An intelligent dosing method and system for sewage treatment based on artificial intelligence

Abstract

The application provides an intelligent dosing method and system for sewage treatment based on artificial intelligence, wherein the method comprises the following steps: collecting sensing data in real time, collecting time characteristics and statistical characteristics of a sampling water section based on dynamic changes of feedforward flow, obtaining sample data and predicted sample data; abstracting a modeling mechanism into four stages, and obtaining four-stage modeling characteristics through processing; taking water quality standard reaching and dosing adjustment time as target variables, obtaining target modeling characteristics through feature processing and screening, modeling in combination with the target variables, obtaining a water quality standard reaching prediction model and a dosing time prediction model, obtaining a dosing adjustment amount model through constraints of minimum combined cost of water quality standard reaching and dosing, inputting the predicted sample data, and outputting dosing adjustment time and target dosing adjustment amount to control dosing time and dosing amount of sewage treatment. The application can dynamically sample, realize continuous and accurate control of dosing time and dosing amount, and ensure water quality standard reaching while reducing dosing cost.

Description

Technical Field

[0001] This invention relates to the field of wastewater treatment technology, and in particular to an intelligent dosing method and system for wastewater treatment based on artificial intelligence, as well as electronic equipment and computer-readable storage media. Background Technology

[0002] In existing technologies, intelligent dosing modeling typically employs two methods. The first is an empirical method based on chemical mechanisms. This method determines the mathematical expression for reagent dosing based on chemical reactions and on-site experience, where the variables and interacting parameters are predetermined. This method relies heavily on expert experience, making it difficult to flexibly handle changing situations in practice. Furthermore, in reality, overdosing often occurs to ensure effluent quality meets standards. The second method is a machine learning-based modeling method. This method models data related to wastewater treatment dosing processes, identifying the relationship between reagent dosage and water quality changes. Based on this relationship, a feedforward water quality approach combined with the target water quality is used to determine a reasonable dosage. This method does not rely heavily on expert experience, learning patterns based on the actual conditions of the wastewater treatment process. It is relatively flexible in practice and can adapt to wastewater treatment situations with varying influent.

[0003] However, the effectiveness of machine learning methods is usually closely related to the quantity and quality of the sample data required for the dosing model. Currently, the acquisition of samples for intelligent dosing modeling includes manual sampling and IoT sampling. Manual sampling involves sampling at fixed points at fixed times based on wastewater treatment experience. This method heavily relies on experience and manpower, and due to the relative shortage of manpower in wastewater treatment plants, the sampling efficiency is often low, sometimes taking 1-2 days to generate a single sample. The limited number of samples often leads to model overfitting and a lack of generalization ability, or results in a very long modeling cycle. IoT sampling, on the other hand, involves IoT devices collecting relevant data. The sampling frequency is relatively high and timely, ensuring rapid acquisition of sample data.

[0004] While IoT sampling can solve the problem of sample quantity, it cannot completely solve the problem of sample quality. The quality of the sample depends on whether the sampling process can better and more dynamically and continuously characterize the dosing mechanism of wastewater treatment. At present, the sampling process still relies more on experience. Although it can characterize the dosing mechanism of wastewater treatment to a certain extent, it still cannot solve the problem of dynamic and continuous characterization.

[0005] Meanwhile, actual wastewater treatment often involves multiple dosing points, and effluent water quality standards typically cover multiple indicators. However, most current intelligent dosing systems only target a single dosing stage in the wastewater treatment process, such as carbon source addition, flocculation and sedimentation, purification and disinfection. They often affect relatively single water quality indicators and do not fully consider the dynamic and comprehensive impact of different dosing stages on effluent water quality. Furthermore, for different agents that affect the same water quality, there may be multiple dosing combinations to achieve the same water quality target. However, due to different unit costs, most current intelligent dosing systems still do not consider whether the determined dosing combination has the lowest cost among feasible dosing combinations. Therefore, it is difficult to achieve the management goal of simultaneously meeting multiple water quality targets and minimizing chemical consumption costs.

[0006] Furthermore, the core of intelligent dosing control includes not only the control of the dosage, but also the control of the dosing adjustment time. However, most intelligent dosing systems at present have only solved the problem of dosage control to a certain extent, and modeling of dosing adjustment time is relatively rare. In practice, they still rely more on experience.

[0007] Therefore, there is an urgent need for an intelligent dosing method for wastewater treatment that can perform dynamic sampling during the wastewater treatment process, achieve precise control of dosing time and dosage, and reduce dosing costs. Summary of the Invention

[0008] Therefore, it is necessary to provide an intelligent dosing method, system, electronic device, and computer-readable storage medium for wastewater treatment based on artificial intelligence to address the aforementioned technical problems.

[0009] An AI-based intelligent dosing method for wastewater treatment includes the following steps: Real-time acquisition of sensing data from wastewater treatment; using influent flow rates at set time intervals as samples; and based on the dynamic changes in feedforward flow, acquiring the temporal and statistical characteristics of the sampled water section passing through sampling nodes of each process stage to obtain sample data and predicted sample data; abstracting the dosing modeling mechanism into four stages; extracting data features from the sampled data according to the characteristics of the four stages; and performing four-stage preprocessing to obtain four-stage modeling features; using water quality compliance and dosing adjustment time as target variables, and further processing the four-stage modeling features... The process involves feature encoding conversion and important feature screening to obtain target modeling features. Based on these features and target variables, a water quality compliance prediction model and a dosing time prediction model are obtained. The water quality compliance prediction model is then combined with a dosing combination cost minimization constraint to obtain a dosing adjustment amount model. The predicted sample data is input into the three constructed models to obtain the dosing adjustment time and target dosing adjustment amount. Based on the dosing adjustment time and target dosing adjustment amount, an instruction for the target dosing adjustment amount is issued at the dosing adjustment time to control the dosing amount at each dosing point during the wastewater treatment process.

[0010] In one embodiment, the real-time acquisition of wastewater treatment sensing data uses influent flow rates at set time intervals as samples, and based on the dynamic changes in feedforward flow, collects the temporal and statistical characteristics of the sampled water segment passing through each process segment sampling node to obtain sample data and predicted sample data. This includes: setting up multiple sampling nodes according to the water flow direction; the sampling nodes include feedforward flow rate points, feedforward water quality points, dosing points, and feedback water quality points; and collecting real-time sensing data from IoT devices at the multiple sampling nodes through a data gateway. The sensing data includes influent flow rate, water quality information, and dosing information, and is stored together with data collection point information and data collection time. Based on the collected sensing data... The cumulative influent flow rate at a set time interval is used as a sample. Based on the dynamic changes in the feedforward flow rate, the time characteristics of the sampled water segment passing through the multiple sampling nodes and the statistical characteristics of all data points are collected sequentially until all sampling nodes have finished sampling, thus obtaining sample data. The time characteristics include the start time of the previous neighboring sampling node, the time interval between the start of sampling of two adjacent sampling nodes, and the time interval required for the sampled water segment to pass through the sampling node. The statistical characteristics include the cumulative value, mean, variance, minimum value, and maximum value during the period. The sampling ends at the sampling node before the sensing data reaches the dosing prediction point, thus obtaining the predicted sample data for the corresponding dosing point.

