Intelligent off-peak operation method and system for latticed storage tank of buried sewage plant

Abstract

The invention relates to the technical field of sewage treatment, in particular to an intelligent off-peak operation method and system for a grid-divided storage tank of a buried sewage plant, and the method comprises the steps: obtaining the operation data of the storage tank, carrying out the data integration of the obtained operation data, and carrying out the construction of a prediction model based on the fused data. Flood season regulation and storage and precipitation control are conducted according to the prediction result, after precipitation is completed, a water outlet gate is opened to conduct drainage operation, washing time is calculated and controlled according to the volume of a regulation and storage pond and the flow of washing liquid, and precipitation of the current batch is completed; according to the method, in the regulation and storage and precipitation control process in the flood season, the mud discharge time and the mud discharge amount are calculated and determined through the PID control algorithm, in the water inlet control stage, the opening degree of the water inlet gate is adjusted in proportion according to the deviation between the liquid level rising speed and the threshold value, accurate control over the water inlet flow can be achieved, and the stability of the water inlet gate is improved. And overflow caused by too fast water inflow of the storage pond is avoided.

Description

Technical Field

[0001] The present invention relates to the technical field of sewage treatment, and in particular to an intelligent peak-shifting operation method and system for a sectional storage tank of an underground sewage treatment plant. Background Art

[0002] In the field of sewage treatment, the front-end storage tank of a sewage treatment plant plays a crucial role in the entire sewage treatment process. It has multiple functions such as peak shaving, valley filling, water quality equalization, and emergency buffering. It can effectively cope with the fluctuations in the quantity and quality of the incoming water, ensure the stable operation of the subsequent treatment process, control the pollution of rainwater overflow, and enhance the urban resilience in the face of extreme weather.

[0003] However, there are many problems that cannot be ignored in the actual operation of the current traditional front-end storage tank of a sewage treatment plant. On the one hand, due to the uncertainty of the incoming water quantity and quality, the liquid level in the storage tank changes greatly. This large fluctuation not only increases the difficulty of operation and management, but also makes the sludge in the tank extremely easy to deposit. It is very difficult to select a suitable mixing device to solve the deposition problem. Even if a mixing device is installed, the actual mixing and flushing effects are not ideal, and the sludge accumulation cannot be effectively avoided, seriously affecting the normal function of the storage tank.

[0004] On the other hand, the storage tank usually has a large volume, which leads to high engineering investment and construction costs. In the actual operation process, the water quantity generally cannot reach the designed maximum volume condition, and a large amount of idle volume is wasted, resulting in unreasonable utilization of resources and increased sewage treatment costs. The existing technical solutions for these problems have limited effects. The traditional mixing devices cannot adapt to the complex and changeable working conditions of the storage tank. The flushing method not only consumes a large amount of water resources, but also is difficult to completely remove the sludge. Moreover, the operation mode lacks flexibility and intelligence, and cannot be effectively adjusted according to different water quantity, water quality conditions and the actual needs of the sewage treatment plant. At present, an intelligent peak-shifting operation method and system for a sectional storage tank of an underground sewage treatment plant are needed. Summary of the Invention

[0005] In order to solve the problems of difficult management caused by the large change range of the liquid level in the storage tank and the waste of the storage tank volume, the present invention provides an intelligent peak-shifting operation method and system for a sectional storage tank of an underground sewage treatment plant.

[0006] In the first aspect, an intelligent peak-shifting operation method for a sectional storage tank of an underground sewage treatment plant provided by the present invention adopts the following technical solutions:

[0007] An intelligent peak-shifting operation method for a sectional storage tank of an underground sewage treatment plant includes:

[0008] Obtain the operation data of the storage tank, and integrate the obtained operation data, including obtaining the storage inlet water volume data and meteorological data, and associating the two in terms of time;

[0009] Construct a prediction model based on the fused data, including constructing an inlet water flow prediction model by using time series analysis combined with long short-term memory network;

[0010] Conduct flood season storage and sedimentation control according to the prediction results, including calculating and determining the sludge discharge time and sludge discharge volume by using the PID control algorithm, and optimizing the control through the genetic algorithm;

[0011] After sedimentation is completed, open the outlet gate for drainage operation, calculate and control the flushing time according to the storage tank volume and flushing liquid flow rate, and complete the sedimentation of the current batch;

[0012] Conduct stability calculation according to the inlet water flow, and adjust the sedimentation time of the next batch by using the feedback control algorithm based on the data of the sedimentation effect of the previous batch;

[0013] Set up an overflow storage tank according to the inlet water flow stability judgment result and the operation state of the storage tank, and construct an overflow risk assessment model to evaluate the overflow event.

[0014] Furthermore, the integration of the obtained operation data includes, based on the obtained storage inlet water volume data and meteorological data, using a filtering algorithm to remove noise, and performing outlier processing through the Z-score method. Taking the timestamp as the basis, associate the storage inlet water volume data and meteorological data at the same moment, establish a linear regression equation for the storage inlet water volume data and meteorological data according to historical data, and determine the relationship between the rainfall threshold, duration and the increase in inlet water flow according to the linear regression equation.

[0015] Furthermore, the construction of the inlet water flow prediction model by using time series analysis combined with long short-term memory network includes using the time series analysis method to process the historical inlet water flow data, extracting statistical features and performing seasonal decomposition, then normalizing the inlet water flow data and meteorological data, combining the features extracted by time series analysis with the original inlet water flow data and meteorological data as the input of the LSTM network, setting multiple LSTM layers, each LSTM layer containing multiple LSTM units, controlling the information transmission and forgetting through the gating mechanism, adding a fully connected layer after the LSTM layer, mapping the features output by the LSTM layer to a one-dimensional vector, and finally training the water flow prediction model.

[0016] Further, the training of the water flow prediction model includes pre-training the prediction model by using transfer learning with historical influent flow data and meteorological data of other sewage treatment plants, and then inputting the data of this sewage treatment plant into the pre-trained LSTM model for fine-tuning. Using the historical influent flow data and real-time meteorological data of this sewage treatment plant as inputs, the fine-tuned prediction model is trained, and the mean square error is used as the loss function and the model is optimized through the Adam algorithm.

