Sewage treatment carbon source addition intelligent control method and system based on hybrid model

Abstract

The invention discloses a sewage treatment carbon source addition intelligent control method and system based on a hybrid model. The method comprises the following steps: determining a biochemical reaction mechanism model and an AI model; setting an expected value of the reference quantity; a real-time detection data acquisition step: acquiring real-time detection data of each sensor of the sewage treatment plant; a single model suggested control value acquisition step: respectively acquiring a suggested control value ME of the biochemical reaction mechanism model and a suggested control value MAI of the AI model; obtaining a suggested control value M of the hybrid model; and an equipment control step: taking the suggested control value M as the final carbon source feeding amount, and controlling feeding equipment of the carbon source feeding system to perform carbon source feeding operation according to the suggested control value M. The sewage treatment carbon source adding intelligent control method and system based on the hybrid model have the advantages that the advantages of the mechanism model and the AI model are comprehensively utilized, the optimal control decision is obtained, the working efficiency of sewage treatment is improved, and the operation and maintenance cost is reduced.

Description

Technical Field

[0001] The invention relates to a sewage treatment technology, in particular to a sewage treatment carbon source addition intelligent control method and system based on a hybrid model. Background Art

[0002] With the continuous development of industrialization, the topic of water pollution has been constantly raised. Sewage treatment technology is also developing rapidly. Sewage treatment refers to the process of purifying sewage so that it meets the water quality requirements for discharge into a water body or reuse. Sewage treatment is widely used in various fields such as construction, agriculture, transportation, energy, petrochemicals, environmental protection, urban landscape, medical care, and catering. At present, sewage treatment is an important part of modern urban infrastructure and is of great significance to environmental protection and social sustainable development. However, there are many challenges in the sewage treatment process, especially in ensuring that the effluent water quality meets the discharge standards while achieving effective utilization of resources and energy.

[0003] In recent years, with the development of information technology and automation technology, the operation and management of sewage treatment plants are undergoing profound changes. At present, most sewage treatment plants have built digital platforms.

[0004] In the digital platform, by installing various sensors and building an interactive display interface, the operating status of the sewage treatment plant is displayed in real time on a large screen or mobile phone, providing an important reference for the operation management and decision-making of the sewage treatment process. The digital platform usually includes data collection and display in multiple aspects such as water quality monitoring, flow control, and equipment status monitoring, which greatly improves the information transparency and management efficiency of the sewage treatment plant. Some advanced sewage treatment plants have introduced automatic control systems based on the digital platform. For example, the precise aeration control system can control the aeration according to the difference between the set DO (dissolved oxygen) value and the real-time detected DO data; the automatic dosing system can change the intermittent batch addition during manual operation into continuous and uniform addition according to the setting, thereby achieving more precise process control. Although the introduction of these automatic control methods can significantly reduce labor costs and improve the stability of sewage treatment to a certain extent, they still have great limitations: it is difficult to adapt to changes in environmental conditions and sudden water quality fluctuations or load shocks, and they need to rely on the experience of process operators. In order to strengthen energy conservation and emission reduction to achieve low-carbon development and adapt to the requirements of ecological civilization construction, further intelligent control technologies are urgently needed to help the operation and management of sewage treatment.

[0005] At present, the operation and control technologies in the field of sewage treatment can be divided into two categories: technologies based on mechanism models and technologies based on AI (Artificial Intelligence) models. Among them, the technology based on mechanism models is relatively mature and is widely used in manual operations and automation systems in current sewage treatment plants, but its adaptability is still limited. The biochemical reactions in the sewage treatment process are extremely complex. The current understanding of the mechanism of the sewage treatment process is not comprehensive and complete, and the existing mechanism models are based on the chemical and biological reactions of the known parts and several assumptions, and the model parameters will change dynamically with the environment and process. Therefore, although the mechanism model can provide a clear understanding of the process, when faced with complex actual working conditions, the mechanism model still relies heavily on the experience of the operator and has a large margin from the optimal decision. The technology based on the AI ​​model is a technology that has been explored and developed in recent years. Its advantage is that by learning a large amount of historical data, it can discover the implicit patterns and laws, establish complex nonlinear relationships, and thus improve the decision support capability. The disadvantage of the AI ​​model is that it lacks an explanation of the process and requires a large amount of training data. In addition, model overfitting caused by data distribution limitations may affect the safe and stable operation of sewage treatment when encountering working conditions that have not been learned. Summary of the invention

[0006] The present invention is to avoid the shortcomings of the above-mentioned existing technologies and provide an intelligent control method and system for carbon source addition in sewage treatment based on a hybrid model, so as to comprehensively utilize the advantages of the mechanism model and the AI ​​model, obtain the optimal control decision, improve the work efficiency of sewage treatment and reduce the operation and maintenance costs.

[0007] The present invention adopts the following technical solutions to solve the technical problems.

