A sewage centralized treatment intelligent control scheduling method based on load prediction

Abstract

The application discloses a sewage centralized treatment intelligent control scheduling method based on load prediction, and relates to the technical field of sewage treatment.The method comprises the following steps: S1, collecting sewage pipe network inflow load multi-source data; S2, preprocessing the data and adopting an STL time sequence decomposition algorithm to decompose the data into a trend component, a seasonal component and a residual component; S3, constructing a WOA-ARIMA-LSTM-Transformer hybrid time sequence prediction model, synergistically optimizing model hyperparameters through a whale optimization algorithm, and predicting the inflow load change trend and peak value in the future 1-7 days; S4, based on the prediction result, adopting a hierarchical scheduling strategy to schedule the pump body, the sedimentation tank and the filter tank in advance; and S5, collecting the effluent water quality and the actual inflow load in real time, adopting a PID control correction model parameter when the deviation exceeds 10%, and updating the scheduling strategy to form a closed-loop management and control.The application realizes high-precision prediction of the inflow load and active scheduling of equipment, improves the sewage treatment compliance rate and the operation stability, and reduces energy consumption.

Description

Technical Field

[0001] This invention relates to the field of wastewater treatment technology, specifically to a method for intelligent control and scheduling of centralized wastewater treatment based on load forecasting. Background Technology

[0002] Currently, academia and industry have attempted to introduce time-series forecasting algorithms into the field of wastewater load forecasting. For example, the paper "ARIMA-Based Forecasting of Wastewater Flow Across Short to Long Time Horizons" (Mathematics 2025, 13(13), 2098) discloses a technical solution for predicting wastewater influent flow using a hybrid model of WOA-ARIMA-LSTM-Transformer. This solution uses the Whale Optimization (WOA) algorithm to collaboratively optimize the ARIMA order, the number of LSTM nodes, and the number of Transformer attention heads to improve prediction accuracy. However, this paper only focuses on offline prediction performance and does not consider the "information leakage" problem faced by the model when it is actually deployed online. That is, decomposition operations such as STL are calculated on the full historical data (including the validation set), resulting in falsely low offline validation errors. When rolling online prediction, the prediction accuracy decreases significantly because future data cannot be obtained. The paper "Investigating LSTM-based time series prediction using dynamic systems measures" (Evolving Systems, 2025, 16, 71) further points out that the combination of LSTM with Transformer or decomposition tools performs poorly on highly sensitive and chaotic time series. Wastewater influent load data precisely mixes quasi-periodic patterns (diurnal and seasonal fluctuations) with chaotic sudden disturbances such as industrial illegal discharges and pipeline leaks. Existing hybrid models employ static weighted fusion strategies, which cannot dynamically adjust the model combination and weights according to real-time changes in data characteristics, leading to nonlinear amplification of prediction errors during abnormal events. Furthermore, the paper "Deficiencies of the Whale Optimization Algorithm and its Validation Method" (Expert Systems with Applications, 2023) systematically reveals the inherent center bias defect of the WOA algorithm—when the global optimal solution deviates from the center of the search space, the convergence accuracy and stability of WOA significantly deteriorate. This means that under the engineering constraint that the model needs to be retrained every 10-30 minutes in a wastewater treatment plant, the instability of the WOA optimizer will cause the hyperparameter combinations output by different training rounds to fluctuate drastically, resulting in model performance oscillations. Consequently, the scheduling strategy becomes inconsistent, making it difficult for operators to trust the system.Meanwhile, literature such as "Research on the Application of DeepLearning Models in Time Series Data" (Highlights in Science, Engineering and Technology, 2025) also acknowledges that the hyperparameter tuning process of deep learning models lacks automation and is prone to overfitting. Multi-model collaborative optimization is almost impossible to meet real-time requirements on industrial edge devices with limited computing resources. On the other hand, the application of digital twin technology in wastewater treatment also has limitations. The paper "A multi-pollutants stress testing and resilience analysis using robust offline digital twin forregenerative wastewater treatment" (Journal of Ecological Engineering, 2026) constructed an offline digital twin framework for multi-pollutant stress testing, but experiments found that the framework's prediction fidelity for suspended solids was significantly low, attributed to the difficulty in modeling the stochasticity of hydraulic shear and sedimentation. This indicates that existing digital twin methods generally do not dynamically calibrate simulation fidelity, causing scheduling strategies that pass offline verification to still fail in actual online operation, and lacking "counterfactual" attribution capabilities, making it impossible to quantitatively evaluate the true contribution of scheduling strategies.

[0003] In summary, existing technologies have significant gaps in multiple aspects, such as "leakage-free online decomposition", "adaptive dynamic fusion of operating conditions", "optimizer engineering hardening", and "digital twin fidelity self-calibration". A complete integrated solution for load forecasting and intelligent control scheduling that can operate stably in industrial wastewater scenarios with limited computing resources, dynamic drift of operating conditions, and stringent real-time requirements has not yet been formed. Summary of the Invention

[0004] The purpose of this invention is to provide an intelligent control and scheduling method for centralized sewage treatment based on load forecasting, so as to solve the problems mentioned in the background art.