[0017] Further, the calculation and determination of the sludge discharge time and sludge discharge volume by using the PID control algorithm include performing storage preparation, influent control, and sedimentation and sludge discharge control according to the prediction results. The influent control is specifically as follows: when the influent peak is predicted, the influent gate is opened, the rising rate is calculated according to the liquid level in the storage tank, a rate threshold is set, and the proportional control algorithm is used to adjust the opening degree of the influent gate proportionally according to the deviation between the liquid level rising rate and the threshold. The calculation formula for the opening degree of the influent gate is:

[0018] Δα = K p E,

[0019] where E is the deviation between the rising rate and the threshold, and K p is the proportionality coefficient.

[0020] Further, the sedimentation and sludge discharge control includes, during the sedimentation process, using the sludge concentration deviation and the sludge level deviation as error inputs, calculating the control quantity by using the PID control algorithm, encoding the parameters of the PID control algorithm as chromosomes, randomly generating an initial population, and evaluating the individuals of the population by using the fitness function to complete the optimization of the PID control algorithm. The formula for the fitness function is:

[0021]

[0022] where n is the number of sampling points, w₁ and w₂ are the weight coefficients of the sludge concentration deviation and the sludge level deviation respectively, E C is the sludge concentration deviation, and E l is the sludge level deviation.

[0023] Further, the adjustment of the sedimentation time for the next batch by using the feedback control algorithm according to the data of the sedimentation effect of the previous batch includes obtaining the influent flow data sequence and calculating the average value, obtaining the flow fluctuation coefficient through the ratio of the standard deviation to the average value, setting a stability threshold and judging the stability according to the fluctuation coefficient. When the influent flow is relatively stable, collecting the data of the sedimentation effect of the previous batch to form a state vector, defining a reward function, and adjusting the sedimentation time for the next batch through the Q-learning algorithm.

[0024] Further, adjusting the sedimentation time of the next batch by using the Q-learning algorithm includes obtaining a state vector composed of suspended solids and sludge sedimentation ratio, setting the adjustment amount of the sedimentation time of the next batch as an action, establishing a reward function, and constructing a Q-learning algorithm formula by using the reward function:

[0025]

[0026] where α is the learning rate, which is used to control the step size when updating the Q value each time, γ is the discount factor, s is the current state, the state vector is composed of suspended solids and sludge sedimentation ratio, a is the adjustment amount of the sedimentation time of the next batch, and R(s,a) is the output value of the reward function. is the maximum Q value among all possible actions a' taken in the new state s'.

[0027] Further, constructing an overflow risk assessment model to evaluate overflow events includes using meteorological data, influent flow data, and the liquid level data of the storage tank as inputs, and constructing an overflow risk assessment model by using a multi-layer perceptron. The hidden layer performs a non-linear transformation on the input data through an activation function to extract features in the data, and the output of the output layer is used as the evaluation result of the overflow risk.

[0028] In a second aspect, an intelligent peak-shifting operation system for a buried sewage treatment plant with divided storage tanks includes:

[0029] A storage tank main body module: at least four divided storage tanks, each storage tank being a closed space for storing and regulating sewage;

[0030] A sludge treatment module: a sludge hopper, a sludge discharge pipe, and a plurality of sludge discharge valves. The sludge hopper is located at the bottom of each storage tank, the sludge discharge pipe is connected to the sludge hopper, and the sludge discharge valves are installed inside the sludge discharge pipe;

[0031] A data processing and decision-making module: a data acquisition unit, including a liquid level meter data acquisition unit and an external sensor interface, for collecting liquid level, flow rate, and sludge concentration data;

[0032] A data processing unit for analyzing and processing the collected data;

[0033] A decision control unit for controlling the actions of the influent gate, effluent gate, overflow flushing gate, sludge discharge valve, and sludge scraper equipment according to the data processing results.

[0034] In summary, the present invention has the following beneficial technical effects: / /

[0035] 1. The present invention removes noise by adopting a filtering algorithm and performs outlier processing using the Z-score method, effectively purifying the regulated water inflow data and meteorological data, ensuring the accuracy and reliability of the data, and providing a high-quality data basis for subsequent analysis and prediction.

[0036] 2. The present invention correlates the regulated water inflow data and meteorological data at the same moment based on timestamps, and establishes a linear regression equation according to historical data to determine the relationship between rainfall threshold, duration, and the increase in inflow rate, enabling the organic combination of data from different sources and providing the possibility for a more in-depth analysis of the influencing factors of the inflow rate.

[0037] 3. The present invention constructs an inflow rate prediction model by combining time series analysis with a long short-term memory network (LSTM). The time series analysis method is used to process historical inflow rate data, extract statistical features and perform seasonal decomposition. At the same time, the inflow rate data and meteorological data are normalized, and a variety of features are combined as the input of the LSTM network. This multi-method fusion approach can fully explore the potential information in the data and improve the accuracy and reliability of the prediction.

[0038] 4. The present invention adopts a transfer learning method. First, the prediction model is pre-trained using the historical inflow rate data and meteorological data of other sewage treatment plants, then the data of this sewage treatment plant is input into the pre-trained LSTM model for fine-tuning, and finally, the historical inflow rate data and real-time meteorological data of this sewage treatment plant are used for training. This method can not only reduce the consumption of training time and computing resources but also make full use of the empirical data of other sewage treatment plants to improve the generalization ability of the model.

[0039] 5. During the flood season regulation and sedimentation control process of the present invention, the PID control algorithm is used to calculate and determine the sludge discharge time and sludge discharge volume, and the genetic algorithm is used for control optimization. In the inlet control stage, according to the deviation between the liquid level rising rate and the threshold, the opening degree of the inlet gate is adjusted proportionally, which can achieve precise control of the inflow rate and prevent the regulating tank from overflowing due to too fast inflow. In the sedimentation and sludge discharge control stage, the sludge concentration deviation and sludge level deviation are used as error inputs, the PID control algorithm is used to calculate the control quantity, and the genetic algorithm is used to optimize the PID parameters, which can improve the sedimentation effect and sludge discharge efficiency and ensure the stable operation of the sewage treatment plant.