[0008] The present invention discloses a method for intelligently controlling carbon source addition in sewage treatment based on a hybrid model, comprising the following steps: Step 1: Determination of the biochemical reaction mechanism model and the AI ​​model; Determine the biochemical reaction mechanism model of the sewage treatment plant; Obtain the historical sensor detection data and historical operation data of the sewage treatment plant, and build an AI model for estimating the carbon source dosage based on the sensor detection historical data and historical operation data and train the AI ​​model; Step 2: setting the expected value of the reference amount; at a designated location of the sewage treatment plant, determining the water quality index of the location, and determining the expected value of the reference amount of the water quality index; Step 3: Real-time detection data acquisition step; obtaining real-time detection data of each sensor of the sewage treatment plant; Step 4: Obtaining the recommended control value of a single model; Obtain the recommended control value M of the biochemical reaction mechanism model respectivelyE and the recommended control value M of the AI ​​model AI ; Step 5: Obtain the suggested control value M of the mixed model; the suggested control value M of the biochemical reaction mechanism model E and the recommended control value M of the AI ​​model AI Perform fusion to obtain a recommended control value M of the fused hybrid model; Step 6: Equipment control step; taking the recommended control value M as the final carbon source dosage, controlling the dosing equipment of the carbon source dosing system to perform the carbon source dosing operation according to the recommended control value M.

[0009] The characteristics of the agent dosing control method of a sewage treatment process of the present invention are also: Furthermore, in step 1, the biochemical reaction mechanism model is an activated sludge model.

[0010] Furthermore, in step 1, the AI ​​model is any one of a machine learning or deep learning model established based on SVM, decision tree, LSTM, GCN or Transformer.

[0011] Furthermore, in step 5, the recommended control value M E and the recommended control value M AI The fusion formula is: M=w E *M E +w AI *M AI (1) In formula (1), w E The suggested control value M for the biochemical reaction mechanism model E The weight value, w AI is the recommended control value M for the AI ​​model AI The weight value of ; M is the recommended control value of the mixed model.

[0012] Furthermore, the weight value w E and weight value w AI dynamic adjustment process.

[0013] Furthermore, in the dynamic adjustment process, the sensitivity S is used to adjust the weight value w E and weight value w AI Perform dynamic adjustment, where 0<S≤1.

[0014] Furthermore, the weight adjustment amount Δw is calculated using the following formula (3); Δw=S*(1 - exp(-(NT) 2 )) (3); In the formula (3), N is the real-time detection value of the reference quantity, and T is the expected value of the reference quantity.

[0015] Furthermore, when the real-time detection value N of the reference quantity is greater than the expected value T of the reference quantity, the weight w of the mechanism model is increased. E , reduce the weight w of the AI ​​model AI ; When the real-time detection value N of the reference quantity is less than the expected value T of the reference quantity, the weight w of the AI ​​model is increased AI , reduce the weight w of the mechanism model E .

[0016] The present invention also discloses a sewage treatment carbon source dosing intelligent control system based on a hybrid model, comprising: A model determination unit, used to determine the biochemical reaction mechanism model of the sewage treatment plant, obtain the historical sensor detection data and historical operation data of the sewage treatment plant, and build an AI model for estimating the carbon source dosage based on the historical sensor detection data and historical operation data and train the AI ​​model; Data input unit: used to set the expected value T of the reference quantity in the sewage treatment process; The sensor unit is used to obtain real-time detection data from sensors at various points in the sewage treatment plant; The recommended control value calculation unit is used to calculate the recommended control value M of the biochemical reaction mechanism model. E and the recommended control value M of the AI ​​model AI ; A data fusion unit for suggesting control values ​​M for the biochemical reaction mechanism model E and the recommended control value M of the AI ​​model AI Perform fusion to obtain a recommended control value M of the fused hybrid model; The control unit is used to use the recommended control value M as the final carbon source dosage, and control the dosing equipment of the carbon source dosing system to perform the carbon source dosing operation according to the recommended control value M.

[0017] Furthermore, the sewage treatment carbon source addition intelligent control system based on the hybrid model also includes an intelligent control engine for calculating the weight w of the physical model. E and the weight w of the AI ​​model AI .

[0018] Compared with the prior art, the beneficial effects of the present invention are as follows: The present invention discloses a method and system for intelligently controlling carbon source addition in sewage treatment based on a hybrid model, comprising the following steps: a step of determining a biochemical reaction mechanism model and an AI model; a step of setting an expected value of a reference amount; a step of acquiring real-time detection data; a step of acquiring real-time detection data of each sensor of a sewage treatment plant; a step of acquiring a recommended control value of a single model, and respectively acquiring a recommended control value M of the biochemical reaction mechanism model. E and the recommended control value M of the AI ​​model AI ; Obtain the recommended control value M of the mixed model; Equipment control step; Use the recommended control value M as the final carbon source dosage, and control the dosing equipment of the carbon source dosing system to perform the carbon source dosing operation according to the recommended control value M.

[0019] The hybrid model-based sewage treatment carbon source addition intelligent control method and system of the present invention utilizes the self-learning ability of the AI ​​model and the theoretical and empirical guarantee of the mechanism model to make the best control decision based on real-time data, which can solve the problems of insufficient flexibility and poor adaptability in the existing automatic control system, and can not only improve the sewage treatment efficiency, but also further reduce the operation and maintenance costs. The system of the present invention can not only be used for adaptive dynamic adjustment of the carbon source addition amount, but also for intelligent automatic adjustment of other operational control quantities in the sewage treatment process.