[0011] In one embodiment, the dosing modeling mechanism is abstracted into four stages. Based on the characteristics of these four stages, data features are extracted from the sampled data, and four-stage preprocessing is performed to obtain four-stage modeling features. These features include: abstracting the dosing modeling mechanism into four stages, including water segment generation, dosing feedforward information, dosing status information, and dosing feedback information. The dosing feedforward information includes target water quality information, and the dosing feedback information mainly consists of target water quality information. Time features are extracted from the sample data of each of the four stages to obtain periodic time features. The cumulative dosing amount of the sampled water segment passing through the dosing point is divided by the corresponding time interval to obtain the dosing amount of the sampled water segment at the dosing point. The dosing adjustment characteristics of the dosing points are described; when multiple agents, i.e., multiple dosing points, are applied to the same water quality, the dosing feedforward information includes the feature information of all sampling nodes before the dosing prediction point, and the dosing status information includes the dosing status information of the current dosing prediction point and all subsequent dosing points, thus obtaining four-stage modeling features; when the same agent is applied to different water qualities, the status information of the current dosing prediction point is added to the four-stage method for different target water qualities, thus obtaining four-stage modeling features for different target water qualities; in the dosing time prediction model, the time features and statistical features of all sampling nodes before the dosing prediction point are extracted to obtain four-stage modeling features.

[0012] In one embodiment, the step of using water quality compliance and chemical dosing adjustment time as target variables, and performing feature encoding transformation and important feature screening on the four-stage modeling features to obtain target modeling features includes: encoding transformation of the four-stage modeling features, including encoding periodic time features, encoding target water quality compliance, and standardizing feature values; randomly sampling the encoded features and sorting them from high to low importance based on the ensemble tree method; calculating the correlation coefficients between all features and the target variables using statistical correlation calculation methods, and screening modeling features according to the pre-set number of important features and correlation thresholds to obtain target modeling features.

[0013] In one embodiment, the process of modeling based on the target modeling features and target variables to obtain a water quality compliance prediction model and a dosing time prediction model, and then combining the water quality compliance constraint with the dosing combination cost minimization constraint to obtain a dosing adjustment amount model, and inputting the prediction sample data into the three constructed models to obtain the dosing adjustment time and target dosing adjustment amount, includes: using the target modeling features as input features and the water quality compliance code as output features, training a neural network model to obtain the water quality compliance prediction model; and using virtual node technology based on feedforward flow sampling, setting the virtual node position of the water flow sampling node in the process flow before the dosing prediction point according to the sampling time interval and the prediction control time interval, acquiring the time features and statistical features of the virtual node and all sampling nodes located before the virtual node, and extracting and preprocessing them to obtain input features, with the sampling starting from the dosing prediction point at the distance from the previous neighboring sampling node. The time interval is used as the output feature, and a neural network model is used for training to obtain the dosing time prediction model. A learning-based PTO optimization algorithm is used, with the water quality compliance prediction result of the water quality compliance prediction model as the first constraint, and the dosage adjustment amount of the dosing prediction point and all subsequent dosing points as the decision variable. The unit variable cost per unit time of the dosage adjustment amount is statistically calculated based on historical operating data. The total dosing cost per unit time is obtained by weighted summation of all dosage adjustments based on the unit variable cost, and the minimization of the dosing combination cost per unit time is used as the second constraint. A population-based metaheuristic optimization algorithm is used to obtain the target dosing adjustment amount that satisfies the first and second constraints, thus completing the construction of the dosing adjustment amount model. The predicted sample data is input into the dosing time prediction model to obtain the dosing adjustment time, and the predicted sample data is input into the water quality compliance prediction model and the dosing adjustment amount model to obtain the target dosing adjustment amount.

[0014] In one embodiment, the method further includes: when the same agent is applied to different water qualities, according to the water quality compliance prediction model and the dosing adjustment model, obtaining multiple optimal dosing adjustments for the same agent under different water qualities, comparing the maximum value of the multiple optimal dosing adjustments, and using it as the output feature of the final dosing adjustment for the same agent, thereby obtaining a multi-objective water quality dosing combination optimization model.

[0015] An AI-based intelligent dosing system for wastewater treatment, used to implement the AI-based intelligent dosing method for wastewater treatment as described above, includes: a sample data acquisition module for real-time acquisition of sensing data from wastewater treatment, using influent flow rates at set time intervals as samples, and based on the dynamic changes in feedforward flow, acquiring the temporal and statistical characteristics of the sampled water section passing through each process section sampling node to obtain sample data and predicted sample data; a data feature extraction module for abstracting the dosing modeling mechanism into four stages, extracting data features from the sampled data according to the characteristics of the four stages, and performing four-stage preprocessing to obtain four-stage modeling features; and a modeling feature acquisition module for determining water quality compliance and dosing adjustment time. The target modeling features are obtained by performing feature encoding transformation and important feature screening on the four-stage modeling features. The dosing model construction module is used to build models based on the target modeling features and target variables to obtain a water quality compliance prediction model and a dosing time prediction model. Based on the water quality compliance prediction model, a dosing adjustment amount model is obtained by combining water quality compliance constraints with the minimum cost constraint of dosing combination. The prediction sample data is input into the three constructed models to obtain the dosing adjustment time and the target dosing adjustment amount. The dosing control module is used to issue an instruction for the target dosing adjustment amount at the dosing adjustment time according to the dosing adjustment time and the target dosing adjustment amount, thereby controlling the dosing amount at each dosing point during the wastewater treatment process.

[0016] An electronic device includes a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the program, it implements the steps of the artificial intelligence-based intelligent dosing method for wastewater treatment described in the above embodiments.

[0017] A computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the artificial intelligence-based intelligent dosing method for wastewater treatment described in the above embodiments.