[0040] 6. The present invention calculates stability based on the influent flow rate. When the influent flow rate is relatively stable, the sedimentation time of the next batch is adjusted using a feedback control algorithm according to the data of the sedimentation effect of the previous batch. By obtaining the suspended solids and the sludge sedimentation ratio to form a state vector, the adjustment amount of the sedimentation time of the next batch is set as an action to establish a reward function, and the Q-learning algorithm is used to adjust the sedimentation time. This flexible way of adjusting the sedimentation time can be dynamically optimized according to the actual influent flow rate and sedimentation effect, improving the sedimentation efficiency and water quality treatment effect.

[0041] 7. The present invention constructs an overflow risk assessment model, taking meteorological data, influent flow rate data, and the liquid level data of the storage tank as inputs, using a multi-layer perceptron to construct the model, performing a non-linear transformation on the input data through the activation function of the hidden layer to extract the features in the data, and taking the output of the output layer as the assessment result of the overflow risk, which can early warn of the occurrence of overflow events and provide guarantee for the safe operation of the sewage treatment plant. BRIEF DESCRIPTION OF THE DRAWINGS

[0042] Figure 1 is the overall flow schematic diagram of an intelligent peak-shifting operation method for a sectional storage tank of an underground sewage treatment plant in Embodiment 1 of the present invention;

[0043] Figure 2 is the structural schematic diagram of a specific storage tank in an intelligent peak-shifting operation system for a sectional storage tank of an underground sewage treatment plant in Embodiment 2 of the present invention.

[0044] Among them, 1. overflow flushing gate; 2. liquid level gauge; 3. outlet gate; 4. inlet gate; 5. inlet channel; 6. outlet channel; 7. sludge discharge valve; 8. sludge discharge pipe; 9. sludge hopper; 10. sludge scraper; 11. storage tank. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0045] The present invention will be further described in detail below with reference to the accompanying drawings.

[0046] Embodiment 1

[0047] Referring to Figure 1 , an intelligent peak-shifting operation method for a sectional storage tank of an underground sewage treatment plant in this embodiment includes:

[0048] Obtain the operation data of the storage tank, and perform data integration on the obtained operation data, including obtaining the storage influent volume data and meteorological data, and performing time correlation on the two;

[0049] Based on the fused data, construct a prediction model, including constructing an influent flow rate prediction model using time series analysis combined with long short-term memory network;

[0050] Conduct flood-season storage and sedimentation control based on the prediction results, including calculating and determining the sludge discharge time and sludge discharge volume using the PID control algorithm, and optimizing the control through the genetic algorithm;

[0051] After sedimentation is completed, open the outlet gate for drainage operation, calculate and control the flushing time according to the storage tank volume and flushing liquid flow rate, and complete the sedimentation of the current batch;

[0052] Conduct stability calculation based on the influent flow rate, and adjust the sedimentation time of the next batch using the feedback control algorithm according to the data of the sedimentation effect of the previous batch;

[0053] Set up an overflow storage tank according to the judgment result of the influent flow rate stability and the operation status of the storage tank, and construct an overflow risk assessment model to evaluate the overflow event.

[0054] Specifically, an intelligent peak-shifting operation method for a sectional storage tank of an underground sewage treatment plant includes the following steps:

[0055] As Figure 1 shown, S1. Obtain the operation data of the storage tank, and integrate the obtained operation data, including obtaining the storage influent volume data and meteorological data, and associating the two in terms of time;

[0056] For the internal data of the storage tank, for the data collected from devices such as level gauges, flow meters, sludge concentration gauges, and sludge level gauges, use a filtering algorithm to remove noise, and perform outlier processing through the Z-score method. For external environmental data, such as rainfall data provided by the meteorological department, according to the distribution characteristics of the data, use methods such as mean and median to fill in the missing values, and standardize the data format so that all data conforms to a unified specification. Time association: Based on the time stamp, associate the internal data of the storage tank and the external environmental data at the same moment or in a similar time period;

[0057] Logical association: Establish a linear regression equation between the historical influent flow rate of the sewage treatment plant and the rainfall data, determine the relationship between the rainfall threshold, duration, and the increase in influent flow rate based on historical data, construct a dedicated database system to store the integrated data, and select a relational database (such as MySQL) or a non-relational database (such as MongoDB) according to the characteristics and application requirements of the data to classify and store the data according to dimensions such as data type and time, establish associations between different types of data through the time field, and regularly maintain and update the database, delete expired or useless data, and ensure the timeliness and accuracy of the data.

[0058] S2. Construct a prediction model based on the fused data, including constructing an influent flow rate prediction model using time series analysis combined with long short-term memory networks;

[0059] Based on the integrated data, time series analysis combined with the Long Short-Term Memory network (LSTM) is adopted, and transfer learning technology is introduced to build an influent flow prediction model. A time series refers to a sequence formed by arranging the values of a certain statistical indicator of a phenomenon at different times in chronological order. In the prediction of influent flow, historical influent flow data is a typical time series. The time series analysis method is used to preliminarily process the historical influent flow data. Calculate the statistical characteristics of the data, such as mean, variance, autocorrelation coefficient, etc. Seasonal decomposition can also be carried out to decompose the time series into a trend term, a seasonal term, and a residual term, so as to grasp the long-term trend and seasonal change law of the data. For example, through seasonal decomposition, it can be found that due to the large water consumption in summer, the influent flow may show a seasonal peak.

[0060] In this step, the main role of time series analysis is to perform preliminary feature extraction and modeling on the influent flow data to capture the periodic and trend information in the data. For example, the order of the time series is determined by calculating the Autocorrelation Function (ACF) and the Partial Autocorrelation Function (PACF), so as to select an appropriate ARIMA or SARIMA model. The calculation formula of the Autocorrelation Function (ACF) is:

[0061]

[0062] where ρ k is the autocorrelation coefficient at lag k, γ k is the autocovariance at lag k, γ0 is the zero-order autocovariance, and the Partial Autocorrelation Function (PACF) is used to measure the correlation between two time points after removing the influence of intermediate lag terms.