[0005] To achieve the above objectives, the present invention provides the following technical solution:

[0006] A method for intelligent control and scheduling of centralized wastewater treatment based on load forecasting includes the following steps: S1. Collect multi-source data on the influent load of the sewage pipe network, including sewage flow rate, water quality parameters, environmental parameters and equipment operating parameters; S2. The collected data is preprocessed and decomposed into trend components, seasonal components and residual components using the STL time series decomposition algorithm; S3. Construct a hybrid time-series prediction model: The ARIMA model, LSTM model, and Transformer model are weighted and fused to obtain the initial prediction values. The whale optimization algorithm is used to optimize the order of the ARIMA model, the number of hidden layer nodes of the LSTM model, and the number of attention heads of the Transformer model. The model is trained using the preprocessed data to predict the trend and peak value of wastewater influent load changes in the next 1-7 days. S4. Based on the load forecast results, a hierarchical scheduling strategy is adopted to schedule the pumps, sedimentation tanks and filters in advance, including: adjusting the pump speed according to the predicted load, adjusting the sludge discharge cycle of the sedimentation tank, and pre-setting the backwashing time of the filter. S5. Real-time collection of effluent water quality and actual influent load. When the deviation from the predicted load exceeds 10%, PID control is used to correct the model parameters and update the scheduling strategy to form a closed-loop management and control system.

[0007] Furthermore, in step S1, the time granularity of data acquisition is 5-60 minutes, and the sensor acquisition accuracy meets the following requirements: flow error ≤2%, water quality parameter detection error ≤3%, and equipment operating parameter detection error ≤1%; data aggregation adopts an industrial gateway that supports multiple protocols such as Modbus and OPC UA, and the data transmission delay is ≤5 seconds.

[0008] Furthermore, in step S2, outlier removal adopts the 3σ criterion, missing value filling adopts linear interpolation, and data normalization adopts Min-Max normalization to map to the [0,1] interval; STL time series decomposition adopts additive decomposition, sets the seasonal period, and calculates the trend component, seasonal component and residual component in sequence.

[0009] Furthermore, in step S3, the ARIMA model is used to capture the linear features of the load time series data, the LSTM model is used to capture long-term dependencies, and the Transformer model is used to capture the parallel association of multi-dimensional features. The outputs of the three models are weighted and fused to obtain the initial prediction value. The whale optimization algorithm is used to jointly optimize the order of the ARIMA model, the number of hidden layer nodes of the LSTM model, and the number of attention heads of the Transformer model.

[0010] Furthermore, in step S4, the hierarchical scheduling strategy is as follows: the first layer is pump scheduling, which adjusts the number of pumps starting and stopping and their speed according to the predicted load; the second layer is sedimentation tank scheduling, which adjusts the sludge discharge cycle and the sludge scraper operating speed according to the predicted load; the third layer is filter scheduling, which presets the filter backwashing time and backwashing intensity according to the predicted load and water quality parameters; the scheduling instructions for each layer are executed by the PLC controller.

[0011] Furthermore, in step S5, the feedback regulation adopts a proportional-integral-derivative (PID) control algorithm. When the deviation between the actual influent load and the predicted load exceeds 10%, or the effluent water quality index exceeds the preset standard, the model parameter correction and scheduling strategy optimization are immediately triggered. The corrected model is updated and put into use within 10-30 minutes.

[0012] Furthermore, it also includes a digital twin simulation module: a full-dimensional digital twin model is constructed based on the physical entity of the sewage treatment system. The load forecast results and scheduling strategies are input into the digital twin model for offline simulation, simulating the scheduling effect under six typical operating conditions, including normal production, sudden load changes, and equipment failure, so as to optimize the scheduling parameters in advance.

[0013] A wastewater centralized treatment intelligent control and scheduling system based on load forecasting includes: The data acquisition unit is used to execute step S1; The data preprocessing unit is used to perform step S2; A load forecasting unit is used to perform step S3; The scheduling execution unit is used to execute step S4; Feedback control unit, used to execute right S5; It also includes a digital twin simulation unit, used to build digital twin models for offline simulation verification.

[0014] The present invention also provides an electronic device, which is a physical device, comprising: The processor and the memory are communicatively connected. The memory is used to store at least one executable instruction executed by the processor, which executes the executable instruction to implement the above-described intelligent control and scheduling method for centralized wastewater treatment based on load forecasting.

[0015] The present invention also provides a computer-readable storage medium storing a computer program, which, when executed by a processor, implements the above-described intelligent control and scheduling method for centralized wastewater treatment based on load forecasting.

[0016] Compared with the prior art, the beneficial effects of the present invention are: 1. The STL time series decomposition method is used to decompose the load data into three independent components: trend, seasonality, and residual. The ARIMA linear model, LSTM long-term dependency model, and Transformer parallel correlation model are combined for weighted fusion. Then, the hyperparameters are optimized by the whale optimization algorithm to keep the prediction error within 5%. The prediction can accurately predict the load change trend, peak and abrupt change nodes 1-7 days in advance. This method overcomes the shortcomings of existing technologies, such as the inability of linear models to capture nonlinear dependencies and the difficulty of single deep learning models to accurately capture abrupt change signals.