[0018] Compared to existing technologies, the advantages and beneficial effects of this invention are as follows: By collecting real-time sensing data from wastewater treatment, using influent flow rates at set time intervals as samples, and based on the dynamic changes in feedforward flow, the invention collects the temporal and statistical characteristics of the sampled water section passing through each process section sampling node, obtaining sample data and predicted sample data. This allows for the collection of dynamic characterization information of water quality points and dosing points based on dynamic changes in influent, achieving dynamic sampling and continuous prediction of samples. The dosing modeling mechanism is abstracted into four stages. Based on the characteristics of these four stages, data features are extracted from the sampled data, and four-stage preprocessing is performed to obtain four-stage modeling features, enabling standardized processing of the mechanism modeling process. Water quality compliance and dosing adjustment time are used as target variables. The system performs feature encoding transformation and important feature screening on the four-stage modeling features to obtain target modeling features, enabling targeted modeling of intelligent dosing. Based on the target modeling features and target variables, a water quality compliance prediction model and a dosing time prediction model are obtained. The water quality compliance prediction model is then combined with a dosing combination cost minimization constraint to obtain a dosing adjustment model. Predictive sample data is input into the three constructed models to obtain the dosing adjustment time and target dosing adjustment amount. Based on the dosing adjustment time and target dosing adjustment amount, the dosing time and amount in the wastewater treatment process are controlled, achieving precise control of the dosing amount and time, and reducing dosing costs while ensuring water quality compliance. Attached Figure Description

[0019] Figure 1 This is a schematic diagram of a wastewater treatment intelligent dosing method based on artificial intelligence in one embodiment;

[0020] Figure 2 This is a block diagram of an artificial intelligence-based intelligent chemical dosing method for wastewater treatment in one embodiment;

[0021] Figure 3 This is a schematic diagram of the structure of an artificial intelligence-based intelligent dosing system for wastewater treatment in one embodiment;

[0022] Figure 4 This is a schematic diagram of the internal structure of the device in one embodiment. Detailed Implementation

[0023] Before describing the specific embodiments of the present invention, the overall concept of the present invention will be explained as follows:

[0024] This invention is mainly based on the development of wastewater treatment processes. Currently, wastewater treatment cannot achieve dynamic and continuous sampling of the mechanism of action, and there is no solution for controlling the dosing time. Therefore, it is impossible to achieve the dual objectives of achieving water quality standards and reducing costs in wastewater treatment dosing control.

[0025] Therefore, this invention proposes an intelligent dosing method for wastewater treatment based on artificial intelligence. By collecting real-time sensing data from wastewater treatment, using influent flow rates at set time intervals as samples, and based on the dynamic changes in feedforward flow, it collects the temporal and statistical characteristics of the sampled water section passing through each process stage sampling node, obtaining sample data and predicted sample data. This method can collect dynamic characterization information of water quality points and dosing points based on dynamic changes in influent, achieving dynamic sampling and continuous prediction of samples. The dosing modeling mechanism is abstracted into four stages. Based on the characteristics of these four stages, data features are extracted from the sampled data, and four-stage preprocessing is performed to obtain four-stage modeling features, enabling standardization of the mechanism modeling process. Water quality compliance and dosing adjustment time are used as target variables, and the four-stage modeling features are further characterized. Encoding conversion and key feature screening yield target modeling features, enabling targeted modeling of intelligent dosing. Based on these features and target variables, models are constructed: a water quality compliance prediction model and a dosing time prediction model. The water quality compliance prediction model is then combined with a dosing combination cost minimization constraint to obtain a dosing adjustment model. Inputting prediction sample data into the three constructed models yields the dosing adjustment time and target dosing adjustment amount. Based on these adjustments, the dosing time and amount in the wastewater treatment process are controlled, achieving precise control of both dosing quantity and timing. This approach achieves the dual objectives of ensuring water quality compliance and reducing dosing costs, ensuring both water quality compliance and cost reduction in the wastewater treatment process.

[0026] Having introduced the overall concept of the present invention, to make the objectives, technical solutions, and advantages of the present invention clearer, the present invention will be further described in detail below through specific embodiments in conjunction with the accompanying drawings. It should be understood that the specific embodiments described herein are merely illustrative of the present invention and are not intended to limit the present invention.

[0027] In one embodiment, such as Figure 1 As shown, an intelligent chemical dosing method for wastewater treatment based on artificial intelligence is provided, including the following steps:

[0028] Step S101: Real-time acquisition of sensing data for wastewater treatment, using influent flow rate at set time intervals as samples, and based on the dynamic changes of feedforward flow, acquisition of time characteristics and statistical characteristics of the sampled water section passing through sampling nodes of each process section, to obtain sample data and predicted sample data.

[0029] Specifically, by collecting real-time sensing data of wastewater treatment and using influent flow rates at set time intervals as samples, a dynamic sampling method based on feedforward flow solves the problem of dynamically representing water quality points and dosing points based on dynamic changes in influent. Through continuous real-time sampling, the dynamically continuous influent is divided into continuous sampling water segments according to the continuous sampling interval, solving the continuous prediction problem in the practical application of the intelligent dosing model. At the same time, by collecting the temporal and statistical characteristics of the sampling water segments, sample data and predicted sample data are obtained, providing effective temporal and statistical characteristic information for establishing the intelligent dosing model.

[0030] Step S101 includes: setting up multiple sampling nodes according to the water flow direction, including feedforward water flow points, feedforward water quality points, dosing points, and feedback water quality points; collecting real-time sensing data from IoT devices at multiple sampling nodes through a data gateway, including influent flow, water quality information, and dosing information, and storing them together with data collection point information and data collection time; based on the collected sensing data, using the cumulative influent flow at a set time interval as a sample, and based on the dynamic changes in feedforward flow, sequentially collecting the time characteristics of the sampling water segment passing through multiple sampling nodes and the statistical characteristics including all data points, until all sampling nodes have finished sampling, thus obtaining sample data; wherein, the time characteristics include the start sampling time of the previous neighboring sampling node, the time interval between the start sampling of two adjacent sampling nodes, and the time interval required for the sampling water segment to pass through the sampling nodes; the statistical characteristics include the cumulative value, mean, variance, minimum value, and maximum value during the period; the sampling ends when the sensing data reaches the dosing prediction point, thus obtaining the predicted sample data for the corresponding dosing point.

[0031] Specifically, in order to obtain sensing data for wastewater treatment, multiple sampling nodes need to be set up sequentially at reasonable locations in each process section of the wastewater treatment plant according to the sampling order. These nodes include, but are not limited to, feedforward flow points for water section sampling, water quality points for characterizing water quality changes, and dosing points for characterizing dosing status. During data collection, sensing data from IoT devices in the wastewater treatment plant, such as flow meters, water quality measuring instruments, and dosing equipment, are collected through a data gateway. The sensing data includes, but is not limited to, influent flow, water quality information, and dosing information. The real-time data collected by the data gateway is stored, including data collection point information, data collection time, and data index values.