[0063] After that, the Long Short-Term Memory network is used for prediction output. LSTM is a special Recurrent Neural Network (RNN) that can effectively process data with long-term dependencies. In the prediction of influent flow, the change of influent flow is not only affected by the current moment, but also may be affected by various factors in the past for a long time, such as previous rainfall, sewage discharge habits, etc. Traditional RNNs are prone to problems such as gradient disappearance or gradient explosion when dealing with long sequence data, while LSTM can effectively solve this problem by introducing gating mechanisms, including input gates, forget gates, and output gates, so as to better capture the long-term dependencies in the data.

[0064] Use historical influent flow data and real-time meteorological data as inputs to train the prediction model. The historical influent flow data can be obtained from the operation records of the sewage treatment plant, including the influent flow values at different time points. The real-time meteorological data can be obtained from the meteorological department, such as rainfall, temperature, wind speed, etc. These meteorological data are closely related to the influent flow. For example, an increase in rainfall usually leads to an increase in the influent flow.

[0065] During the training process, the input data is divided into certain time windows. Each time window contains a certain amount of historical data and corresponding meteorological data. The data of these time windows are sequentially input into the LSTM network. Through the backpropagation algorithm, the weights and biases of the network are continuously adjusted to make the output of the network as close as possible to the actual influent flow value. The mean squared error (MSE) is used as the loss function during the training process, and its calculation formula is:

[0066]

[0067] where n is the number of samples, y i is the actual influent flow value, is the influent flow value predicted by the model. By minimizing the MSE, the prediction result of the model is made more accurate. The model is regularly optimized according to newly collected data, and the model parameters are continuously adjusted. As time goes by, the operation situation of the sewage treatment plant and the external environment may change. Therefore, it is necessary to update the model in a timely manner to adapt to these changes. Optimization algorithms such as stochastic gradient descent (SGD) and adaptive moment estimation (Adam) can be used to update the model parameters to improve the accuracy and stability of the model.

[0068] The prediction formula can be expressed as:

[0069]

[0070] where, is the predicted influent flow, Q historical is the historical influent flow data, M weather is the meteorological data, and f is the functional relationship of the model. Here, f is actually the functional relationship represented by the LSTM network. It has learned the mapping relationship between the historical influent flow data, meteorological data and the future influent flow through training.

[0071] Through this formula, it is possible to predict the future peak time and flow rate of the influent flow based on the current historical data and meteorological conditions. In specific applications, the historical influent flow data and real-time meteorological data at the current moment are input into the trained model, and the model will output the predicted influent flow value. By analyzing the prediction results over a period of time, the peak time and flow rate of the influent flow can be determined, providing a decision-making basis for the operation and management of the sewage treatment plant.

[0072] Finally, transfer learning technology is introduced. The model is pre-trained using the historical data of other similar sewage treatment plants, and then fine-tuned on the data of this sewage treatment plant. The historical data of sewage treatment plants similar to this one in terms of other geographical locations, scales, treatment processes, etc. are selected and input into the LSTM network for pre-training. Since the data of these similar sewage treatment plants have certain similarities, during the pre-training process, the model can learn some general features and rules, thereby reducing the training time and cost on the data of this sewage treatment plant. After the pre-training is completed, the historical data and meteorological data of this sewage treatment plant are used as inputs to fine-tune the pre-trained model. During the fine-tuning process, only some parameters of the model need to be adjusted to enable the model to better adapt to the specific situation of this sewage treatment plant. Through transfer learning technology, the generalization ability and prediction accuracy of the model can be improved, especially when the historical data of this sewage treatment plant is scarce, the effect is more obvious.

[0073] S3. Conduct flood season storage and sedimentation control according to the prediction results, including using the PID control algorithm to calculate and determine the sludge discharge time and sludge discharge volume, and optimizing the control through the genetic algorithm;

[0074] During the flood season, the influent flow rate of the sewage treatment plant fluctuates greatly, which may impact the storage tank and the entire sewage treatment system. By implementing refined storage and sedimentation control based on the influent flow rate prediction results, the response ability of the storage tank can be effectively improved, ensuring its safe and stable operation. At the same time, the efficiency and effect of the sedimentation and sludge discharge links can be enhanced, providing good conditions for the subsequent sewage treatment process. When the system judges based on the influent flow rate prediction in the second step that the influent peak is approaching, it will automatically switch to the storage preparation state. At this time, all gates (inlet gate, outlet gate, overflow flushing gate, etc.), valves (sludge discharge valve, etc.) and sludge scraper are set to the closed state. This operation can ensure that the storage tank is in a stable initial state, avoiding problems such as sewage leakage and short circuit caused by abnormal equipment status, and preparing for the upcoming large amount of influent.

[0075] When the influent peak is predicted, the system will open the corresponding inlet gate to let the sewage flow into the storage tank. During this process, the liquid level gauge monitors the liquid level H in the tank in real time. To ensure the safe operation of the storage tank, a liquid level rise rate threshold V is preset max , and the liquid level rise rate is calculated through the liquid level control formula . In practical applications, the discrete time method is usually used for approximate calculation, that is where H t and H t+Δt are the liquid levels at times t and t + Δt respectively.

[0076] If the calculated liquid level rising rate V is greater than the threshold value V max , it indicates that the liquid level is rising too fast, which may threaten the safety of the storage tank. At this time, the system starts the intelligent control algorithm to adjust the opening degree of the inlet sluice. The intelligent control strategy of this embodiment is to adopt the proportional control algorithm. According to the deviation between the liquid level rising rate and the threshold value, the opening degree of the inlet sluice is adjusted proportionally. Let the deviation E = V - V max , and the adjustment amount of the inlet sluice opening degree Δα = K p E, where K p is the proportional coefficient, which is determined according to the characteristics of the storage tank and the actual operation. By adjusting the opening degree of the inlet sluice in real time, the liquid level rising rate is maintained within a safe range.