[0017] 2. Based on high-precision prediction results, a three-layered collaborative scheduling strategy for pumps, sedimentation tanks, and filters was constructed. When a load increase is predicted, the pump speed is increased in advance, the sludge discharge cycle is shortened, and the backwashing time is preset. When a load decrease is predicted, energy consumption is reduced. Compared with the traditional "passive response" mode, this invention achieves "advance prediction and proactive scheduling," effectively avoiding the risk of effluent exceeding standards caused by sudden load changes, and achieving an effluent compliance rate of over 99.5%.

[0018] 3. By precisely scheduling equipment operating parameters, blind full-load operation and frequent equipment start-ups and shutdowns are avoided, resulting in a reduction of more than 20% in power consumption per unit of wastewater treatment, reduced equipment wear, and a significant decrease in the number of unplanned shutdowns.

[0019] 4. A closed-loop management and control mechanism of "collection-prediction-scheduling-feedback" has been constructed. The PID control algorithm is adopted. When the deviation between the actual load and the prediction exceeds 10% or the effluent water quality exceeds the standard, the model parameter correction and scheduling strategy optimization are automatically triggered. The corrected model is updated within 10-30 minutes, enabling the system to adapt to changes in operating conditions and continuously maintain the optimal operating state.

[0020] 5. By constructing a full-dimensional digital twin model, the scheduling strategies under six typical working conditions are optimized through offline simulation. Potential failure risks are identified in advance, reducing the trial and error costs in actual operation and improving the overall reliability and availability of the system. Attached Figure Description

[0021] Figure 1 A flowchart illustrating an intelligent control and scheduling method for centralized wastewater treatment based on load forecasting, provided in an embodiment of the present invention. Figure 2 A schematic diagram of the structure of the hybrid temporal prediction model (WOA-ARIMA-LSTM-Transformer) provided in an embodiment of the present invention; Figure 3 An architecture diagram of the hierarchical intelligent control scheduling mechanism provided in this embodiment of the invention; Figure 4 This is a schematic diagram of digital twin simulation and closed-loop feedback provided in an embodiment of the present invention; Figure 5This is a schematic diagram of the structure of an electronic device provided in an embodiment of the present invention.

[0022] Reference numerals: 21, memory; 22, processor; 23, communication interface; 24, communication bus. Detailed Implementation

[0023] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0024] The method in this embodiment is executed by a terminal, which can be a mobile phone, computer, PDA, laptop or desktop computer, etc. Of course, it can also be other devices with similar functions, and this embodiment does not limit them.

[0025] See Figures 1 to 4 This embodiment provides a method for intelligent control and scheduling of centralized sewage treatment based on load forecasting, which is applied to a centralized urban sewage treatment station with a daily treatment capacity of 20,000 tons.

[0026] Step S1: Data Acquisition At key nodes of the sewage pipe network (three nodes selected in this embodiment), the inlet of the treatment plant, and core treatment units such as pumps, sedimentation tanks, and filters, a total of 20 sensors are deployed, including flow sensors, online COD monitors, ammonia nitrogen monitors, total phosphorus monitors, temperature sensors, humidity sensors, pump speed sensors, sedimentation tank level gauges, sludge scraper operation status sensors, and filter backwash cycle recorders. The data acquisition granularity is set to 15 minutes. Sensor acquisition accuracy requirements: flow error not exceeding 2%, water quality parameter (COD, NH3-N, TP) detection error not exceeding 3%, and equipment operation parameter detection error not exceeding 1%. The collected data is aggregated in real time to the data processing center through an industrial gateway supporting the OPC UA protocol, with a data transmission delay of no more than 5 seconds. The collected parameters include: wastewater flow rate (m³ / h), chemical oxygen demand (COD) (mg / L), ammonia nitrogen (NH3-N) (mg / L), total phosphorus (TP) (mg / L), ambient temperature (°C), ambient humidity (%RH), pump speed (r / min), sedimentation tank level (m), sludge scraper operating speed (m / min), and filter backwashing cycle (h).

[0027] Step S2: Data Preprocessing Preprocessing of the multi-source heterogeneous data collected in step S1 specifically includes: Outliers were removed using the 3σ criterion: For each parameter, its mean and standard deviation were calculated, and data that exceeded the mean ± 3 times the standard deviation were considered outliers and removed.

[0028] Missing values ​​are filled using linear interpolation: For missing values ​​between two consecutive valid data points, interpolation is performed according to the linear relationship over time. In this embodiment, the missing value rate is controlled within 5%.

[0029] Min-Max normalization is used to map all data to the [0,1] interval to eliminate the influence of units.

