustment strategy; When the reward function value reaches the target or the dosage adjustment strategy is stable, stop training.

7. An automatic dosing control device for removing the hardness of industrial high-salt wastewater, characterized in that: include: The data acquisition module is used to collect the raw water quality parameters before adding the agent, the effluent water quality parameters after adding the agent and the agent dosage, and clean and standardize the collected data; The central processing module has built-in supervised learning models and reinforcement learning models. Based on historical data and real-time input data, it can quickly analyze water quality trends and predict the optimal dosage of reagents. It also has self-learning capabilities and can continuously optimize algorithm parameters based on treatment effect feedback to improve treatment efficiency and accuracy. The automatic dosing unit receives the dosage of the reagent output by the central processing module and dynamically adjusts the dosage of the reagent to ensure that the water hardness continues to meet the preset standard.

Images