[0020] The hybrid model-based intelligent control method and system for carbon source addition in sewage treatment of the present invention have the advantages of comprehensively utilizing the advantages of the mechanism model and the AI ​​model, obtaining the optimal control decision, improving the working efficiency of sewage treatment and reducing operation and maintenance costs. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] Figure 1 The present invention is a flow chart of a method for controlling the dosage of a chemical in a sewage treatment process.

[0022] Figure 2 The invention discloses a schematic diagram for obtaining an optimized carbon source dosage amount using a hybrid model of a method for controlling the dosage of a reagent in a sewage treatment process.

[0023] Figure 3 is a suggested control value M for the biochemical reaction mechanism model of a method for controlling the dosage of a reagent in a sewage treatment process of the present invention. E and the recommended control value M of the AI ​​model AI Schematic diagram of the process of performing fusion.

[0024] Figure 4 It is the effluent nitrate nitrogen detection value used as a reference amount in a certain period of time when the present invention is applied to a certain sewage treatment plant.

[0025] Figure 5 for Figure 4Real-time adjustment results of AI model weights and mechanism model weights during runtime.

[0026] Figure 6 for Figure 4 The relationship between the mechanism recommendation value, AI recommendation value, and fusion recommendation value and the manual operation value is shown in the form of a scatter plot.

[0027] The present invention will be further described below through specific implementation modes in conjunction with the accompanying drawings. DETAILED DESCRIPTION

[0028] See also Figure 1 to Figure 6 The present invention discloses a method for intelligently controlling carbon source addition in sewage treatment based on a hybrid model, comprising the following steps: Step 1: Determination of the biochemical reaction mechanism model and the AI ​​model; Determine the biochemical reaction mechanism model of the sewage treatment plant; Obtain the historical sensor detection data and historical operation data of the sewage treatment plant, and build an AI model for estimating the carbon source dosage based on the sensor detection historical data and historical operation data and train the AI ​​model; In the present invention, machine learning or deep learning technology is used to build and train a data-driven AI model based on the historical sensor detection data and historical operation records of the sewage treatment plant, and a maturely trained AI model is obtained to estimate the amount of carbon source added in the sewage treatment process.

[0029] The input of the AI ​​model is mainly part or all of the historical detection data of various sensors on the sewage treatment process line, including but not limited to inlet and outlet water volume and water quality data (such as total phosphorus TP, total nitrogen TN, chemical oxygen demand COD, ammonia nitrogen NH3-N, temperature, pH, etc.) and process sensor data (such as dissolved oxygen DO, air volume, mixed liquid suspended solids concentration MLSS, etc.), including recorded data or current detection data during historical operation, and its output is the carbon source dosage. The process sensor refers to other output instruments on the sewage treatment process line except for the inlet and outlet sensors.

[0030] Step 2: setting the expected value of the reference amount; at a designated location of the sewage treatment plant, determining the water quality index of the location, and determining the expected value of the reference amount of the water quality index; The reference quantity refers to one or more water quality indicators that need to be monitored at a certain point in the sewage treatment process line. Under the premise of safe operation, the upper limit or range value allowed by these indicators is called the expected value T of the reference quantity. The expected value T is expressed by a vector. There can be only one value or multiple values ​​in the vector, including the expression of the data interval. The key points of sewage treatment include water inlet, lift pump room, grille, sedimentation tank, anaerobic tank, anoxic tank, aerobic tank, secondary sedimentation tank, intermediate lift pump room, phosphorus removal coagulation tank, filter tank and outlet, etc. These points each undertake different tasks in the sewage treatment process, and jointly ensure the efficiency and stability of sewage treatment.

[0031] The expected value of the reference amount is determined and set by the operator based on experience. As long as the reference amount at the site is maintained below the expected value or within the range, the effluent quality can be guaranteed to meet the standard. For example, for the calculation of the carbon source dosage, depending on the different dosing locations, one or more of the indicators such as anoxic pool ammonia nitrogen, effluent nitrate nitrogen, total nitrogen TN, COD, sludge characteristics, etc. can be used as reference amounts, but not limited to.

[0032] Step 3: Real-time detection data acquisition step; obtaining real-time detection data of each sensor of the sewage treatment plant; Obtain the real-time data of each sensor on the sewage treatment process line of the sewage treatment plant, including the real-time detection value N of the reference quantity mentioned in step 2; the real-time detection value N corresponds one-to-one to the expected value T of the reference quantity, and N is expressed by a vector. There can be only one value or multiple values ​​in the vector, including the expression of the data interval. When the operator deems it necessary, for example, when the sensor is being calibrated as per routine, manual detection data or data before calibration can be used as the real-time detection data of the sensor.

[0033] Step 4: Obtaining the recommended control value of a single model; Obtain the recommended control value M of the biochemical reaction mechanism model respectively E and the recommended control value M of the AI ​​model AI ; Based on the current sensor data, use the AI ​​model built and trained in step 1 to obtain the AI ​​model estimated value of the required carbon source dosage as the recommended control value M of the AI ​​model. AI At the same time, according to the current sensor data, the mechanism model estimation value of the required carbon source dosage is obtained based on the sewage treatment biochemical reaction mechanism model as the recommended control value M of the mechanism model. E ; Step 5: Obtain the suggested control value M of the mixed model; the suggested control value M of the biochemical reaction mechanism model E and the recommended control value M of the AI ​​model AIPerform fusion to obtain a recommended control value M of the fused hybrid model; Referring to the real-time change trend of the real-time detection value N of the controlled target and its difference with the expected value, the above M obtained in step 4 is E and M AI After fusion and adjustment, the optimized required carbon source dosage is obtained, which is called the recommended control value M of the hybrid model.