[0030] Subsequently, the Seasonal-Trend Decomposition (STL) algorithm using LOESS was employed to decompose the preprocessed influent load time series data. The seasonality period was set to 24 hours (corresponding to diurnal load fluctuations). In this embodiment, other key parameters for STL decomposition were set as follows: the trend smoothing window length was set to 169 observation points (corresponding to 7 days × 24 hours / 15 minutes), the seasonal smoothing window length was set to 25 observation points, the low-pass filter window length was set to 61 observation points, and the robustness iteration count was set to 5. These parameters were optimized and determined on the validation set through grid search. An additive decomposition model was adopted, the basic form of which is:

[0031] in, Let t be the original influent load value. As a trend component, it reflects the long-term changing pattern of the load (annual growth or decline trend). It is a seasonal component that reflects the periodic fluctuations in load (daytime peaks and valleys, weekday / weekend differences). The residual components reflect random load fluctuations and abnormal impact signals such as rainstorm runoff and industrial illegal discharges. The time-series features of each component are extracted to form a standardized prediction input dataset.

[0032] Step S3: Load Forecasting Construct a hybrid time series prediction model of WOA-ARIMA-LSTM-Transformer, the specific structure of which is as follows: Figure 2 As shown.

[0033] The ARIMA model is used to capture the linear characteristics of load time series data. In this embodiment, the initial order is set to (2,1,2).

[0034] The LSTM model is used to capture long-term dependencies in load time-series data. The forward propagation process of its core unit is described by the following set of formulas:

[0035] in, Enter the current time. This is the hidden state from the previous moment. , , These are the forget gate, input gate, and output gate, respectively. In cellular state, It is the Sigmoid activation function. This is an element-wise multiplication. In this embodiment, the number of nodes in the LSTM hidden layer is initially set to 64.

[0036] The Transformer model captures the parallel associations of multi-dimensional features through a multi-head self-attention mechanism. Given a query matrix... Key matrix Sum matrix The formula for calculating scaled dot product attention is:

[0037] in, Let be the dimension of the key vector. Multi-head attention concatenates the outputs of multiple attention heads and then performs a linear transformation. In this embodiment, the number of Transformer attention heads is initially set to 4.

[0038] The outputs of the three are weighted and fused to obtain the initial predicted value, with the weighting coefficients initially set to 1 / 3.

[0039] The Whale Optimization (WOA) algorithm is used to jointly optimize the order (p, d, q) of the ARIMA model, the number of hidden layer nodes in the LSTM model, and the number of attention heads in the Transformer model. The prediction input dataset obtained in step S2 is divided into training and test sets in a 7:3 ratio. Supervised training is performed on the training set using the mean squared error (MSE) loss function, and the model parameters are adjusted with the goal of minimizing the loss function. The trained model is validated using the test set, and the model parameters are adjusted until the prediction error does not exceed 5%. In this embodiment, the optimized model has a prediction error of 4.2%.

[0040] The preprocessed data, updated in real time, is input into the optimized model to predict the trend of wastewater influent load changes, load peaks, and abrupt change points over the next 7 days. In this embodiment, the load peak prediction error is ≤4%, and the early warning time for abrupt change points is ≥2 hours.

[0041] Step S4: Intelligent Control and Scheduling Based on the load forecast results from step S3, and combined with GB18918-2002 "Pollutant Discharge Standard for Urban Wastewater Treatment Plants" and equipment operating thresholds, a multi-equipment collaborative intelligent control and scheduling strategy is constructed, employing a hierarchical scheduling mechanism (such as...). Figure 3(As shown).

[0042] First layer: Pump scheduling: Adjust the number of pumps started and stopped, and their speed, based on predicted load changes. When an increase in influent load is predicted (e.g., the load is predicted to increase from 1800 m³ / h to 2200 m³ / h during the morning peak from 7:00 to 9:00 the next day), start the standby pumps one hour in advance and increase the operating pump speed from 800 r / min to 900 r / min to ensure stable influent flow. When a decrease in load is predicted (e.g., the load is predicted to drop to 1200 m³ / h during the night from 0:00 to 6:00), shut down one standby pump 30 minutes in advance and reduce the operating pump speed to 600 r / min to reduce energy consumption.

[0043] Second layer: Sedimentation tank scheduling: Adjust the sludge discharge cycle and scraper speed of the sedimentation tank according to the predicted load and water quality parameters. When the predicted load increases, shorten the sludge discharge cycle from 4 hours to 3 hours, and increase the scraper speed from 1.2 m / min to 1.5 m / min, controlling the sludge concentration in the sedimentation tank at 2000-3000 mg / L. When the predicted load decreases, extend the sludge discharge cycle to 5 hours, and reduce the scraper speed to 1.0 m / min.

[0044] Third layer: Filter scheduling: Based on the predicted load and effluent water quality requirements, preset the filter backwash time and intensity. When the predicted influent COD concentration is expected to rise to 500 mg / L, schedule one set of filters for backwashing 2 hours in advance, with a backwash time of 15 minutes and a backwash intensity of 12 L / (m²·s). When the predicted load is reduced and the COD concentration is below 300 mg / L, extend the filter backwash cycle from 24 hours to 36 hours.

