Sewage treatment control method based on AI intelligent driving

Abstract

The invention relates to the technical field of sewage treatment, and particularly discloses a sewage treatment control method based on AI intelligent driving, and the method comprises the following steps: S1, collecting parameters in real time; s2, inputting the parameters into a pre-trained random forest model to obtain an output aeration rate set value, a sludge reflux ratio and a carbon source feeding rate; s3, adjusting the rotating speed of an aeration fan of the aerobic tank according to an aeration rate set value, adjusting the frequency of a sludge reflux pump according to a sludge reflux ratio, and adjusting a metering pump to add a sodium acetate solution according to a carbon source adding rate; and S4, when the turbidity of the MBR membrane pool exceeds a turbidity threshold value, giving out an early warning, further judging whether the transmembrane pressure difference of the MBR membrane pool exceeds a transmembrane pressure difference threshold value or whether the membrane flux attenuation exceeds a membrane flux attenuation threshold value, and if so, starting a pulse backwashing program.

Description

Technical Field

[0001] This invention relates to the field of wastewater treatment technology, and in particular to a wastewater treatment control method based on AI intelligent drive. Background Technology

[0002] With social development, wastewater treatment is a crucial aspect of water supply for local residents and tourism services in mountainous areas. It not only solves environmental problems in mountainous areas but also ensures safe drinking water for local residents. Furthermore, a safe and stable water supply can improve the quality of tourism services and promote the green and healthy development of mountainous areas.

[0003] Mountainous villages and towns are often geographically dispersed, and their sewage quality and quantity vary greatly due to factors such as tourism and temperature, exhibiting obvious seasonal characteristics. This presents a completely different problem compared to urban sewage treatment, making it unsuitable to directly apply urban sewage treatment methods. Specific methods need to be developed to address the unique climatic environment and other factors in mountainous areas. Summary of the Invention

[0004] This invention provides an AI-driven wastewater treatment control method to solve the problems encountered in wastewater treatment in mountainous areas.

[0005] The technical solution adopted in this invention is: An AI-driven wastewater treatment control method is applied to a wastewater treatment system, which includes a water quality equalization tank, an anaerobic tank, an anoxic tank, an aerobic tank, and an MBR membrane tank. The control method includes the following steps: S1. Real-time parameter acquisition, including the instantaneous flow rate of wastewater at the inlet of the water quality equalization tank. Chemical oxygen demand in the water quality equalization tank ammonia nitrogen Total phosphorus pH value and water temperature nitrate concentration at the outlet of the anoxic pool Dissolved oxygen at the outlet of the aerobic tank Turbidity of MBR membrane tank and transmembrane pressure difference ; S2, Instantaneous flow rate of sewage at the inlet of the water quality equalization tank. Chemical oxygen demand in the water quality equalization tank ammonia nitrogen Total phosphorus pH value and water temperature nitrate concentration at the outlet of the anoxic pool Dissolved oxygen at the outlet of the aerobic tank Input the pre-trained random forest model to obtain the output aeration setpoint, sludge return ratio, and carbon source addition acceleration rate; S3. Adjust the speed of the aeration blower in the aerobic tank according to the aeration volume setting value, adjust the frequency of the sludge return pump according to the sludge return ratio, and adjust the metering pump to add sodium acetate solution according to the carbon source addition acceleration rate. S4, When the turbidity of the MBR membrane tank Exceeding the turbidity threshold It issues an early warning and further assesses the transmembrane pressure difference in the MBR membrane tank. Does it exceed the upper limit threshold for transmembrane pressure difference? or membrane flux decline Does it exceed the upper limit threshold for membrane flux decline? If the value exceeds the limit, the pulse backwashing procedure will be initiated.

[0006] Preferably, in step S2, the chemical oxygen demand, ammonia nitrogen, and total phosphorus in the water quality conditioning tank are dynamically compensated for before being input into the random forest model. The formula for calculating the dynamic fluctuation compensation coefficient is as follows: ; in, The standard deviation of the parameter within a sliding window of a preset time is given. This is the average value of the parameter over a sliding window with a preset time interval. This is the terrain compensation coefficient; The chemical oxygen demand, ammonia nitrogen, and total phosphorus in the water quality regulation tank are multiplied by the corresponding dynamic fluctuation compensation coefficients before being input into the random forest model.