[0034] Step 6: Equipment control step; taking the recommended control value M as the final carbon source dosage, controlling the dosing equipment of the carbon source dosing system to perform the carbon source dosing operation according to the recommended control value M.

[0035] The optimized carbon source dosage obtained in step 5 is transmitted to the carbon source addition system for automatic carbon source addition operation, or is transmitted as a recommended control value to the operator for reference and operation.

[0036] The present invention proposes a hybrid model of intelligent control method and system for carbon source addition in sewage treatment, which combines the biochemical reaction basis of the mechanism model with the data-driven advantages of the AI ​​model, and organically integrates the above two types of technologies by imitating human decision-making strategies. The carbon source dosage is adaptively and dynamically adjusted according to the influent load and process sensor detection value, which can not only ensure the safe and stable operation of sewage treatment, but also help achieve energy conservation and consumption reduction by optimizing the carbon source dosage.

[0037] During specific implementation, in step 1, the biochemical reaction mechanism model is an activated sludge model.

[0038] In specific implementation, the biochemical reaction mechanism model of sewage treatment can be an activated sludge series model, or a calculation formula derived by operators based on experience, etc. The mechanism model can use the existing operating program of the sewage treatment plant, or it can be rebuilt and written in addition.

[0039] In the specific implementation, in step 1, the AI ​​model is any one of the machine learning or deep learning models established based on SVM, decision tree, LSTM, GCN or Transformer.

[0040] The AI ​​model may be, but is not limited to, a model based on SVM (Support Vector Machine), decision tree or LSTM (Long Short-Term Memory), GCN (Graph Convolutional Network), Transformer, etc.

[0041] ‌SVM is a generalized linear classifier that performs binary classification of data in a supervised learning manner. Its decision boundary is the maximum margin hyperplane for solving learning samples. Decision Tree is a decision analysis method that expresses the mapping relationship between object attributes and object values ​​by constructing a decision tree based on the known probability of occurrence of various situations, and then evaluates project risks and judges its feasibility by obtaining the probability that the expected value of the net present value is greater than or equal to zero. It is a graphical method for intuitively using probability analysis. Since this decision branch is drawn into a graph that looks like a tree branch, it is called a decision tree. LSTM is a special recurrent neural network (RNN). By introducing a gating mechanism: including a forget gate, an input gate, and an output gate, it can reduce the gradient vanishing or gradient exploding problems faced by traditional RNN when processing long sequence data. LSTM is suitable for processing and predicting important events with very long intervals and delays in time series. GCN is a convolutional neural network that can directly act on graphs and use their structural information. It generalizes the convolution operation from traditional array data (images or grids) to graph data. GCN methods are divided into two categories, spectral domain-based methods and spatial domain-based methods. Spectral domain-based methods define graph convolution by introducing filters from the perspective of graph signal processing, and spatial domain-based methods represent graph convolution as aggregating feature information from neighbors. The Transformer model is a deep learning architecture originally proposed by Vaswani et al. in 2017 and used for natural language processing (NLP) tasks. The core innovation of the Transformer model is the introduction of the self-attention mechanism, which enables the Transformer model to perform well when processing sequence data, especially in capturing long-distance dependencies. In the specific implementation, in step 5, the recommended control value M E and the recommended control value M AI The fusion formula is: M=w E *M E +w AI *M AI (1) In formula (1), w E The suggested control value M for the biochemical reaction mechanism model E The weight value, w AI is the recommended control value M for the AI ​​model AI The weight value of ; M is the recommended control value of the mixed model.

[0042] Fusion means that, in most cases, the proposed control value M of the AI ​​model obtained in step 4 AI The suggested control value M of the mechanism model Ewill not be equal, so it is necessary to obtain a unique value based on these two values ​​through calculation to determine the final carbon source dosage, that is, to merge these two values ​​so as to provide them to operators or execution systems for carbon source dosage operations. The recommended control value M of the hybrid model finally calculated is used as the carbon source dosage for specific operations.

[0043] In specific implementation, any other linear or nonlinear function may be used to make M satisfy the following condition (2): M∈[min(M AI , M E ), max(M AI , M E )] (2) In formula (2), M is AI and M E The range formed by the two values.

[0044] In the specific implementation, it also includes the weight value w E and weight value w AI dynamic adjustment process.

[0045] In specific implementation, the weight value w E And weight value is adjusted w AI Adjustment means that E and M AI The weight w for weighted averaging E and w AI It is not a fixed empirical value, but will be adaptively and dynamically adjusted according to the actual operating conditions of the system to ensure that the optimized recommended control value M is obtained, that is, the optimal carbon source dosage M is obtained.