[0032] Based on the collected real-time time-series data, the influent flow rate at a set time interval is used as a sample. The influent flow rate is the cumulative influent volume during the sampling period. Based on the dynamic changes of the feedforward flow rate (including the flow rate changed in the process, such as the booster pump), the time characteristics of the sampling water segment passing through each sampling node and the statistical characteristics of all data points in each sampling node are collected sequentially until all points in all designated sampling nodes are sampled, thus forming a complete sample and obtaining the corresponding sample data.

[0033] The time features include the start time of the previous neighboring sampling node, the time interval between the start of sampling at the current sampling node and the start of sampling at the previous neighboring sampling node, and the time interval between the current sampling water segment passing through the current sampling node; the statistical features include, but are not limited to, the cumulative value, mean, variance, minimum value, and maximum value during the period.

[0034] If the sensing data collection ends only at the sampling node before the dosing point, a prediction sample is formed, and the prediction sample data is obtained. The corresponding dosing adjustment time and the optimal dosing adjustment amount can be obtained through the prediction sample data. If there are multiple dosing nodes in the process, multiple prediction samples can be generated continuously for a sampling water section.

[0035] Step S102: The drug dosing modeling mechanism is abstracted into four stages. Based on the characteristics of the four stages, data features are extracted from the sampled data, and four-stage preprocessing is performed to obtain the four-stage modeling features.

[0036] Specifically, in order to better describe the changes in water quality before and after chemical dosing, the chemical dosing modeling mechanism is abstracted into four stages, and the sampling data of the four stages are subjected to feature extraction and preprocessing to obtain four-stage modeling features. This enables rapid and standardized extraction of feature information, which can be used for complex chemical dosing processes with multiple chemicals and different water qualities.

[0037] Step S102 includes: abstracting the dosing modeling mechanism into four stages, including water segment generation, dosing feedforward information, dosing status information, and dosing feedback information, where both dosing feedforward and dosing feedback information contain the target water quality; extracting time features from the sample data of the four stages to obtain periodic time features, and dividing the cumulative dosing amount of the sampled water segment passing through the dosing point by the corresponding time interval to obtain the dosing adjustment amount feature of the sampled water segment at the dosing point; when multiple agents, i.e., multiple dosing points, act on the same water quality, the dosing feedforward information contains the feature information of all sampling nodes before the dosing prediction point, and the dosing status information contains the dosing status information of the current dosing prediction point and all subsequent dosing points, to obtain the four-stage modeling features; when the same agent acts on different water qualities, the current dosing prediction point status information is added to the four-stage methods for different target water qualities to obtain the four-stage modeling features for different target water qualities; in the dosing time prediction model, extracting the time features and statistical features of all sampling nodes before the dosing prediction point to obtain the four-stage modeling features.

[0038] Specifically, the wastewater treatment dosing mechanism is uniformly and abstractly expressed as a four-stage method. Among them, water segment generation is mainly based on the influent flow rate to generate water segment samples at a set sampling time interval. The dosing feedforward information includes at least the target water quality and may also include other information such as water flow rate. The dosing status information mainly includes dosing adjustment information, and the dosing feedback information mainly includes the target water quality. Water segment generation represents the influent flow rate of the dosing sample, and the other three stages together represent the intrinsic interaction law between the dosing sample and the water quality.

[0039] For situations where multiple agents (i.e., multiple dosing points) act on the same water quality, the dosing feedforward information includes the feature information of all sampling nodes before the current dosing prediction point, while the dosing status information not only includes the dosing status information of the current dosing prediction point, but also includes the dosing status information of all subsequent dosing points acting on the same water quality, resulting in a four-stage modeling feature.

[0040] For situations where the same agent acts on different water qualities, the current dosing prediction point status information can be incorporated into the four-stage method of dosing mechanism for different water qualities, based on the method of multiple agents (i.e. multiple dosing points) acting on the same water quality, to obtain four-stage modeling features.

[0041] For the dosing time prediction model, only the time and statistical features of all sampling nodes before the dosing prediction point are extracted as the four-stage modeling features.

[0042] Step S103: Water quality compliance and chemical dosing adjustment time are taken as target variables, and feature encoding conversion and important feature screening are performed on the four-stage modeling features to obtain target modeling features.

[0043] Specifically, the four-stage modeling features are transformed by feature encoding, such as encoding time period features, encoding target water quality standards, and standardizing feature values. Based on the preset number of important features and correlation thresholds, the important features required for modeling are selected from the four-stage modeling features to obtain the target modeling features.

[0044] Step S103 includes: encoding and converting the four-stage modeling features, including encoding periodic time features, encoding target water quality compliance, and standardizing feature values; randomly sampling the encoded features and sorting them from high to low importance based on the ensemble tree method; calculating the correlation coefficients between all features and the target variable using a statistical correlation calculation method; and filtering the modeling features according to the number of important features and the correlation threshold set in advance to obtain the target modeling features.

[0045] Specifically, after obtaining the four-stage modeling features, they need to be preprocessed. First, the four-stage modeling features are encoded and transformed, including time period features (e.g., month, week, hour, etc.), the encoding of target water quality compliance (e.g., compliance is 1, non-compliance is 0), and the standardization of feature values ​​(e.g., scaling the data to between 0 and 1).

[0046] To avoid the perturbation of the target variable by low-relevance features in the neural network, it is necessary to screen important features. Features after encoding and transformation are randomly sampled, and a random seed is set to ensure the reproducibility of results. An ensemble tree method is used to sort the features from high to low importance. The correlation coefficients between all features and the target variable are calculated through statistical correlation. Feature screening is performed based on a pre-set number of important features (e.g., 50, selecting the top 50 features in the sort) and a correlation threshold (e.g., 0.2, selecting the corresponding feature when the correlation coefficient is not less than 0.2). This yields the target modeling features and also verifies whether there is a statistically significant relationship between the agent and water quality, thereby updating the dosing mechanism.

[0047] Step S104: Model the target based on the target modeling features and target variables to obtain the water quality compliance prediction model and the dosing time prediction model. Then, combine the water quality compliance constraint with the dosing combination cost minimization constraint based on the water quality compliance prediction model to obtain the dosing adjustment amount model. Input the prediction sample data into the three constructed models to obtain the dosing adjustment time and the target dosing adjustment amount.