[0007] Preferably, in step S2, the chemical oxygen demand (COD), ammonia nitrogen, and total phosphorus in the water quality equalization tank are dynamically compensated for fluctuations, then standardized, and then input into the random forest model. The standardization formula is: ; in, These are the real-time collected values ​​of the parameters. The historical mean of the parameter, The historical standard deviation of the parameter is denoted as .

[0008] Preferably, in step S2, a threshold split is used to control the dissolved oxygen at the outlet of the aerobic tank. When the dissolved oxygen input to the outlet of the aerobic tank reaches a certain threshold, the threshold is adjusted accordingly. Less than the lower limit threshold of dissolved oxygen At that time, 80% of the decision trees in the random forest model chose to increase the aeration rate; a linear weighting method was used for the nitrate concentration at the outlet of the anoxic tank, and when the nitrate concentration at the outlet of the anoxic tank... greater than the upper limit threshold for nitrate concentration At that time, the increment of the carbon source acceleration rate is: ; in, The final carbon source acceleration rate is determined by the sum of the carbon source acceleration rate output by the random forest model and the carbon source acceleration rate increment.

[0009] Preferably, in step S2, each decision tree of the random forest model independently outputs a control suggestion value, and the final output value is a weighted average of the suggestion values ​​of each decision tree, with the weights determined by the historical prediction accuracy of the decision trees.

[0010] Preferably, in step S1, the real-time collected parameters also include the concentration of suspended solids in the mixed liquor at the outlet of the aerobic tank. In step S3, when the concentration of suspended solids in the mixture... Greater than the upper limit threshold of the concentration of suspended solids in the mixture At that time, the preset low sludge return ratio is read. The sludge return ratio output by the random forest model is covered, based on a preset low sludge return ratio. Adjust the sludge return pump frequency when the mixed liquor suspended solids concentration Less than the lower limit threshold of the concentration of suspended solids in the mixture At that time, read the preset high sludge return ratio. The sludge return ratio output by the random forest model is covered, and a high sludge return ratio is preset. Adjust the sludge return pump frequency; when the lower limit threshold of the mixed liquor suspended solids concentration. <Concentration of suspended solids in the mixture> Upper limit threshold for suspended solids concentration in mixed liquid At times, such as the nitrate concentration at the outlet of the anoxic pool. greater than the upper limit threshold for nitrate concentration The sludge return ratio is then adjusted to cover the output of the random forest model according to the following formula: ; For example, the nitrate concentration at the outlet of the anoxic pool. Less than the lower limit threshold of nitrate concentration The sludge return ratio is then adjusted to cover the output of the random forest model according to the following formula: ; in, The adjusted sludge return ratio, This represents the current real-time sludge return ratio.

[0011] Preferably, in step S1, the parameters collected in real time also include dissolved oxygen at the anaerobic tank effluent outlet. In step S2, the voting divergence rate of the random forest output carbon source acceleration rate is calculated. Greater than the preset divergence rate threshold Then, the carbon source input acceleration rate that optimizes the carbon source input acceleration rate coverage of the random forest model output is calculated according to the following formula: ; in, To optimize the carbon source input acceleration rate, For the target C / N ratio, The density of sodium acetate solution, This represents the concentration of the sodium acetate solution.

[0012] The beneficial effects of this invention are: By combining real-time data acquisition with dynamic fluctuation compensation calculation, it effectively addresses the pain point of large water quality fluctuations and improves the system's anti-interference capability; it utilizes a random forest dynamic control model to achieve precise optimization, maintaining the aeration rate within a suitable range to ensure stable dissolved oxygen, and controlling the sludge return ratio to maintain MLSS concentration and reduce energy consumption; it extends membrane life through membrane fouling control mechanisms and reduces maintenance frequency; ultimately, it ensures stable effluent, significantly improves treatment efficiency, and reduces manual operation and maintenance costs, making it particularly suitable for applications in remote mountainous areas. Attached Figure Description

[0013] Figure 1 This is a flowchart illustrating the control method in an embodiment of the present invention; Figure 2 This is a schematic diagram of the dual closed-loop control process for the sludge return ratio in an embodiment of the present invention. Detailed Implementation

[0014] The present invention will now be described in detail with reference to the accompanying drawings and embodiments.