[0046] In the specific implementation, the sensitivity S is used to adjust the weight value w in the dynamic adjustment process. E and weight value w AI Perform dynamic adjustment, where 0<S≤1.

[0047] In specific implementation, considering the complexity of the sewage treatment biochemical system, the weights of different carbon source dosage estimation models are not always constant during fusion, and the speed of change can also be adjusted according to the situation. Sensitivity S is introduced in the present invention to adjust the speed of change of weights of different models. The value range of sensitivity S is set to (0,1], that is, 0<S≤1. A higher sensitivity S means a faster response speed, and the model weight can be adjusted quickly to obtain a better recommended control value M more quickly, but it may also cause the recommended control value M to fluctuate greatly, affecting the stability of the dosing system and sewage treatment. On the contrary, although the lower sensitivity S has a slower response, it can provide a smoother output, which helps to maintain the stable operation of the system. The specific value of sensitivity S needs to be based on the actual operation of the specific sewage treatment plant, and a balance should be struck between rapid optimization and stable operation.

[0048] In specific implementation, the weight adjustment amount Δw is calculated using the following formula (3); Δw=S*(1 - exp(-(NT) 2 )) (3); In the formula (3), N is the real-time detection value of the reference quantity, and T is the expected value of the reference quantity.

[0049] Weight value w E and weight value w AI The basis for adjustment is mainly the difference between the expected value T of the reference quantity set in step 2 and the real-time detection value N of the reference quantity obtained in step 3, and is also related to the set sensitivity S. In addition to formula (3), other forms of formulas can also be used to calculate the weight adjustment amount Δw. The calculated weight adjustment amount Δw will be respectively in the weight w E and w AI Add to one and subtract from the other.

[0050] In specific implementation, when the real-time detection value N of the reference quantity is greater than the expected value T of the reference quantity, the weight w of the mechanism model is increased. E , reduce the weight w of the AI ​​model AI ; When the real-time detection value N of the reference quantity is less than the expected value T of the reference quantity, the weight w of the AI ​​model is increased AI , reduce the weight w of the mechanism model E .

[0051] The strategy of weight adjustment is: when the real-time detection value N of the reference quantity is greater than the expected value T of the reference quantity, increase the weight w of the mechanism model E , reduce the weight w of the AI ​​model AI ; That is, when N>T, w E ↑,w AI↓; On the contrary, when the real-time detection value N of the reference quantity is less than the expected value T of the reference quantity, the weight w of the AI ​​model is increased AI , reduce the weight w of the mechanism model E ; That is, when N≤T, w AI ↑,w E ↓.

[0052] When making specific adjustments, set the high value S of the sensitivity S respectively. up and the low value S of sensitivity S down , S up and S down The two values ​​can be equal or unequal.

[0053] Compare the reference value N with the expected value T. If N>T, then take S = S up , calculate Δw according to formula (3), and then adjust w according to formula (4) AI And adjust w accordingly E .

[0054] w AI =max(w AI -Δw, 0), w E =1-w AI (4) Compare the reference value N with the expected value T. If N≤T, then take S = S down , calculate Δw according to formula (3), and then adjust w according to formula (5) AI And adjust w accordingly E .

[0055] w AI =min(w AI +Δw,1), w E =1-w AI (5) Generally speaking, it can be considered that when the real-time detection value N of the reference quantity is greater than the expected value T of the reference quantity, a higher sensitivity S tends to be selected in order to respond more quickly to reduce the real-time detection value N of the reference quantity.

[0056] Figure 2 The hybrid model of the present invention is used to obtain the optimal carbon source dosage. The intelligent control engine is used to calculate the weight w of the physical model. E and the weight w of the AI ​​model AI .

[0057] The present invention also discloses a sewage treatment carbon source dosing intelligent control system based on a hybrid model, comprising: A model determination unit, used to determine the biochemical reaction mechanism model of the sewage treatment plant, obtain the historical sensor detection data and historical operation data of the sewage treatment plant, and build an AI model for estimating the carbon source dosage based on the historical sensor detection data and historical operation data and train the AI ​​model; Data input unit: used to set the expected value T of the reference quantity in the sewage treatment process; The sensor unit is used to obtain real-time detection data from sensors at various points in the sewage treatment plant; The recommended control value calculation unit is used to calculate the recommended control value M of the biochemical reaction mechanism model. E and the recommended control value M of the AI ​​model AI ; A data fusion unit for suggesting control values ​​M for the biochemical reaction mechanism model E and the recommended control value M of the AI ​​model AI Perform fusion to obtain a recommended control value M of the fused hybrid model; The control unit is used to use the recommended control value M as the final carbon source dosage, and control the dosing equipment of the carbon source dosing system to perform the carbon source dosing operation according to the recommended control value M.

[0058] In specific implementation, the sewage treatment carbon source dosing intelligent control system based on the hybrid model also includes an intelligent control engine for calculating the weight w of the physical model. E and the weight w of the AI ​​model AI .

[0059] The operation process of the sewage treatment carbon source addition intelligent control system based on the hybrid model mainly includes the following steps: Step 01: Start the system and perform initial settings; When the carbon source dosing intelligent control system of the present invention is connected to the sewage treatment process line for operation, it is initially set to operate according to the mechanism model, that is, take w AI = 0, w E = 1.