[0077] During the sedimentation process, the sludge concentration meter measures the sludge concentration C in real time, and the sludge level meter measures the sludge level L in real time. To accurately determine the sludge discharge time and sludge discharge volume, this embodiment adopts an improved adaptive fuzzy-PID composite control algorithm. The basic formula of PID control is:

[0078]

[0079] Among them, u(t) is the control quantity, corresponding to the sludge discharge time or sludge discharge volume, and e(t) is the error. In the sludge discharge control, the sludge concentration deviation E c = C set - C or the sludge level deviation E l = L set - L is used as the error input, C set and L set are the set sludge concentration target value and sludge level target value respectively, and K p , K i and K d are the proportional, integral, and differential coefficients respectively. The fuzzy control enhances the adaptability of the controller by adjusting the PID parameters in real time. Taking the sludge concentration deviation EC c and the deviation change rate as the inputs of the fuzzy controller, E c and EC c are divided into multiple fuzzy subsets, and the corresponding membership functions are defined. This embodiment adopts the triangular membership function. Through fuzzy reasoning and defuzzification methods, the values of K p , K i and K d after real-time adjustment are obtained.

[0080] Finally, the parameters of the fuzzy-PID controller are globally optimized by the genetic algorithm, and K p , K i and K dEncode it as chromosomes and randomly generate an initial population. Define a fitness function, and use the reciprocal of the weighted sum of squares of the sludge concentration deviation and the sludge level deviation over a period of time as the fitness, that is

[0081] where n is the number of sampling points, w1 and w2 are the weight coefficients of the sludge concentration deviation and the sludge level deviation respectively, E C is the sludge concentration deviation, and E l is the sludge level deviation. Through genetic operations such as selection, crossover, and mutation, continuously iterate the population until the maximum number of iterations is reached. The optimal chromosomes finally obtained correspond to K p , K i and K d values, which are the optimized control parameters to improve the accuracy and adaptability of sludge discharge control.

[0082] S4. After precipitation is completed, open the outlet gate for drainage operation, and calculate and control the flushing time according to the volume of the storage tank and the flushing liquid flow rate to complete the precipitation of the current batch;

[0083] After precipitation is completed, the storage tank stores the treated sewage and the sludge deposited at the bottom. Start the high-efficiency drainage and intelligent flushing operations. On the one hand, the treated sewage can be timely transported to the subsequent treatment link of the sewage treatment plant. On the other hand, the residual sludge and impurities in the tank can be removed by flushing to ensure the cleanliness of the storage tank, maintain its normal operation, and create good conditions for the next round of storage and precipitation work. After the precipitation process ends, the system automatically issues an instruction to open the outlet gate. At this time, the sewage in the storage tank, under the action of gravity, is transported to the sewage treatment plant through the outlet channel and enters the subsequent treatment process. During the drainage process, the liquid level sensor continuously monitors the liquid level change in the storage tank and continuously feeds the liquid level data back to the control system. When the liquid level sensor monitors that the liquid level in the storage tank drops to the set minimum liquid level H min , it indicates that most of the sewage in the storage tank has been discharged. At this time, the control system issues an instruction to open the overflow flushing gate. The relatively clear clarified liquid in the upper layer of the adjacent storage tank flows into the storage tank that needs to be flushed under the action of the liquid level difference. Selecting the clarified liquid in the upper layer of the adjacent storage tank as the flushing liquid can not only effectively utilize water resources, but also reduce the dependence on external water sources and lower the operating cost.

[0084] To achieve precise control of the flushing time, based on the volume V of the storage tank, the flushing liquid flow rate Q, and the preset flushing intensity I, use a dynamic adjustment algorithm to calculate the flushing time T. The flushing intensity I represents the total amount of flushing liquid required per unit volume of the storage tank, and it is determined through experiments and experience according to factors such as the actual operating conditions of the storage tank and the degree of sludge accumulation.

[0085] The basic calculation formula for the flushing time is: However, in actual operation, considering various factors that may occur during the flushing process, such as uneven distribution of the flushing liquid, influence of the internal structure of the storage tank, etc., it is necessary to dynamically adjust this formula. For example, by introducing a correction coefficient k, the actual flushing time calculation formula becomes The correction coefficient k is determined by conducting multiple flushing experiments on the storage tank and combining data analysis. Its value range is usually around 1 and is finely adjusted according to different storage tank characteristics and operating conditions.

[0086] After calculating the flushing time, the control system opens the overflow flushing gate according to the set time to flush the storage tank. During the flushing process, the system continuously monitors the liquid level of the storage tank to ensure the normal progress of the flushing operation. When the flushing time reaches the set value and the liquid level in the tank drops to the lowest liquid level again, it indicates that the flushing is completed. At this time, the control system issues an instruction to close the overflow flushing gate and the outlet gate to stop the flushing and drainage operations.

[0087] Subsequently, start the sludge emptying equipment to discharge the remaining sediment in the sludge hopper from the storage tank. The sludge emptying equipment usually includes a sludge pump, sludge pipeline, etc. The sludge pump pumps out the sludge in the sludge hopper and transports it through the sludge pipeline to the sludge treatment system for subsequent treatment and disposal, thereby completing the precipitation treatment process of the current batch.

[0088] S5. Perform stability calculation based on the influent flow rate, and adjust the precipitation time of the next batch using a feedback control algorithm according to the data of the precipitation effect of the previous batch;

[0089] During the sewage treatment process, the stability of the influent flow rate has a significant impact on the treatment effect. An unstable influent flow rate may make it difficult to accurately control the treatment process parameters and affect the effluent quality. By dynamically evaluating the stability of the influent flow rate, when the flow rate is relatively stable, adjusting the precipitation time of the next batch according to the precipitation effect of the previous batch can enable the treatment process to better adapt to the influent situation, improve the treatment efficiency, and achieve goals such as carbon source conservation and hydrolysis acidification.