[0015] Example An AI-driven wastewater treatment control method is applied to a wastewater treatment system, which includes a water quality equalization tank, an anaerobic tank, an anoxic tank, an aerobic tank, and an MBR membrane tank, such as... Figure 1 As shown, the control method includes the following steps: S1. Real-time parameter acquisition, including the instantaneous flow rate of wastewater at the inlet of the water quality equalization tank. Chemical oxygen demand in the water quality equalization tank ammonia nitrogen Total phosphorus pH value and water temperature nitrate concentration at the outlet of the anoxic pool Dissolved oxygen at the outlet of the aerobic tank Turbidity of MBR membrane tank Transmembrane pressure difference The concentration of suspended solids in the mixed liquor at the outlet of the aerobic tank. Dissolved oxygen at the effluent outlet of the anaerobic tank .

[0016] S2, Instantaneous flow rate of sewage at the inlet of the water quality equalization tank. Chemical oxygen demand in the water quality equalization tank ammonia nitrogen Total phosphorus pH value and water temperature nitrate concentration at the outlet of the anoxic pool Dissolved oxygen at the outlet of the aerobic tank Input a pre-trained random forest model to obtain the output aeration setpoint, sludge return ratio, and carbon source acceleration rate.

[0017] In this process, the chemical oxygen demand, ammonia nitrogen, and total phosphorus in the water quality regulation tank are first standardized, then dynamically compensated for fluctuations, and finally input into the random forest model.

[0018] The standardized formula is: ; in, These are the real-time collected values ​​of the parameters. The historical mean of the parameter, The historical standard deviation of the parameter is denoted as .

[0019] Standardization eliminates differences in parameter dimensions, making the input data more uniform and facilitating efficient training and inference of the random forest model. Combined with dynamic fluctuation compensation, standardization improves data consistency, reduces the risk of model overfitting, helps achieve accurate optimization, and ensures stable system operation when dealing with fluctuating water quality.

[0020] The formula for calculating the dynamic fluctuation compensation coefficient is as follows: ; in, The standard deviation of the parameter within a sliding window of a preset time is given. This is the average value of the parameter over a sliding window with a preset time interval. This is the terrain compensation coefficient, which can generally be taken as a value; Dynamic fluctuation compensation involves multiplying the chemical oxygen demand, ammonia nitrogen, and total phosphorus in the water quality equalization tank by the corresponding dynamic fluctuation compensation coefficient.

[0021] By compensating for dynamic fluctuations in water quality parameters, the system's adaptability to seasonal water quality changes in mountainous areas is further enhanced, improving the accuracy of model input. Combining real-time data acquisition with dynamic fluctuation compensation calculations effectively addresses the challenge of large water quality fluctuations, improves the system's anti-interference capabilities, and thus optimizes the prediction accuracy of the random forest model while reducing control errors.

[0022] Specifically, a threshold splitting method is used for dissolved oxygen at the aerobic tank outlet. When the dissolved oxygen input to the aerobic tank outlet... Less than the lower limit threshold of dissolved oxygen At that time, 80% of the decision trees in the random forest model chose to increase the aeration rate; a linear weighting method was used for the nitrate concentration at the outlet of the anoxic tank, and when the nitrate concentration at the outlet of the anoxic tank... greater than the upper limit threshold for nitrate concentration At that time, the increment of the carbon source acceleration rate is: ; in, The final carbon source acceleration rate is determined by the sum of the carbon source acceleration rate output by the random forest model and the carbon source acceleration rate increment.