[0060] Step 02: The same as the step of setting the expected value T of the reference quantity in step 2 above; Steps 03 to 06: correspond to steps 3 to 6 of the intelligent control method for carbon source addition in sewage treatment based on the hybrid model, respectively; Step 07: Continue looping; If the expected value T of the reference quantity does not need to be changed, repeat steps 3 to 6; if the expected value T of the reference quantity needs to be changed, repeat steps 2 to 6.

[0061] like Figure 3 is the suggested control value M for the biochemical reaction mechanism modelE and the recommended control value M of the AI ​​model AI Schematic diagram of the process of performing fusion.

[0062] Figure 3 In the figure, the horizontal dotted line indicates the optimal carbon source dosage that makes the effluent meet the standard at the current moment, which is an ideal value. The dotted line indicates the projection of the value range space of the carbon source dosage solution on a certain space slice. Figure 3 In the figure, the solid triangle represents the suggested control value M of the biochemical reaction mechanism model. E , the solid circle represents the recommended control value M of the AI ​​model AI ; The hollow circle represents the recommended control value M after the hybrid model is integrated; M is in M AI and M E They are all values ​​obtained within a reasonable search space. When the system is running stably, the AI ​​estimation value generally uses less dosage than the mechanism estimation value.

[0063] Figure 3 In the figure, the operation status of four moments, namely, moment 0, moment 1, moment 2 and moment 3, is reflected.

[0064] At time 0, the real-time detection value N of the reference quantity is qualified (qualified means that the real-time detection value N of the reference quantity is less than the expected value T), and the hybrid model uses the recommended control value M of the AI ​​model. AI ; At time 1, due to sudden changes in the environment or water inflow, the real-time detection value N of the reference quantity suddenly increases. At this time, the recommended control value M E , recommended control value M AI and the recommended control value M are both increasing. If the recommended control value M AI If the water quality cannot meet the standard, the hybrid model will make M move to the recommended control value M E near; At time 2, since the actual dosage used exceeds the optimal dosage, the problem of excessive effluent is solved, and the reference quantity detection value N is restored to qualified. At this time, the hybrid model still tends to adopt the mechanism model; At time 3, after the system stabilizes, the hybrid model can re-propose the control value M AI Approach and fully adopt the recommended control value M of the AI ​​model AI to run.

[0065] In the early stage of implementation of the method and system of the present invention, we tend to choose the recommended control value M EAs the output of M. Then, the M value at each moment is determined by comparing the detection value N of the reference amount with the expected value T: if the detection value N of the reference amount is greater than the expected value T, it means that the effluent is at risk of exceeding the standard, and the carbon source dosage needs to be shifted to the model with a larger estimated value to ensure that the effluent meets the standard; conversely, if the detection value N is already less than the expected value T, then the carbon source dosage can gradually tend to adopt the model with a smaller estimated value to achieve energy saving and consumption reduction in process operation.

[0066] In the present invention, different sensitivities can be used for the two situations of effluent exceeding the standard and effluent meeting the standard. When the effluent exceeds the standard, the adjustment of the carbon source dosage should be a quick response to reduce the impact of the impulse as soon as possible. When the effluent meets the standard, the model can be gradually adjusted to a smaller carbon source dosage to make the system run more stably. In this way, the method and system of the present invention can not only effectively respond to emergencies, but also achieve rational use of resources while ensuring the quality of effluent.

[0067] The method and system of the present invention are implemented in a sewage treatment plant in Anhui Province, and automatic control of carbon source addition is performed in the anoxic tank.

[0068] In the system of a sewage treatment plant in Anhui, the reference amount used is effluent nitrate nitrogen. The detection frequency of effluent nitrate nitrogen is about once every 5 minutes, and the expected value of effluent nitrate nitrogen is set to T=6.5. The AI ​​model is built and trained based on the historical data of the plant, and its input uses all the sensor data that can be obtained at each key section of the plant's process line. The mechanism model is built based on the knowledge and experience of water experts and the plant's management and operation personnel.

[0069] When the system of the present invention takes over the operation, the weight w of the mechanism model E is 1, the weight w of the AI ​​model AI It is 0. During operation, the optimized carbon source dosage M output by the hybrid model is obtained by comparing the sensor detection value N of the reference effluent nitrate nitrogen with the set effluent nitrate nitrogen expected value T; if the effluent nitrate nitrogen detection value N is less than the set effluent nitrate nitrogen expected value T, indicating that the effluent meets the standard, then the fusion optimization of the carbon source dosage will gradually lean towards the recommended control value M of the AI ​​model. AI In this embodiment, according to the sensitivity S=0.2 (that is, S takes S down ) settings gradually increase the weight w of the AI ​​model AI ; When the detected value N of effluent nitrate nitrogen is greater than the set expected value T of effluent nitrate nitrogen, it indicates that there is a risk of exceeding the standard in effluent, then the weight w of the mechanism model needs to be increased quickly. E To ensure that the water meets the standard, in this embodiment, according to the sensitivity S=0.5 (that is, S takes S up ) settings quickly bias the mechanistic model estimates.