[0090] Let the real-time monitored influent flow rate data sequence be q1, q2, …, q n , where n is the number of monitoring time points. First, calculate the average value of the influent flow rate within this time period The calculation formula is:

[0091]

[0092] Then calculate the standard deviation σ of the flow rate, and the formula is:

[0093]

[0094] Flow fluctuation coefficient C vIt is defined as the ratio of the standard deviation to the mean, that is:

[0095]

[0096] The flow fluctuation coefficient reflects the degree of fluctuation of the inlet flow within a certain period of time. v The larger the value, the more drastic the flow fluctuation. Conversely, the more stable the flow is. A stability threshold is set based on factors such as the historical operation data of the sewage treatment plant, the characteristics of the treatment process, and the bearing capacity of the equipment. When the calculated flow fluctuation coefficient When the influent flow rate is determined to be relatively stable, data on the previous batch of sedimentation results is collected. This includes effluent quality indicators such as chemical oxygen demand (COD), biochemical oxygen demand (BOD), and suspended solids (SS). These indicators reflect the degree of pollutant removal from wastewater after sedimentation and are important parameters for measuring sedimentation effectiveness.

[0097] Sludge Settling Ratio (SVR): This refers to the volume ratio of settled sludge to mixed liquor after a certain volume of mixed liquor has been allowed to stand in a graduated cylinder for a certain period of time. It reflects the settling performance and cohesiveness of the sludge and is a key indicator for evaluating the settling effect and sludge quality.

[0098] Afterwards, the feedback control algorithm based on reinforcement learning adjusts the sedimentation time, and takes the effluent quality indicators (such as COD, BOD, SS) and sludge settling ratio of the previous batch as state variables to form the state vector: s =

[0099] [COD,BOD,SS,SVR]

[0100] The action is the adjustment amount Δt for the next batch of sedimentation time, which can be set to a set of discrete adjustment values, such as

[0101] {-Δt max ,…,-Δt1,0,Δt1,…,Δt max}

[0102] Where Δt max is the maximum allowable adjustment amount. The reward function R(s,a) is used to measure the treatment effect after taking action a in state s. The reward function can be designed based on the effluent water quality index and sludge settling ratio, for example:

[0103]

[0104] Among them, COD target 、BOD target , SS target and SVR targetThey are respectively the target values of the effluent water quality index and the sludge settling ratio. w1, w2, w3, and w4 are the weight coefficients of each index, and The higher the reward value, the better the treatment effect. The agent selects an action a based on the current state s and applies this action to the next batch of precipitation processes. After the precipitation process ends, the environment returns a new state s' and a reward R(s, a). The agent updates its policy based on the reward value to maximize the long-term cumulative reward. Commonly used reinforcement learning algorithms such as Q-learning or policy gradient algorithms can be used to update the policy. Taking Q-learning as an example, the Q-table Q(s, a) stores the expected cumulative reward for taking action a in state s. The update formula for the Q-table is:

[0105]

[0106] where α is the learning rate, which is used to control the step size for each update of the Q value, γ is the discount factor, s is the current state, the state vector is composed of suspended solids and the sludge settling ratio, a is the adjustment amount of the precipitation time for the next batch, and R(s, a) is the output value of the reward function. is the maximum Q value among all possible actions a' in the new state s'. By continuously interacting with the environment, the agent gradually learns the optimal precipitation time adjustment strategy in different states, thereby realizing the dynamic optimization of the precipitation time and achieving the best treatment effect.

[0107] S6. Set up an overflow storage tank according to the judgment result of the influent flow stability and the operation status of the storage tank, and construct an overflow risk assessment model to evaluate the overflow event.

[0108] The space of the underground sewage treatment plant is relatively closed. Once an overflow or flooding accident occurs, it will not only affect the normal operation of the sewage treatment plant, but may also cause serious consequences such as equipment damage and environmental pollution. Therefore, establishing an intelligent protection and collaborative scheduling system can monitor the operation status in real time, predict the overflow risk in advance, and respond quickly when an accident occurs to ensure the safe and stable operation of the underground plant. Based on the judgment result of the influent flow stability in S5 and the operation status (data such as liquid level and sludge volume) of the storage tank in S3 and S4, adjust the setting of the overflow storage tank. When S5 judges that the influent flow is stable and S3 and S4 feedback that the overall operation of the storage tank is normal, the system will select some storage tanks from many storage tanks as the overflow storage tanks for flood prevention in the underground plant. For the selected overflow storage tanks, their overflow flushing gates remain open to ensure that the overflow water can flow in smoothly; while other related gates such as the influent gate, effluent gate, and sludge discharge valve are in the closed state to prevent the normal inflow and outflow of sewage from interfering with the emergency function of the overflow storage tank.

[0109] Based on deep learning technology, an overflow risk assessment model is established. This model takes meteorological data, predicted influent flow data, and the operating status data of the storage tank as inputs to evaluate the overflow risk in real time. In this embodiment, a multi-layer perceptron is used to construct the overflow risk assessment model:

[0110] Input layer: The input data includes real-time collected meteorological data (such as rainfall, rainfall intensity, etc.), the influent flow data predicted in S2, and the operating status data of the storage tank obtained from the liquid level sensor and pressure sensor. Hidden layer: The hidden layer performs a non-linear transformation on the input data through an activation function to extract the features in the data. The ReLU activation function is used, and its expression is f(x) = max(0, x). In the multi-layer perceptron, the output of each hidden layer serves as the input of the next layer, continuously extracting and abstracting the features of the data. Output layer: The output of the output layer is the evaluation result of the overflow risk, usually represented by a numerical value indicating the level of risk. For example, 0 indicates no overflow risk, and 1 indicates an overflow risk.