[0023] By introducing threshold splitting for dissolved oxygen and linear weighting for nitrate concentration, the model's response speed to key parameters is enhanced. For example, it can rapidly increase aeration when dissolved oxygen is insufficient or incrementally add carbon source when nitrate levels are excessive. This optimizes the real-time performance of the treatment, reduces energy consumption and reagent waste, maintains the aeration rate within a suitable range to ensure dissolved oxygen stability, and improves carbon source utilization efficiency.

[0024] It is important to note that each decision tree in the random forest model independently outputs a control suggestion value, and the final output value is a weighted average of the suggestion values ​​of each decision tree. The weights are determined by the historical prediction accuracy of the decision trees.

[0025] The weighting mechanism improves the reliability and robustness of the model output, avoiding the bias of a single decision tree. By dynamically adjusting the weights based on historical accuracy, the system can adaptively learn the optimal control strategy, improve prediction accuracy, and achieve precise optimization by using a random forest dynamic control model, reducing control fluctuations and ensuring processing efficiency.

[0026] In addition, when the concentration of suspended solids in the mixture Greater than the upper limit threshold of the concentration of suspended solids in the mixture At that time, the preset low sludge return ratio is read. The sludge return ratio output by the random forest model is covered, based on a preset low sludge return ratio. Adjust the sludge return pump frequency when the mixed liquor suspended solids concentration Less than the lower limit threshold of the concentration of suspended solids in the mixture At that time, read the preset high sludge return ratio. The sludge return ratio output by the random forest model is covered, and a high sludge return ratio is preset. Adjust the sludge return pump frequency; when the lower limit threshold of the mixed liquor suspended solids concentration. <Concentration of suspended solids in the mixture> Upper limit threshold for suspended solids concentration in mixed liquid At times, such as the nitrate concentration at the outlet of the anoxic pool. greater than the upper limit threshold for nitrate concentration The sludge return ratio is then adjusted to cover the output of the random forest model according to the following formula: ; For example, the nitrate concentration at the outlet of the anoxic pool. Less than the lower limit threshold of nitrate concentration The sludge return ratio is then adjusted to cover the output of the random forest model according to the following formula: ; in, The adjusted sludge return ratio, This represents the current real-time sludge return ratio.

[0027] The sludge return ratio is dynamically adjusted based on MLSS and nitrate concentrations. When MLSS exceeds the limit, the preset return ratio is used to cover the model output, or the adjustment value is calculated based on the nitrate concentration within the normal range. This achieves dual closed-loop control of the sludge return ratio, effectively maintaining the MLSS concentration within the ideal range, preventing excessive or insufficient sludge accumulation, reducing energy consumption, and improving the system's adaptability to load changes.

[0028] For example, the voting divergence rate of the carbon source output acceleration rate in random forests. Greater than the preset divergence rate threshold Then, the carbon source input acceleration rate that optimizes the carbon source input acceleration rate coverage of the random forest model output is calculated according to the following formula: ; in, To optimize the carbon source input acceleration rate, For the target C / N ratio, The density of sodium acetate solution, This represents the concentration of the sodium acetate solution.

[0029] When the voting divergence rate in the random forest model exceeds a threshold, a formula is used to optimize the carbon source addition acceleration rate. This provides a backup control strategy when model uncertainty is high, ensuring the reliability of carbon source addition. By calculating optimized values ​​to cover the model output, processing fluctuations caused by model divergence are reduced. Combined with real-time data and dynamic compensation, the system's anti-interference capability is improved, ultimately ensuring stable effluent and reducing maintenance frequency.

[0030] S3. Adjust the speed of the aeration blower in the aerobic tank according to the aeration rate setting, adjust the frequency of the sludge return pump according to the sludge return ratio, and adjust the metering pump to add sodium acetate solution according to the carbon source addition acceleration rate.

[0031] S4, When the turbidity of the MBR membrane tank Exceeding the turbidity threshold It issues an early warning and further assesses the transmembrane pressure difference in the MBR membrane tank. Does it exceed the upper limit threshold for transmembrane pressure difference? or membrane flux decline Does it exceed the upper limit threshold for membrane flux decline? If the value exceeds the limit, the pulse backwashing procedure will be initiated.