[0045] The scheduling instructions at each level are executed precisely by the PLC controller, with an execution accuracy rate of ≥99%. The execution accuracy rate is defined as the percentage of times the PLC controller correctly receives, parses, and drives the corresponding equipment (pump, scraper, backwash valve) to complete the expected action out of 1000 consecutive scheduling instructions issued. Field testing showed that this embodiment achieved an execution accuracy rate of 99.3% (993 successful out of 1000 attempts, with 7 failures due to momentary communication interruptions). For load surge warnings, such as those predicting a sudden increase of 100% in load, an emergency scheduling mode is triggered, rapidly adjusting the operating parameters of each device within 5 minutes to ensure continuous and stable processing.

[0046] Step S5: Feedback Adjustment The system collects real-time data on effluent quality indicators (COD, NH3-N, TP), operational status of each device, and actual influent load from the water treatment plant. This data is then compared with the predicted load and preset treatment standards to calculate the deviation. A proportional-integral-derivative (PID) control algorithm is employed, and its discrete-form control law is as follows:

[0047] in, This refers to the control output at the current moment (pump speed adjustment and sludge discharge cycle adjustment). This represents the deviation between the actual value and the set value at the current moment (the difference between the actual influent load and the predicted load, or the difference between the effluent quality and the discharge standard). , , These are the proportional, integral, and differential coefficients, respectively. For the sampling time interval, the PID controller parameters in this embodiment are tuned as follows: proportional coefficient K_p = 0.6, integral coefficient K_i = 0.1, derivative coefficient K_d = 0.05, and sampling period Δt = 5 seconds. These parameters were initially tuned based on the Ziegler-Nichols method and then fine-tuned on-site. It should be noted that this embodiment uses three independent PID controllers for pump speed, sludge discharge cycle, and filter backwashing time, each calculating its control input independently based on its corresponding deviation. For simplicity, u(k) in the formula represents the output of any controller. When the actual influent load deviates from the predicted load by more than 10%, or when the effluent quality indicators exceed the preset standards (COD≤50mg / L, NH3-N≤5mg / L, TP≤0.5mg / L), model parameter correction and scheduling strategy optimization are immediately triggered: on the one hand, the hyperparameters of the hybrid time-series prediction model (ARIMA order, LSTM node count, Transformer attention head count) are corrected to improve prediction accuracy; on the other hand, equipment scheduling parameters are adjusted, such as increasing pump speed to 950r / min and shortening the sedimentation tank sludge removal cycle to 2.5 hours. The corrected model is updated and put into use within 10-30 minutes, forming a closed-loop management system of "collection-prediction-scheduling-feedback" (e.g., data collection-prediction-scheduling-feedback). Figure 4 (As shown).

[0048] Digital twin simulation module This embodiment also includes a digital twin simulation module. The construction method of this module is as follows: (I) Modeling Method: A hybrid modeling strategy was adopted to construct a full-dimensional digital twin model based on the physical entities of the wastewater treatment system. For the mechanism part, based on the ASM2d activated sludge model and fluid dynamics equations, mechanism models were established for the influent pumping station, grit chamber, biological treatment tank, secondary sedimentation tank, filter, and dosing system respectively. For the data-driven part, an LSTM network was used to compensate for the prediction residuals of the mechanism model to improve simulation fidelity. The above mechanism-data fusion model was integrated into the Unity3D simulation platform, with a simulation step size set to 1 minute.

[0049] (II) Definition of Typical Operating Conditions: The following six types of typical operating conditions will be simulated, with each type of operating condition simulated continuously for 72 hours: ① Normal daily load fluctuation condition: The influent load fluctuates randomly within ±20% of the daily average value; ② Heavy rain runoff causing sudden load increase: The load suddenly increases from the baseline value to 150% within 1 hour, and recovers after 2 hours; ③ Industrial illegal discharge causing a sudden increase in COD: COD concentration increases from 300 mg / L to 600 mg / L within 30 minutes; ④ Inlet pump failure condition: One main pump stops, the flow rate decreases by 30%, and this lasts for 4 hours; ⑤ Filter clogging condition: The filter backwash valve is stuck, and the backwash cycle is extended from 24 hours to 48 hours; ⑥ Sludge scraper jamming in sedimentation tank: The scraper speed is reduced to 50% of the normal value for 6 hours.

[0050] (III) Fidelity Verification and Parameter Optimization: Historical actual operating data was compared with the digital twin simulation results, and the fidelity was evaluated using the mean absolute percentage error. Verification showed that the digital twin model in this embodiment achieved a prediction fidelity of ≥95% for effluent COD, NH3-N, and TP. Load prediction results and scheduling strategies were input into the digital twin model, and offline simulations were performed under the aforementioned six operating conditions. Based on the effluent quality curves and equipment energy consumption data output from the simulation, scheduling parameters (such as pump speed adjustment, sludge discharge cycle advance, and filter backwash trigger threshold) were dynamically adjusted until the effluent quality met standards and unit power consumption was minimized. The optimized scheduling parameters were then written into the actual PLC control program.