[0048] Specifically, based on the obtained target modeling features and set target variables, a water quality compliance prediction model and a dosing time prediction model are obtained. The water quality compliance prediction model is then combined with a minimum dosing combination cost constraint. A dosing adjustment model is constructed using PTO (Predict-then-Optimize) technology. Predicted sample data is input into the dosing time prediction model to obtain the dosing adjustment time. This predicted sample data is then input into both the water quality compliance prediction model and the dosing adjustment model to obtain the target dosing adjustment amount that minimizes the dosing combination cost. This target dosing adjustment amount minimizes the dosing combination cost while achieving water quality compliance, and is therefore the optimal dosing adjustment amount. It is applicable to dosing processes for various chemicals and water qualities, achieving the goal of minimizing dosing costs while ensuring water quality compliance, thus reducing overall dosing costs.

[0049] Step S104 includes: using the target modeling features as input features and the water quality compliance code as output features, training a neural network model to obtain a water quality compliance prediction model; based on the virtual node technology of feedforward flow sampling, setting the virtual node position of the water flow sampling node in the process flow before the dosing prediction point according to the sampling time interval and the prediction control time interval; acquiring the time features and statistical features of the virtual node and all previous sampling nodes, extracting and preprocessing them to obtain input features, using the time interval between the start of sampling at the dosing prediction point and the start of sampling at the previous neighboring sampling node as output features, training a neural network model to obtain a dosing time prediction model; and using a learning-based PTO optimization algorithm to optimize the water quality compliance prediction model. The water quality compliance prediction result of the model is used as the first constraint. The dosage adjustment amount of the dosing agent at the prediction point and all subsequent dosing points is used as the decision variable. The unit variable cost of the dosage adjustment amount per unit time is calculated based on historical operating data. The total dosing cost per unit time is obtained by weighted summation of all dosage adjustments based on the unit variable cost. The second constraint is minimizing the combined cost of dosing per unit time. A population-based metaheuristic optimization algorithm is used to obtain the target adjustment dosage that satisfies the first and second constraints, thus completing the construction of the dosing adjustment amount model. The predicted sample data is input into the dosing time prediction model to obtain the dosing adjustment time. The predicted sample data is then input into the water quality compliance prediction model and the dosing adjustment amount model to obtain the target dosing adjustment amount.

[0050] Specifically, the water quality compliance prediction model is used to predict whether the effluent water quality after chemical treatment meets the standards, providing a constraint basis for the chemical dosing adjustment combination optimization model. It takes the water quality target modeling features as input and the water quality compliance code as output, and is trained using a neural network model.

[0051] Leveraging the flexibility of feedforward flow sampling and considering the requirements of dosing time modeling, a virtual node technology based on feedforward flow sampling is adopted. This involves setting a virtual flow sampling node before the dosing prediction point. The specific location is determined based on the sampling time interval and the required interval for predictive control time. For example, if the distance from the dosing prediction point is t and the predictive control time interval is Δt, the virtual sampling node can be set at time t-Δt. The predictive control time interval can be set according to actual conditions, such as the time interval required after the issuance of historical instructions. The time features acquired by this virtual node, along with the time and flow information from all previous sampling nodes, are extracted and preprocessed as model inputs. The time interval between the start of sampling at the dosing prediction point and the start of sampling at the previous neighboring sampling node is used as the output. A neural network model is used for training to obtain the dosing time prediction model. Finally, the predicted dosing time interval is added to the start time of sampling at the previous neighboring sampling node to obtain the final dosing adjustment time. When the dosing time prediction model makes predictions, the dosing adjustment time is determined as t+Δt based on the start time t of sampling at the previous neighboring sampling node and the predictive control time interval Δt, thus achieving the prediction of the dosing adjustment time. By combining feedforward flow and virtual nodes, the problem of large time prediction errors caused by the distance between the dosing prediction point and the previous sampling node due to the long interval between sampling nodes can be avoided, thus improving the accuracy of time prediction. In addition, the virtual node technology is flexible and can be designed on demand according to the actual situation of the wastewater treatment plant and the requirements for the interval of prediction control time, so as to expand the scope of application.

[0052] The dosing adjustment model is used to optimize the dosing adjustment to minimize the total cost of dosing per unit time under the constraint of achieving the target water quality standard. The water quality standard prediction result from the water quality standard prediction model is used as the first constraint. The dosing adjustment amounts of all agents that can affect the same water quality at the predicted dosing point and all subsequent dosing points are used as decision variables. Based on historical operational data, the unit variable cost per unit time for the adjustment amounts of various agents is statistically analyzed. The total cost of dosing per unit time is obtained by weighted summation of the dosing adjustment amounts of all agents. Minimizing the total cost of dosing per unit time is used as the second constraint. Within the feasible region of the dosing adjustment amount, the Particle Swarm Optimization (PSO) algorithm is used to obtain the optimal dosing adjustment combination that simultaneously satisfies the first and second constraints, thus constructing the dosing adjustment model.

[0053] Based on the water quality compliance prediction model, dosing time prediction model, and dosing adjustment model obtained from the above steps, it can be applied to dosing processes with multiple agents and multiple water qualities, and obtain the dosing adjustment amount that minimizes the total dosing cost while ensuring that the water quality meets the standards, thereby achieving control over dosing costs.

[0054] The predicted sample data of the dosing prediction point is input into the dosing time prediction model, and the output is the dosing adjustment time. At the same time, it is input into the water quality compliance prediction model and the dosing adjustment amount model. Through the water quality compliance constraint and the minimum total dosing cost constraint, the target dosing adjustment amount is obtained, so as to control the dosing time and dosing amount in the sewage treatment process and minimize the total dosing cost, thereby realizing intelligent dosing in sewage treatment.

[0055] This also includes: when the same agent is applied to different water qualities, based on the water quality compliance prediction model and the dosing adjustment model, multiple optimal dosing adjustments for the same agent under different water qualities are obtained, and the maximum value of the multiple optimal dosing adjustments is compared and used as the output feature of the final dosing adjustment for the same agent, thus obtaining a multi-objective water quality dosing combination optimization model.

[0056] Specifically, for the case where the same agent affects multiple water quality objectives, the PTO optimization algorithm is used to perform optimization modeling for each water quality, obtaining multiple optimal dosage adjustments for the same agent under different water quality conditions. In order to meet the requirement that all different water quality standards are met, the maximum value of the different optimal dosage adjustments is compared and used as the output of the final dosage adjustment for the same agent, thus obtaining a multi-objective water quality dosing combination optimization model. This achieves optimization under the constraint of meeting multiple water quality standards, enabling the lowest dosing cost while ensuring water quality compliance, thereby achieving cost control of dosing.