[0111] During the model training process, a large amount of historical data, including the actual overflow situation data under different meteorological conditions, influent flow, and the operating status of the storage tank, is used to train the model. The cross-entropy loss function is used to measure the difference between the model prediction result and the actual situation, and its formula is:

[0112]

[0113] where n is the number of samples, y i is the actual overflow situation (0 or 1), is the overflow risk value predicted by the model. When the overflow risk assessment model predicts that an overflow or flooding accident may occur, the automated system quickly responds, automatically opens the preset relevant channels, and guides the overflow water into the overflow storage tank to prevent the overflow water from spreading in the underground plant and ensure the safe operation of the underground plant.

[0114] Embodiment 2

[0115] The difference between this embodiment and Embodiment 1 is that this embodiment provides an intelligent peak-shifting operation system for a buried sewage treatment plant's sectional storage tank, including:

[0116] As Figure 2As shown in the figure, during the flood season, the regulation + sedimentation mode is adopted: when it is judged by the influent flow prediction model that the influent peak is about to arrive, the system enters the initial preparation state, and all gates, valves and sludge scrapers 10 are closed. When the influent peak arrives, the data processing and decision-making layer issues an instruction, and the equipment control layer opens the influent gate 4 of the first storage tank 11, and the sewage enters the first storage tank 11. Among them, the influent channel 5 is located at the front end of the storage tank and is the channel for sewage to enter the storage tank from the influent pipe, playing a role in guiding and distributing sewage. The liquid level gauge 2 monitors the liquid level H in the tank in real time. When the set highest liquid level is reached, the data processing and decision-making layer issues an instruction based on the liquid level data, and the equipment control layer closes the influent gate 4 of the first storage tank 11, opens the influent gate 4 of the second storage tank 11 and the sludge scraper 10 of the first storage tank 11. The second storage tank 11 starts to receive water, and the first storage tank 11 enters the sedimentation mode. During the sedimentation process, the sludge concentration gauge and the sludge level gauge transmit data to the data processing and decision-making layer. The decision-making layer selects the appropriate sludge discharge method and time according to the sedimentation situation, and controls the sludge discharge valve 7 to carry out sludge discharge operation through the sludge discharge pipe 8. The second and third storage tanks 11 are started in this order. If there is no subsequent influent, the fourth storage tank 11 remains idle, reducing the energy consumption of equipment operation. At night, when the influent of the sewage treatment plant reaches the trough, the data processing and decision-making layer issues a drainage instruction, and the equipment control layer opens the effluent gate 3 of the first storage tank 11, and the sewage enters the sewage treatment plant through the effluent channel 6 for treatment. When the liquid level H in the tank reaches the set lowest liquid level, the equipment control layer opens the overflow flushing gate 1 of the second storage tank 11, and uses the clarified liquid on the upper layer of the second storage tank 11 to flush the first storage tank 11. After the flushing is completed and the liquid level in the tank reaches the lowest again, the equipment control layer closes the influent gate 4 of the first storage tank 11 and the sludge scraper 10 of the first storage tank 11, and controls the sludge system to empty the remaining sediment in the sludge hopper 9 in the tank. And so on, the drainage and flushing operations of the second and third storage tanks 11 are completed. After all operations are completed, all equipment and gates are closed, and the system resumes the standby mode.

[0117] Sedimentation mode: On the basis of the flood season regulation + sedimentation mode, the system switches to the sedimentation mode of continuous sequential batch operation. The data processing and decision-making layer adjusts the sedimentation time of the next batch by using a feedback control algorithm based on reinforcement learning according to the sedimentation effect data of the previous batch (such as effluent water quality indicators, sludge sedimentation ratio, etc.). The equipment control layer controls the influent, sedimentation and drainage operations of each storage tank 11 according to the instructions of the decision-making layer, without flushing the tank body, making full use of the storage tank 11 to improve the sedimentation efficiency, and achieving the purposes of saving carbon source, hydrolysis acidification, etc.

[0118] Underground plant flood prevention and soaking mode: When the overflow risk assessment model based on deep learning predicts that an overflow or soaking accident may occur, the system automatically switches to the underground plant flood prevention and soaking mode. The data processing and decision-making layer issues an instruction, and the equipment control layer operates the first, second, and third storage ponds 11 in sequence batch sedimentation mode, and closes the first, second, and third overflow flushing gates 1; closes the inlet gate 4, outlet gate 3, and sludge discharge valve 7 of the fourth storage pond 11, and opens the overflow flushing gate 1 of the fourth storage pond 11. When an overflow or soaking accident occurs in a certain unit of the underground plant, the overflow water passes through the overflow flushing channel and enters the storage pond 11 through the opened overflow flushing gate 1 of the fourth storage pond 11 to ensure the safety of the underground plant. At the same time, the equipment control layer continuously monitors the liquid level of the overflow storage pond (the fourth storage pond 11). When the liquid level approaches the warning liquid level, an alarm is issued through multiple channels to remind the staff to take corresponding measures, and the operation mode of other storage ponds 11 is dynamically adjusted according to the actual situation to ensure the safe and stable operation of the entire underground plant. The above are all preferred embodiments of the present invention, and the protection scope of the present invention is not limited thereby. Therefore, all equivalent changes made according to the structure, shape, and principle of the present invention shall be covered within the protection scope of the present invention.

Claims

1. An intelligent peak-shifting operation method for a sectional storage tank of an underground sewage treatment plant, characterized in that Including: Obtain the operation data of the storage tank, and integrate the obtained operation data, including obtaining the storage inflow data and meteorological data, and associating the two in terms of time; Construct a prediction model based on the fused data, including constructing an influent flow prediction model by using time series analysis combined with long short-term memory network; Conduct flood season storage and sedimentation control according to the prediction results, including calculating and determining the sludge discharge time and sludge discharge volume by using the PID control algorithm, and optimizing the control through the genetic algorithm; After sedimentation is completed, open the effluent gate for drainage operation, calculate and control the flushing time according to the volume of the storage tank and the flushing liquid flow rate, and complete the sedimentation of the current batch; Conduct stability calculation according to the influent flow rate, and adjust the sedimentation time of the next batch by using the feedback control algorithm based on the data of the sedimentation effect of the previous batch; Set an overflow storage tank according to the influent flow rate stability judgment result and the operation state of the storage tank, and construct an overflow risk assessment model to evaluate the overflow event.