[0070] Figure 4 It is the nitrate nitrogen detection value of the plant's effluent as a reference over a certain period of time. Figure 4 The data time period is: 10:14 on July 20, 2024 to 3:33 on July 21, 2024.

[0071] Figure 5 It is the real-time adjustment result of the AI ​​model weight and the mechanism model weight when the system of the present invention is running during this period of time.

[0072] The plant also adopted a concomitant operation mode to implement it to test the effect of the method and system of the present invention, which means that on the process line, the operator still determines and actually operates the carbon source dosage of the anoxic tank, and this dosage is called the "manual operation value"; at the same time, the sensor data obtained on the process line is input into the invented system running in the computer to obtain the carbon source dosage output by the mechanism model, called the "mechanism recommended value", the carbon source dosage output by the AI ​​model, called the "AI recommended value", and the carbon source dosage output by the hybrid model, called the "fusion recommended value", and these three recommended values ​​are compared with the manual operation values ​​respectively.

[0073] Figure 6 The relationship between the mechanism recommendation value, AI recommendation value, fusion recommendation value and manual operation value is shown in the form of a scatter plot. In the figure, the manual operation value is represented as the value on the horizontal axis, and the values ​​on the vertical axis are the mechanism recommendation value, AI recommendation value and fusion recommendation value respectively.

[0074] from Figure 6 You can see: The points where the mechanism-recommended values ​​are compared with the manual operation values ​​are evenly distributed on the straight line y=x and its neighbors, and the fitting straight line can basically coincide with y=x, indicating that the manual operation values ​​are basically equivalent to the mechanism-recommended values, which is consistent with the fact that the manual operation values ​​are derived from the empirical formula, that is, from the mechanism; Most of the points where AI recommended values ​​are compared with manual operation values ​​are distributed below the straight line y=x, that is, most AI recommended values ​​are smaller than manual operation values, indicating that the AI ​​model can statistically find a better estimate of the carbon source dosage, and the AI ​​recommended values ​​can save more than the manual operation values ​​or the mechanism recommended values; Compared with the manual operation values, more points of the fusion recommended values ​​are distributed below the straight line y=x, indicating that statistically the fusion recommended values ​​can combine the strengths of the mechanism model and the AI ​​model to achieve optimization and obtain further savings in dosage, and the fusion optimization is effective.

[0075] The present invention provides a hybrid model sewage treatment carbon source addition intelligent control method and system, which has the following characteristics.

[0076] 1. The method of model fusion and adaptive dynamic adjustment of weights includes the method of weighted average fusion of mechanism model and artificial intelligence model, as well as the algorithm and logic of dynamic adjustment of weights according to the system operation status. It can combine the strengths of both mechanism model and AI model and avoid their weaknesses to obtain the most optimized control method.

[0077] 2. Sensitivity setting and application, the concept of sensitivity, the setting of value range and the application method under different operating conditions (water outlet exceeding the standard and meeting the standard). This unique adjustment mechanism helps to balance system performance and improve operational efficiency.

[0078] 3. Carry out the logic of fusion optimization, and select the specific logic and process of fusion optimization value between the AI ​​model recommended value and the mechanism model recommended value according to different situations (normal, environmental mutation, water quality impact, whether the effluent meets the standard, etc.). The system can generate a more optimized carbon source dosage.

[0079] 4. Overall control strategy based on feedback. An overall feedback control strategy is used to adjust the carbon source dosage based on the comparison between the reference value and the expected value, including the conditions and operation methods for shifting to different models, which reduces resource consumption while ensuring that the effluent meets the standards, thereby achieving the goal of energy conservation and emission reduction.

[0080] The present invention provides a hybrid model of intelligent control method and system for carbon source addition in sewage treatment. The system can combine the advantages of mechanism model and AI model and avoid their disadvantages. By dynamically adjusting the fusion ratio of the two models, the system can generate a more optimized carbon source dosage, while ensuring that the effluent meets the standards, reducing resource consumption, thereby achieving the goal of energy conservation and emission reduction.

[0081] The present invention provides a mixed model sewage treatment carbon source addition intelligent control method and system, which can realize automatic calculation and adjustment of carbon source addition amount, reduce the labor intensity of operators and improve operation efficiency.

[0082] In specific implementation, the monomer model in the method of the present invention is not limited to the AI ​​model and the mechanism model, and other forms of models may also be used; the number of models used for mixing is not limited to two models, and a larger number of models may also be used. In addition, the method of the present invention is not only applicable to the intelligent control of carbon source addition, but also to the intelligent control of other operational control quantities that need to be regulated in the sewage treatment process, such as air volume control, iron salt addition control, etc.

[0083] The hybrid model of the sewage treatment carbon source addition intelligent control method and system of the present invention also provides an interface for manual intervention. Authorized operators can change the expected value of the reference amount, the sensitivity setting value, etc. at any time, providing a basis for further energy conservation and emission reduction.

[0084] It will be apparent to those skilled in the art that the invention is not limited to the details of the exemplary embodiments described above and that the invention can be implemented in other specific forms without departing from the spirit or essential features of the invention. Therefore, the embodiments should be considered exemplary and non-limiting in all respects, and the scope of the invention is defined by the appended claims rather than the foregoing description, and it is intended that all variations falling within the meaning and scope of the equivalent elements of the claims be included in the invention. Any reference numeral in a claim should not be considered as limiting the claim to which it relates.