[0057] Step S105: Based on the dosing adjustment time and the target dosing adjustment amount, issue an instruction for the target dosing adjustment amount at the dosing adjustment time to control the dosing amount at each dosing point during the wastewater treatment process.

[0058] Specifically, the dosing adjustment time is predicted based on the dosing time prediction model, and the optimal dosing adjustment amount is calculated based on the (multi-objective water quality) dosing adjustment amount model. At the dosing adjustment time, the calculated optimal dosing adjustment amount instruction is sent to the dosing control device, which then controls the corresponding automatic dosing equipment according to the optimal dosing adjustment amount. This achieves precise control of dosing time and amount, and reduces dosing costs while ensuring water quality meets standards.

[0059] In this embodiment, by collecting real-time sensing data of wastewater treatment, using influent flow rates at set time intervals as samples, and based on the dynamic changes in feedforward flow, the temporal and statistical characteristics of the sampled water section passing through each process section sampling node are collected to obtain sample data and predicted sample data. This allows for the collection of dynamic characterization information of water quality points and dosing points based on dynamic changes in influent, achieving dynamic sampling and continuous prediction of samples. The dosing modeling mechanism is abstracted into four stages. Data features are extracted from the sampled data according to the characteristics of these four stages, and four-stage preprocessing is performed to obtain four-stage modeling features, enabling standardization of the mechanism modeling process. Water quality compliance and dosing adjustment time are used as target variables, and the four stages are modeled accordingly. The system performs feature encoding and conversion, and selects important features to obtain target modeling features, enabling targeted modeling of intelligent dosing. Based on the target modeling features and target variables, it generates a water quality compliance prediction model and a dosing time prediction model. Then, based on the water quality compliance prediction model, it combines water quality compliance constraints with a minimum dosing combination cost constraint to obtain a dosing adjustment model. By inputting prediction sample data into the three constructed models, it obtains the dosing adjustment time and the target dosing adjustment amount. Based on the dosing adjustment time and the optimal dosing adjustment amount, it controls the dosing time and amount during the wastewater treatment process, achieving precise control of the dosing amount and time, and reducing dosing costs while ensuring water quality compliance.

[0060] like Figure 3 As shown, an artificial intelligence-based intelligent wastewater treatment dosing system 30 is provided to implement the artificial intelligence-based intelligent wastewater treatment dosing method described above. The system includes: a sample data acquisition module 31, a data feature extraction module 32, a modeling feature acquisition module 33, a dosing model construction module 34, and a dosing control module 35, wherein:

[0061] The sample data acquisition module 31 is used to collect the sensing data of sewage treatment in real time. It uses the influent flow rate at a set time interval as a sample, and based on the dynamic change of the feedforward flow, it collects the time characteristics and statistical characteristics of the sampled water section passing through the sampling nodes of each process section to obtain sample data and predicted sample data.

[0062] The data feature extraction module 32 is used to abstract the drug dosing modeling mechanism into four stages, extract data features from the sampled data according to the characteristics of the four stages, and perform four-stage preprocessing to obtain four-stage modeling features.

[0063] The modeling feature acquisition module 33 is used to take water quality compliance and chemical dosing adjustment time as target variables, and to perform feature encoding conversion and important feature screening on the four-stage modeling features to obtain target modeling features;

[0064] The dosing model construction module 34 is used to build models based on target modeling features and target variables to obtain a water quality compliance prediction model and a dosing time prediction model. Based on the water quality compliance prediction model, the dosing adjustment amount model is obtained by combining water quality compliance constraints with the minimum cost constraint of dosing combination. The dosing adjustment time and the target dosing adjustment amount are obtained by inputting prediction sample data into the three constructed models.

[0065] The dosing control module 35 is used to issue an instruction for the target dosing adjustment amount at the dosing adjustment time according to the dosing adjustment time and the target dosing adjustment amount, so as to control the dosing amount at each dosing point in the sewage treatment process.

[0066] In one embodiment, the sample data acquisition module 31 is specifically used for: setting up multiple sampling nodes according to the water flow direction, the sampling nodes including feedforward water flow points, feedforward water quality points, dosing points, and feedback water quality points; collecting sensing data from IoT devices at multiple sampling nodes in real time through a data gateway, the sensing data including influent flow, water quality information, and dosing information, and storing them together with data collection point information and data collection time; based on the collected sensing data, taking the cumulative influent flow at a set time interval as a sample, and based on the dynamic changes of the feedforward flow, sequentially collecting the time characteristics of the sampling water segment passing through multiple sampling nodes and the statistical characteristics including all data points, until all sampling nodes have finished sampling, thus obtaining sample data; wherein, the time characteristics include the start sampling time of the previous neighboring sampling node, the time interval between the start sampling of two adjacent sampling nodes, and the time interval required for the sampling water segment to pass through the sampling nodes; the statistical characteristics include the cumulative value, mean, variance, minimum value, and maximum value during the period; the sampling node before the sensing data reaches the dosing prediction point ends, thus obtaining the predicted sample data for the corresponding dosing point.

[0067] In one embodiment, the data feature extraction module 32 is specifically used to: abstract the dosing modeling mechanism into four stages, including water segment generation, dosing feedforward information, dosing state information, and dosing feedback information, where both the dosing feedforward and dosing feedback information contain the target water quality; extract time features from the sample data of the four stages to obtain periodic time features, and divide the cumulative dosing amount of the sampled water segment passing through the dosing point by the corresponding time interval to obtain the dosing adjustment amount feature of the sampled water segment at the dosing point; when multiple agents, i.e., multiple dosing points, act on the same water quality, the dosing feedforward information contains the feature information of all sampling nodes before the dosing prediction point, and the dosing state information contains the dosing state information of the current dosing prediction point and all subsequent dosing points, to obtain the four-stage modeling features; when the same agent acts on different water qualities, the current dosing prediction point state information is added to the four-stage methods for different target water qualities to obtain the four-stage modeling features for different target water qualities; in the dosing time prediction model, extract the time features and statistical features of all sampling nodes before the dosing prediction point to obtain the four-stage modeling features.

[0068] In one embodiment, the modeling feature acquisition module 33 is specifically used to: encode and convert the four-stage modeling features, including encoding periodic time features, encoding target water quality compliance, and standardizing feature values; randomly sample the encoded features and sort them from high to low importance based on the ensemble tree method; calculate the correlation coefficients between all features and the target variable using a statistical correlation calculation method, and filter the modeling features according to the number of important features and the correlation threshold set in advance to obtain the target modeling features.