[0051] Through the aforementioned digital twin offline simulation, the risks of excessively high water levels due to insufficient pump speed adjustment under operating condition ④ and the problem of increased filter clogging after prolonged backwashing cycles under operating condition ⑤ were identified in advance. The scheduling thresholds were adjusted accordingly, reducing the trial-and-error costs in actual operation. After one month of trial operation, the effluent quality compliance rate in this embodiment increased from 97.2% to 99.6%, the unit wastewater treatment power consumption decreased by 25.3%, and the number of unplanned equipment shutdowns decreased from 1-2 times per month to 0 times, significantly improving operational stability.

[0052] Example 2 The difference between this embodiment and Embodiment 1 is that the method is applied to a centralized industrial wastewater treatment plant with a daily processing capacity of 50,000 tons, and the data acquisition time granularity is set to 5 minutes. Furthermore, the prediction model optimization uses a particle swarm optimization algorithm instead of a whale optimization algorithm (it should be noted that the particle swarm optimization algorithm is only an alternative implementation in this embodiment and is not a limitation of the invention. The whale optimization algorithm defined in the claims is still the preferred solution). The remaining steps are basically the same as in Embodiment 1; only the differences are described below.

[0053] In step S1, the data acquisition time granularity is set to 5 minutes, and the number of sensors is increased to 35 to meet the requirements of higher precision real-time monitoring.

[0054] In step S3, the same WOA-ARIMA-LSTM-Transformer hybrid time-series prediction model is constructed, but the hyperparameter optimization uses the particle swarm optimization algorithm. After training and testing, the model's prediction error is controlled within 4.8%, meeting the requirement of ≤5%.

[0055] In step S4, in addition to the tiered scheduling of the pumps, sedimentation tank, and filter, the scheduling of the aeration system is also added: the aeration intensity is adjusted according to the predicted load. When the predicted load increases, the blower operating frequency is increased to increase the aeration volume and maintain the dissolved oxygen concentration at 2.0-3.0 mg / L; when the predicted load decreases, the blower frequency is appropriately reduced to reduce power consumption. The scheduling instructions are also executed through the PLC controller.

[0056] After trial operation, this embodiment showed that the power consumption for COD removal was reduced by 30.1%, the dissolved oxygen control accuracy was improved by 62.5%, and the effluent quality compliance rate was maintained at over 99.5%.

[0057] Example 3 This embodiment further optimizes the feedback adjustment mechanism based on Embodiment 1. In step S5, in addition to using the PID control algorithm, an adaptive threshold adjustment strategy is introduced. Specifically, a sliding time window (window length of 24 hours) is set to statistically analyze the deviation distribution between the actual influent load and the predicted load over the past 24 hours, and dynamically update the deviation threshold that triggers correction. When the deviation exceeds the current dynamic threshold (initially 10%), model parameter correction is triggered. Simultaneously, the model update cycle is adaptively adjusted according to the deviation change rate: when the deviation increases rapidly (change rate > 5% / h), the model update cycle is shortened to 10 minutes; when the deviation tends to stabilize, the update cycle is extended to 30 minutes. The other steps in this embodiment are the same as in Embodiment 1. Through adaptive threshold and variable cycle updates, unnecessary model retraining times are further reduced, computational resource consumption is lowered, and the effluent water quality compliance rate is maintained at ≥99.7%.

[0058] The above description is merely a preferred embodiment of the present invention and is not intended to limit the invention. Those skilled in the art can make various improvements and modifications without departing from the spirit and principle of the invention, and these improvements and modifications should also be considered within the scope of protection of the present invention. An optional embodiment provides a wastewater centralized treatment intelligent control and scheduling system based on load forecasting, used to implement the method described in Embodiment 1. The system includes: The data acquisition unit consists of sensor arrays deployed at key nodes of the wastewater pipeline network, treatment plant inlets, and core treatment units. These include flow sensors, online COD monitors, ammonia nitrogen monitors, total phosphorus monitors, temperature sensors, humidity sensors, pump speed sensors, sedimentation tank level gauges, sludge scraper operation status sensors, and filter backwash cycle recorders. The acquisition time granularity for each sensor is set to 15 minutes, with acquisition accuracy meeting the requirements of flow error ≤2%, water quality parameter detection error ≤3%, and equipment operation parameter detection error ≤1%. This unit aggregates the acquired data to the data processing center in real time via an industrial gateway supporting the OPC UA protocol, with a data transmission delay ≤5 seconds.

[0059] Data preprocessing unit: Deployed on the server in the data processing center, it includes a data cleaning module (using the 3σ criterion to remove outliers), a missing value imputation module (using linear interpolation), and a data normalization module (using Min-Max normalization to map to the [0,1] interval). This unit also includes an STL time series decomposition module, which sets the seasonal period to 24 hours and decomposes the inflow load time series data into trend components, seasonal components, and residual components to form a standardized prediction input dataset.