2. The intelligent peak-shifting operation method of the sectional storage tank of the buried sewage treatment plant according to claim 1, characterized in that, The integration of the obtained operation data includes, based on the obtained storage inflow data and meteorological data, using a filtering algorithm to remove noise, and performing outlier processing through the Z-score method. Based on the time stamp, associate the storage inflow data and meteorological data at the same moment, establish a linear regression equation for the storage inflow data and meteorological data according to historical data, and determine the relationship between the rainfall threshold, duration and the increase in influent flow rate according to the linear regression equation.

3. The intelligent peak-shifting operation method of the sectional storage tank of the buried sewage treatment plant according to claim 1, characterized in that, The construction of the influent flow prediction model by using time series analysis combined with long short-term memory network includes using the time series analysis method to process the historical influent flow data, extracting statistical features and performing seasonal decomposition. Then, normalize the influent flow data and meteorological data, combine the features extracted by the time series analysis with the original influent flow data and meteorological data as the input of the LSTM network, set multiple LSTM layers, each LSTM layer contains multiple LSTM units, control the information transfer and forgetting through the gating mechanism, add a fully connected layer after the LSTM layer, map the features output by the LSTM layer to a one-dimensional vector, and finally train the water flow prediction model.

4. The intelligent peak-shifting operation method of the sectional storage tank of the in-ground sewage treatment plant according to claim 3, characterized in that, The training of the water flow prediction model includes using the transfer learning method to pre-train the prediction model with the historical influent flow data and meteorological data of other sewage treatment plants, then input the data of this sewage treatment plant into the pre-trained LSTM model for fine-tuning, use the historical influent flow data and real-time meteorological data of this sewage treatment plant as the input, train the fine-tuned prediction model, use the mean square error as the loss function and optimize the model through the Adam algorithm.

5. The intelligent peak-shifting operation method for the sectional storage tank of an in-ground sewage treatment plant according to claim 1, characterized in that, Calculating and determining the sludge discharge time and sludge discharge volume using the PID control algorithm includes making storage preparations, influent control, and sedimentation and sludge discharge control according to the prediction results. The influent control is specifically as follows: when the influent peak is predicted, open the influent gate, calculate the rising rate based on the liquid level in the storage tank, set a rate threshold, and use the proportional control algorithm to adjust the opening degree of the influent gate proportionally according to the deviation between the liquid level rising rate and the threshold. The formula for calculating the opening degree of the influent gate is: Δα = K p E, where E is the deviation between the rising rate and the threshold value, and K p is the proportionality coefficient.

6. The intelligent peak-shifting operation method of the sectional storage tank of the buried sewage treatment plant according to claim 5, characterized in that, The sedimentation and sludge discharge control includes, during the sedimentation process, using the sludge concentration deviation and the sludge level deviation as error inputs, calculating the control quantity using the PID control algorithm, encoding the parameters of the PID control algorithm as chromosomes, randomly generating an initial population, and evaluating the individuals in the population using the fitness function to complete the optimization of the PID control algorithm. The formula for the fitness function is: Among them, n is the number of sampling points, w1 and w2 are the weight coefficients of the sludge concentration deviation and the sludge level deviation respectively, E C is the sludge concentration deviation, and E l is the sludge level deviation.

7. The intelligent peak-shifting operation method of the sectional storage tank of the buried sewage treatment plant according to claim 6, characterized in that Adjusting the sedimentation time of the next batch using the feedback control algorithm based on the data of the sedimentation effect of the previous batch includes obtaining the influent flow data sequence and calculating the average value, obtaining the flow fluctuation coefficient through the ratio of the standard deviation to the average value, setting a stability threshold and judging the stability according to the fluctuation coefficient. When the influent flow is relatively stable, collect the sedimentation effect data of the previous batch to form a state vector, define the reward function, and adjust the sedimentation time of the next batch through the Q-learning algorithm.

8. The intelligent peak-shifting operation method of the sectional storage tank of the buried sewage treatment plant according to claim 7, characterized in that, Adjusting the sedimentation time of the next batch through the Q-learning algorithm includes obtaining the suspended solids and the sludge sedimentation ratio to form a state vector, setting the adjustment amount of the sedimentation time of the next batch as the action, and defining the reward function to construct the Q-learning algorithm formula: Among them, α is the learning rate, which is used to control the step size for each update of the Q value, γ is the discount factor, s is the current state, the state vector is composed of the suspended solids and the sludge sedimentation ratio, a is the adjustment amount of the sedimentation time for the next batch, and R(s,a) is the output value of the reward function. It is the maximum Q value among all possible actions a' taken in the new state s'.

9. The intelligent peak-shifting operation method of the sectional storage tank of the buried sewage treatment plant according to claim 1, characterized in that Constructing an overflow risk assessment model to evaluate overflow events includes using meteorological data, influent flow data, and the liquid level data of the storage tank as inputs, using a multi-layer perceptron to construct an overflow risk assessment model, and the hidden layer performing a non-linear transformation on the input data through the activation function to extract the features in the data, and using the output of the output layer as the evaluation result of the overflow risk.

10. An intelligent peak-shifting operation system for a sectional storage tank of an underground sewage treatment plant, which executes the method described in any one of claims 1-9, characterized in that, Including: Storage tank main module: at least four partitioned storage tanks, each storage tank being a closed space for storing and regulating sewage; Sludge treatment module: sludge hoppers, sludge discharge pipes, and multiple sludge discharge valves. The sludge hoppers are located at the bottom of each storage tank, the sludge discharge pipes are connected to the sludge hoppers, and the sludge discharge valves are installed inside the sludge discharge pipes; Data processing and decision-making module: data acquisition unit, including a liquid level meter data acquisition unit and an external sensor interface, for collecting liquid level, flow rate, and sludge concentration data; Data processing unit, analyzing and processing the collected data; Decision control unit, controlling the actions of the influent gate, effluent gate, overflow flushing gate, sludge discharge valve, and sludge scraper equipment according to the data processing results.

Images