[0069] In one embodiment, the combined model construction module 34 is specifically used for: taking the target modeling features as input features and the water quality compliance code as output features, training a neural network model to obtain a water quality compliance prediction model; based on the virtual node technology of feedforward flow sampling, setting the virtual node position of the water flow sampling node in the process flow before the dosing prediction point according to the sampling time interval and the prediction control time interval; acquiring the time features and statistical features acquired by the virtual node and all previous sampling nodes, and extracting and preprocessing them to obtain input features, taking the time interval between the start of sampling at the dosing prediction point and the start of sampling at the previous neighboring sampling node as output features, training a neural network model to obtain a dosing time prediction model; and using a learning-based PTO optimization algorithm to... The water quality compliance prediction result of the water quality compliance prediction model is used as the first constraint. The dosage adjustment amount of the dosing agent at the dosing prediction point and all subsequent dosing points is used as the decision variable. The unit variable cost of the dosage adjustment amount per unit time is calculated based on historical operating data. The total dosing cost per unit time is obtained by weighted summation of all dosage adjustment amounts based on the unit variable cost. The second constraint is minimizing the combined cost of dosing per unit time. A population-based metaheuristic optimization algorithm is used to obtain the target dosage adjustment amount that satisfies the first and second constraints, thus completing the construction of the dosing adjustment amount model. The predicted sample data is input into the dosing time prediction model to obtain the dosing adjustment time. The predicted sample data is then input into the water quality compliance prediction model and the dosing adjustment amount model to obtain the target dosing adjustment amount.

[0070] In one embodiment, an electronic device is provided, which may be a server, and its internal structure diagram may be as follows: Figure 4As shown, the device includes a processor, memory, network interface, and database connected via a system bus. The processor provides computing and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system, computer programs, and database. The internal memory provides the environment for the operating system and computer programs in the non-volatile storage media to run. The database stores configuration templates and can also be used to store target web page data. The network interface communicates with external terminals via a network connection. When the computer program is executed by the processor, it implements an AI-based intelligent dosing method for wastewater treatment.

[0071] Those skilled in the art will understand that Figure 4 The structure shown is merely a block diagram of a portion of the structure related to the present application and does not constitute a limitation on the device to which the present application is applied. Specific devices may include more or fewer components than those shown in the figure, or may combine certain components, or may have different component arrangements.

[0072] In one embodiment, a computer-readable storage medium may also be provided, the storage medium storing a computer program, the computer program including program instructions, which, when executed by a computer, cause the computer to perform the method described in the foregoing embodiments, the computer being part of the aforementioned AI-based intelligent dosing system for wastewater treatment.

[0073] Those skilled in the art will understand that all or part of the processes in the above embodiments can be implemented by a computer program instructing related hardware. The program can be stored in a computer-readable storage medium, and when executed, it can include the processes of the embodiments of the above methods. The storage medium can be a magnetic disk, optical disk, read-only memory (ROM), or random access memory (RAM), etc.

[0074] It is obvious to those skilled in the art that the modules or steps of the present invention described above can be implemented using general-purpose computing devices. They can be centralized on a single computing device or distributed across a network of multiple computing devices. Optionally, they can be implemented using computer-executable program code, thereby storing them in a computer storage medium (ROM / RAM, magnetic disk, optical disk) for execution by the computing device. In some cases, the steps shown or described can be performed in a different order than those described herein, or they can be fabricated as separate integrated circuit modules, or multiple modules or steps can be fabricated as a single integrated circuit module. Therefore, the present invention is not limited to any particular hardware and software combination.

[0075] The above description, in conjunction with specific embodiments, provides a further detailed explanation of the present invention. It should not be construed that the specific implementation of the present invention is limited to these descriptions. For those skilled in the art, various simple deductions or substitutions can be made without departing from the concept of the present invention, and all such deductions or substitutions should be considered within the scope of protection of the present invention.