[0060] Load forecasting unit: Deployed on a server, it integrates a hybrid time-series forecasting model of WOA-ARIMA-LSTM-Transformer. The ARIMA module captures linear load characteristics (order (2,1,2)), the LSTM module captures long-term dependencies (64 hidden layer nodes), and the Transformer module captures multi-dimensional parallel associations (4 attention heads). The outputs of these three modules are weighted and fused to obtain the predicted value. This unit also integrates a Whale Optimization (WOA) module for collaborative optimization of the ARIMA order, LSTM node count, and Transformer attention head count. The forecasting unit receives standardized data from the data preprocessing unit and outputs the wastewater influent load change trend, peak values, and abrupt change points for the next 1-7 days, with a prediction error ≤5%.

[0061] The scheduling and execution unit includes a PLC controller and controlled equipment. The PLC controller receives the load forecasting results from the load forecasting unit and generates and issues commands according to a hierarchical scheduling strategy: the first level controls the pumps (frequency converter adjusts speed and starts / stops), the second level controls the sedimentation tank (sludge discharge cycle, scraper speed), and the third level controls the filter (backwashing time, backwashing intensity). The PLC is connected to each equipment actuator via an industrial Ethernet network, achieving an execution accuracy of ≥99%.

[0062] Feedback control unit: Deployed on the server, it includes a data acquisition interface (for real-time acquisition of effluent water quality indicators, equipment operating status, and actual influent load), a deviation calculation module (for comparison with predicted values), and a PID controller. The PID controller operates according to discrete control laws. Calculate the correction amount. When the deviation exceeds 10% or the effluent water quality exceeds the standard, the hyperparameter correction of the load prediction unit and the scheduling strategy update of the scheduling execution unit are triggered. The corrected model is updated and put into use within 10-30 minutes.

[0063] Digital Twin Simulation Unit: Deployed on a high-performance server, this unit employs a hybrid modeling strategy (mechanistic model based on ASM2d, data-driven model based on LSTM) to construct a full-dimensional digital twin model, covering facility layout, process flow, equipment status, and water quality and quantity scenarios. This unit receives load forecast results and scheduling strategies, and offline simulates the following six typical operating conditions: ① Normal daily load fluctuation; ② Sudden 50% increase in storm runoff load; ③ Instantaneous 100% increase in COD; ④ 30% load reduction due to influent pump failure; ⑤ Filter clogging; ⑥ Sludge scraper jamming. Each condition is simulated continuously for 72 hours, with a simulation step size of 1 minute. Simulation fidelity is ≥95%. This unit outputs a simulation evaluation report (including effluent water quality curves, energy consumption data, and equipment action response times) and feeds it back to the scheduling execution unit to optimize scheduling parameters.

[0064] The above units communicate with each other via industrial Ethernet or internal data bus to form an integrated closed-loop system of "acquisition-prediction-scheduling-feedback-simulation".

[0065] Based on the above embodiments, such as Figure 5 As shown, the present invention also provides an electronic device, the electronic device comprising: The processor 22 includes at least one processor 22, at least one memory 21, a communication interface 23, and a communication bus 24, wherein the processor 22 is communicatively connected to the memory 21. In this embodiment, the memory 21 can be implemented in any suitable manner, for example, the memory 21 can be a read-only memory, a hard disk drive, a solid-state drive, or a USB flash drive, etc.; the memory 21 is used to store at least one executable instruction executed by the processor; In this embodiment, the processor 22 can be implemented in any suitable manner. For example, the processor 22 can take the form of a microprocessor or processor and a computer-readable medium storing computer-readable program code (e.g., software or firmware) that can be executed by the (micro)processor, logic gates, switches, application-specific integrated circuits (ASICs), programmable logic controllers, and embedded microcontrollers, etc.; the processor is used to execute the executable instructions to implement a wastewater centralized treatment intelligent control and scheduling method based on load forecasting as described above.

[0066] Based on the above embodiments, the present invention also provides a computer-readable storage medium storing a computer program, which, when executed by a processor, implements the above-described intelligent control and scheduling method for centralized sewage treatment based on load forecasting.

[0067] Those skilled in the art will recognize that the modules and method steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementations should not be considered beyond the scope of this invention.

[0068] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working processes of the devices, equipment, and modules described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0069] In addition, the functional modules in the various embodiments of the present invention can be integrated into one processing module, or each module can exist physically separately, or two or more modules can be integrated into one module.

[0070] If the aforementioned functions are implemented as software functional modules and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this invention, or the part that contributes to the prior art, or a portion of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this invention. The aforementioned storage medium includes various media capable of storing program instructions, such as USB flash drives, portable hard drives, read-only storage servers, random access storage servers, magnetic disks, or optical disks.

[0071] Furthermore, it should be noted that the combination of the various technical features in this case is not limited to the combination methods described in the claims of this case or the combination methods described in the specific embodiments. All technical features described in this case can be freely combined or combined in any way, unless they contradict each other.

[0072] It should be noted that the above examples are merely specific embodiments of the present invention, and the present invention is obviously not limited to the above embodiments, with many similar variations. All modifications that can be directly derived or conceived by those skilled in the art from the content disclosed in this invention should fall within the protection scope of this invention.