[0085] In addition, it should be understood that although the present specification is described according to implementation modes, not every implementation mode contains only one independent technical solution. This description of the specification is only for the sake of clarity. Those skilled in the art should regard the specification as a whole. The technical solutions in each embodiment may also be appropriately combined to form other implementation modes that can be understood by those skilled in the art.

Claims

1. An intelligent control method for carbon source addition in sewage treatment based on a hybrid model, characterized in that: The steps include: Step 1: Determination of the biochemical reaction mechanism model and the AI ​​model; Determine the biochemical reaction mechanism model of the sewage treatment plant; Obtain the historical sensor detection data and historical operation data of the sewage treatment plant, and build an AI model for estimating the carbon source dosage based on the sensor detection historical data and historical operation data and train the AI ​​model; Step 2: setting the expected value of the reference amount; at a designated location of the sewage treatment plant, determining the water quality index of the location, and determining the expected value of the reference amount of the water quality index; Step 3: Real-time detection data acquisition step; obtaining real-time detection data of each sensor of the sewage treatment plant; Step 4: Obtaining the recommended control value of a single model; Obtain the recommended control value M of the biochemical reaction mechanism model respectively E and the recommended control value M of the AI ​​model AI ; Step 5: Obtain the suggested control value M of the mixed model; the suggested control value M of the biochemical reaction mechanism model E and the recommended control value M of the AI ​​model AI Perform fusion to obtain a recommended control value M of the fused hybrid model; Step 6: Equipment control step; taking the recommended control value M as the final carbon source dosage, controlling the dosing equipment of the carbon source dosing system to perform the carbon source dosing operation according to the recommended control value M.

2. According to claim 1, a method for intelligent control of carbon source addition in sewage treatment based on a hybrid model is characterized in that: In the step 1, the biochemical reaction mechanism model is an activated sludge model.

3. According to claim 1, a method for intelligent control of carbon source addition in sewage treatment based on a hybrid model is characterized in that: In step 1, the AI ​​model is any one of the machine learning or deep learning models established based on SVM, decision tree, LSTM, GCN or Transformer.

4. According to claim 1, a method for intelligent control of carbon source addition in sewage treatment based on a hybrid model is characterized in that: In step 5, the recommended control value M E and the recommended control value M AI The fusion formula is: M=w E *M E +w AI *M AI (1) In formula (1), w E The suggested control value M for the biochemical reaction mechanism model E The weight value, w AI is the recommended control value M for the AI ​​model AI The weight value of ; M is the recommended control value of the mixed model.

5. The method for intelligently controlling carbon source addition in sewage treatment based on a hybrid model according to claim 4 is characterized in that: It also includes the weight value w E and weight value w AI dynamic adjustment process.

6. The method for intelligently controlling carbon source addition in sewage treatment based on a hybrid model according to claim 5 is characterized in that: In the dynamic adjustment process, the sensitivity S is used to adjust the weight value w E and weight value w AI Perform dynamic adjustment, where 0<S≤1.

7. The method for intelligently controlling carbon source addition in sewage treatment based on a hybrid model according to claim 6 is characterized in that: The weight adjustment amount Δw is calculated using the following formula (3); Δw=S*(1 - exp(-(N-T) 2 )) (3); In the formula (3), N is the real-time detection value of the reference quantity, and T is the expected value of the reference quantity.

8. The method for intelligently controlling carbon source addition in sewage treatment based on a hybrid model according to claim 7 is characterized in that: When the real-time detection value N of the reference quantity is greater than the expected value T of the reference quantity, the weight w of the mechanism model is increased. E , reduce the weight w of the AI ​​model AI ; When the real-time detection value N of the reference quantity is less than the expected value T of the reference quantity, the weight w of the AI ​​model is increased AI , reduce the weight w of the mechanism model E .

9. An intelligent control system for carbon source addition in sewage treatment based on a hybrid model according to claim 1, characterized in that: include: A model determination unit, used to determine the biochemical reaction mechanism model of the sewage treatment plant, obtain the historical sensor detection data and historical operation data of the sewage treatment plant, and build an AI model for estimating the carbon source dosage based on the historical sensor detection data and historical operation data and train the AI ​​model; Data input unit: used to set the expected value T of the reference quantity in the sewage treatment process; The sensor unit is used to obtain real-time detection data from sensors at various points in the sewage treatment plant; The recommended control value calculation unit is used to calculate the recommended control value M of the biochemical reaction mechanism model. E and the recommended control value M of the AI ​​model AI ; A data fusion unit for suggesting control values ​​M for the biochemical reaction mechanism model E and the recommended control value M of the AI ​​model AI Perform fusion to obtain a recommended control value M of the fused hybrid model; The control unit is used to use the recommended control value M as the final carbon source dosage, and control the dosing equipment of the carbon source dosing system to perform the carbon source dosing operation according to the recommended control value M.

10. An intelligent control system for carbon source addition in sewage treatment based on a hybrid model according to claim 9, characterized in that: It also includes an intelligent control engine for calculating the weights w of the physical model. E and the weight w of the AI ​​model AI .

Images