Claims

1. An artificial intelligence-based intelligent dosing method for sewage treatment, characterized in that, The method comprises the following steps: Real-time collection of perception data of sewage treatment, taking the influent flow at a set time interval as a sample, and based on the dynamic changes of the feedforward flow, collecting the time characteristics and statistical characteristics of the sampling water section passing through each process section sampling node to obtain sample data and predicted sample data; The dosing modeling mechanism is abstracted into four stages, and the sample data is extracted according to the characteristics of the four stages and preprocessed to obtain four-stage modeling features, including: the dosing modeling mechanism is abstracted into four stages, including water section generation, dosing feedforward information, dosing state information and dosing feedback information, the dosing feedforward information contains target water quality information, and the dosing feedback information is target water quality information; the sample data of the four stages is extracted respectively to obtain periodic time characteristics, and the dosing cumulative amount of the sampling water section passing through the dosing point is divided by the corresponding time interval to obtain the dosing adjustment amount characteristics of the sampling water section at the dosing point; when there are multiple dosing points acting on the same water quality, the dosing feedforward information contains the feature information of all sampling nodes before the dosing prediction point, and the dosing state information contains the dosing state information of the current dosing prediction point and all subsequent dosing points, to obtain four-stage modeling features; when the same dosing agent acts on different water qualities, the state information of the current dosing prediction point is added to the four-stage method of different target water qualities respectively to obtain four-stage modeling features of different target water qualities; in the dosing time prediction model, the time characteristics and statistical characteristics of all sampling nodes before the dosing prediction point are extracted to obtain four-stage modeling features; The water quality compliance and dosing adjustment time are taken as target variables, and the four-stage modeling features are converted and screened to obtain target modeling features; Modeling is performed according to the target modeling features and target variables to obtain a water quality compliance prediction model and a dosing time prediction model, and the water quality compliance prediction model is combined with the minimum cost constraint of dosing combination to obtain a dosing adjustment amount model, and the predicted sample data is input into the three constructed models to obtain the dosing adjustment time and the target dosing adjustment amount; According to the dosing adjustment time and the target dosing adjustment amount, the instruction of the target dosing adjustment amount is issued at the dosing adjustment time to control the dosing amount of each dosing point in the sewage treatment process. 2.The AI-based intelligent dosing method for sewage treatment according to claim 1, wherein, The real-time collection of perception data of sewage treatment, taking the influent flow at a set time interval as a sample, and based on the dynamic changes of the feedforward flow, collecting the time characteristics and statistical characteristics of the sampling water section passing through each process section sampling node to obtain sample data and predicted sample data, comprises: A plurality of sampling nodes are arranged according to the water flow direction, the sampling nodes include feedforward water flow points, feedforward water quality points, dosing points and feedback water quality points, perception data of the Internet of Things equipment at the plurality of sampling nodes are collected in real time through a data gateway, the perception data includes influent flow, water quality information and dosing information, and are stored together with data collection point information and data collection time; Based on the collected perception data, the inflow cumulative flow at a set time interval is taken as a sample, and based on the dynamic changes of the feedforward flow, the time characteristics of the sampling water section passing through the plurality of sampling nodes and the statistical characteristics containing all data points are sequentially collected until all sampling nodes are sampled, to obtain sample data; The time characteristics include the starting sampling time of the previous adjacent sampling node, the time interval between the starting sampling of adjacent two sampling nodes, and the time interval required for the sampling water section to pass through the sampling node; and the statistical characteristics include the cumulative value during the period, the mean value, the variance, the minimum value, and the maximum value. The sampling node before the perception data collection reaches the dosing prediction point is ended, and the prediction sample data corresponding to the dosing point is obtained. 3.The AI-based intelligent dosing method for sewage treatment according to claim 2, wherein, The water quality standard and the dosing adjustment time are taken as target variables, and the four-stage modeling features are subjected to feature encoding conversion and important feature screening to obtain target modeling features, including: The four-stage modeling features are subjected to encoding conversion, including encoding of periodic time characteristics, encoding of target water quality standard, and standardization processing of feature values; The features after the encoding conversion are subjected to random sampling, and based on the integrated tree method, the features are sorted in descending order of importance; The correlation coefficients of all features and target variables are calculated by a statistical correlation calculation method, the important feature quantity and the correlation threshold are pre-set, the modeling features are screened, and the target modeling features are obtained. 4.The AI-based intelligent dosing method for sewage treatment according to claim 3, wherein, The target modeling features and target variables are modeled to obtain a water quality standard prediction model and a dosing time prediction model, and the water quality standard prediction model is used to constrain the water quality standard combined with the minimum cost constraint of dosing to obtain an intelligent dosing combination optimization model, the prediction sample data is input into the three constructed models to obtain the dosing adjustment time and the target dosing adjustment amount, including: The target modeling features are taken as input features, the water quality standard encoding is taken as output features, a neural network model is trained to obtain a water quality standard prediction model; Based on the virtual node technology of the feedforward flow sampling, before the dosing prediction point, the virtual node position of the water flow sampling node in the process flow is set according to the interval between the sampling time interval and the prediction control time, the time characteristics and statistical characteristics of the virtual node and all sampling nodes located before the virtual node are obtained, the input features are extracted and preprocessed, the time interval between the sampling start of the dosing prediction point and the sampling start of the previous adjacent sampling node is taken as the output features, a neural network model is trained to obtain a dosing time prediction model; The learning-based PTO optimization algorithm is adopted, the water quality compliance prediction result of the water quality compliance prediction model is taken as a first constraint, the action medicament adjustment amount in the dosing prediction point and all subsequent dosing points is taken as a decision variable, the unit time unit variation cost of the action medicament adjustment amount is counted according to historical operation data, the unit time total dosing cost is obtained by weighted summation of all action medicament adjustment amounts according to the unit time unit variation cost, and minimization of the unit time total dosing cost is taken as a second constraint, a group-based meta-heuristic optimization algorithm is adopted to obtain the target adjustment dosing amount satisfying the first constraint and the second constraint, and construction of the dosing adjustment amount model is completed. The prediction sample data is input into the dosing time prediction model to obtain a dosing adjustment time, and the prediction sample data is input into the water quality compliance prediction model and the dosing adjustment amount model to obtain a target dosing adjustment amount. 5.The AI-based intelligent dosing method for sewage treatment according to claim 4, wherein, Further comprising: When the same medicament acts on different water qualities, a plurality of optimal dosing adjustment amounts of the same medicament under different water qualities are obtained according to the water quality compliance prediction model and the dosing adjustment amount model, the maximum value of the plurality of optimal dosing adjustment amounts is compared to obtain an output feature of the final dosing adjustment amount of the same medicament, and a multi-target water quality dosing combination optimization model is obtained.

6. An artificial intelligence-based intelligent dosing system for sewage treatment, characterized by, A sewage treatment intelligent dosing method based on artificial intelligence is used to realize the sewage treatment intelligent dosing method based on artificial intelligence, comprising: A sample data acquisition module is configured to collect perception data of sewage treatment in real time, take influent flow at a set time interval as a sample, collect time characteristics and statistical characteristics of a sampling water section passing through each process section sampling node based on dynamic changes of feedforward flow, and obtain sample data and prediction sample data. A data feature extraction module is configured to abstract a dosing modeling mechanism into four stages, extract data features of sample data according to four stage division characteristics, and perform four stage preprocessing to obtain four stage modeling features, including: abstracting a dosing modeling mechanism into four stages, including water section generation, dosing feedforward information, dosing state information, and dosing feedback information, the dosing feedforward information including target water quality information, and the dosing feedback information being target water quality information; extracting time characteristics of sample data of the four stages respectively to obtain periodic time characteristics, and dividing dosing cumulative amount of a sampling water section passing through a dosing point by a corresponding time interval to obtain dosing adjustment amount characteristics of the sampling water section at the dosing point; when multiple medicaments, i.e., multiple dosing points, act on the same water quality, the dosing feedforward information includes feature information of all sampling nodes before a dosing prediction point, and the dosing state information includes dosing state information of the current dosing prediction point and all subsequent dosing points, to obtain four stage modeling features; when the same medicament acts on different water qualities, the current dosing prediction point state information is added to the four stage method of different target water qualities respectively to obtain four stage modeling features of different target water qualities; in the dosing time prediction model, time characteristics and statistical characteristics of all sampling nodes before the dosing prediction point are extracted to obtain four stage modeling features. The modeling feature acquisition module is configured to take water quality compliance and dosing adjustment time as target variables, and perform feature coding conversion and important feature screening on the four-stage modeling features to obtain target modeling features; The dosing model construction module is configured to perform modeling according to the target modeling features and the target variables to obtain a water quality compliance prediction model and a dosing time prediction model, and perform water quality compliance constraint combined with a minimum combination cost constraint according to the water quality compliance prediction model to obtain a dosing adjustment amount model, input the three constructed models with the prediction sample data to obtain the dosing adjustment time and the target dosing adjustment amount; The dosing control module is configured to issue an instruction of the target dosing adjustment amount at the dosing adjustment time according to the dosing adjustment time and the target dosing adjustment amount, and control dosing amounts of each dosing point in the sewage treatment process.

7. An electronic device comprising a memory, a processor, and a computer program stored on the memory and executable on the processor, characterized in that, The processor executes the computer program to implement the steps of the method in any one of claims 1 to 5.

8. A computer-readable storage medium having stored thereon a computer program, characterized in that The computer program is executed by the processor to implement the steps of the method in any one of claims 1 to 5.

Images