[0073] The above are merely preferred embodiments of the present invention and are not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for intelligent control and scheduling of centralized wastewater treatment based on load forecasting, characterized in that, Includes the following steps: S1. Collect multi-source data on the influent load of the sewage pipe network, including sewage flow rate, water quality parameters, environmental parameters and equipment operating parameters; S2. The collected data is preprocessed and decomposed into trend components, seasonal components and residual components using the STL time series decomposition algorithm; S3. Construct a hybrid time-series prediction model: The ARIMA model, LSTM model, and Transformer model are weighted and fused to obtain the initial prediction values. The whale optimization algorithm is used to optimize the order of the ARIMA model, the number of hidden layer nodes of the LSTM model, and the number of attention heads of the Transformer model. The model is trained using the preprocessed data to predict the trend and peak value of wastewater influent load changes in the next 1-7 days. S4. Based on the load forecast results, a hierarchical scheduling strategy is adopted to schedule the pumps, sedimentation tanks and filters in advance, including: adjusting the pump speed according to the predicted load, adjusting the sludge discharge cycle of the sedimentation tank, and pre-setting the backwashing time of the filter. S5. Real-time collection of effluent water quality and actual influent load. When the deviation from the predicted load exceeds 10%, PID control is used to correct the model parameters and update the scheduling strategy to form a closed-loop management and control system.

2. The intelligent control and scheduling method for centralized sewage treatment based on load forecasting according to claim 1, characterized in that, In step S1, the time granularity of data acquisition is 5-60 minutes, and the sensor acquisition accuracy meets the following requirements: flow rate error ≤2%, water quality parameter detection error ≤3%, and equipment operating parameter detection error ≤1%. Data aggregation uses an industrial gateway that supports multiple protocols such as Modbus and OPC UA, with data transmission latency of ≤5 seconds.

3. The intelligent control and scheduling method for centralized sewage treatment based on load forecasting according to claim 1, characterized in that, In step S2, outlier removal uses the 3σ criterion, missing value filling uses linear interpolation, and data normalization uses Min-Max normalization to map to the [0,1] interval; STL time series decomposition uses additive decomposition, sets a seasonal period, and calculates the trend component, seasonal component, and residual component in sequence.

4. The intelligent control and scheduling method for centralized sewage treatment based on load forecasting according to claim 1, characterized in that, In step S3, the ARIMA model is used to capture the linear features of the load time series data, the LSTM model is used to capture long-term dependencies, and the Transformer model is used to capture the parallel association of multi-dimensional features. The outputs of the three models are weighted and fused to obtain the initial prediction value. The whale optimization algorithm is used to optimize the order of the ARIMA model, the number of hidden layer nodes of the LSTM model, and the number of attention heads of the Transformer model in a coordinated manner.

5. The intelligent control and scheduling method for centralized sewage treatment based on load forecasting according to claim 1, characterized in that, In step S4, the hierarchical scheduling strategy is as follows: the first layer is pump scheduling, which adjusts the number of pumps starting and stopping and their speed according to the predicted load; the second layer is sedimentation tank scheduling, which adjusts the sludge discharge cycle and the sludge scraper operating speed according to the predicted load; the third layer is filter scheduling, which presets the filter backwashing time and backwashing intensity according to the predicted load and water quality parameters; the scheduling instructions for each layer are executed by the PLC controller.

6. The intelligent control and scheduling method for centralized sewage treatment based on load forecasting according to claim 1, characterized in that, In step S5, the feedback regulation adopts the proportional-integral-derivative (PID) control algorithm. When the deviation between the actual influent load and the predicted load exceeds 10%, or the effluent water quality index exceeds the preset standard, the model parameter correction and scheduling strategy optimization are immediately triggered. The corrected model is updated and put into use within 10-30 minutes.

7. The intelligent control and scheduling method for centralized sewage treatment based on load forecasting according to claim 1, characterized in that, It also includes a digital twin simulation module: a full-dimensional digital twin model is built based on the physical entity of the sewage treatment system. The load forecast results and scheduling strategies are input into the digital twin model for offline simulation, simulating the scheduling effect under six typical operating conditions, including normal production, load changes, and equipment failure, so as to optimize the scheduling parameters in advance.

8. An electronic device, characterized in that, The electronic device includes a processor (22) and a memory (21), wherein the memory (21) is communicatively connected to the processor (22); The memory (21) is used to store at least one executable instruction executed by the processor (22), the processor (22) being used to execute the executable instruction to implement the intelligent control and scheduling method for centralized sewage treatment based on load forecasting as described in any one of claims 1 to 7.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the program is executed by the processor, it implements the intelligent control and scheduling method for centralized sewage treatment based on load prediction as described in any one of claims 1 to 7.

10. A centralized wastewater treatment intelligent control and scheduling system based on load forecasting, characterized in that, include: The data acquisition unit is used to perform step S1 as described in claim 1; The data preprocessing unit is used to perform step S2 as described in claim 1; A load forecasting unit, configured to perform step S3 as described in claim 1; A scheduling execution unit is used to execute step S4 as described in claim 1; A feedback control unit is used to perform step S5 as described in claim 1; It also includes a digital twin simulation unit, used to build digital twin models for offline simulation verification.

Images