[0032] The above method, by combining real-time data acquisition with AI models, effectively addresses the challenge of large fluctuations in wastewater quality in mountainous areas and enhances the system's resilience to interference. Specifically, a random forest model is used to dynamically regulate aeration rate, sludge return ratio, and carbon source addition acceleration rate, ensuring that aeration rate is maintained within a suitable range to stabilize dissolved oxygen. The sludge return ratio is adjusted to maintain MLSS concentration, thereby reducing energy consumption. Simultaneously, membrane fouling control mechanisms (such as turbidity warning and pulse backwashing) extend membrane life and reduce maintenance frequency. Ultimately, this method ensures stable effluent, significantly improves treatment efficiency, and reduces manual operation and maintenance costs, making it particularly suitable for applications in remote mountainous areas.

[0033] The embodiments described above are merely illustrative of specific implementations of the present invention, and while the descriptions are detailed, they should not be construed as limiting the scope of the present invention. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of the present invention, and these modifications and improvements all fall within the scope of protection of the present invention.

Claims

1. A sewage treatment control method based on AI intelligent driving, applied to a sewage treatment system, the sewage treatment system comprising a water quality conditioning tank, an anaerobic tank, an anoxic tank, an aerobic tank and an MBR membrane tank, characterized in that, The control method comprises the following steps: S1, collecting parameters in real time, the parameters including instantaneous flow of sewage at the inlet of the water quality adjusting tank , chemical oxygen demand in the water quality adjusting tank , ammonia nitrogen , total phosphorus , pH value and water temperature , nitrate concentration at the outlet of the anoxic tank , dissolved oxygen at the outlet of the aerobic tank , turbidity of the MBR membrane tank and transmembrane pressure difference ; S2, instantaneous flow of sewage at the inlet of the water quality adjusting tank , chemical oxygen demand in the water quality adjusting tank , ammonia nitrogen , total phosphorus , pH value and water temperature , nitrate concentration at the outlet of the anoxic tank , dissolved oxygen at the outlet of the aerobic tank input a pre-trained random forest model to obtain output aeration setting value, sludge return ratio and carbon source dosage rate; S3, according to the aeration quantity setting value, adjust the rotation speed of the aeration fan of the aerobic tank, according to the sludge return ratio, adjust the frequency of the sludge return pump, and according to the carbon source adding rate, adjust the metering pump to add sodium acetate solution; S4, When the turbidity of the MBR membrane tank Exceeding the turbidity threshold It issues an early warning and further assesses the transmembrane pressure difference in the MBR membrane tank. Does it exceed the upper limit threshold for transmembrane pressure difference? or membrane flux decline Does it exceed the upper limit threshold for membrane flux decline? If the value exceeds the limit, the pulse backwashing procedure will be initiated.

2. The control method according to claim 1, characterized by, In step S2, the chemical oxygen demand, ammonia nitrogen and total phosphorus in the water quality adjusting tank are dynamically compensated, and then input into the random forest model. The calculation formula of the dynamic fluctuation compensation coefficient is: ; wherein, is the standard deviation of the parameter over a sliding window of a preset time, is the average value of the parameter over a sliding window of a preset time, and is a terrain compensation coefficient; The chemical oxygen demand, ammonia nitrogen and total phosphorus in the water quality adjusting tank are multiplied by the corresponding dynamic fluctuation compensation coefficient of the parameters, and then input into the random forest model.

3. The control method according to claim 2, characterized by, In step S2, the chemical oxygen demand, ammonia nitrogen and total phosphorus in the water quality adjusting tank are dynamically compensated, and then standardized and input into the random forest model. The standardization formula is: ; wherein, is a real-time collected value of the parameter, is a historical mean value of the parameter, is a historical standard deviation of the parameter.

4. The control method according to claim 3, characterized by, In step S2, a threshold split is applied to the dissolved oxygen at the aerobic tank outlet. When the dissolved oxygen input to the aerobic tank outlet... Less than the lower limit threshold of dissolved oxygen At that time, 80% of the decision trees in the random forest model chose to increase the aeration rate; the nitrate concentration at the effluent outlet of the anoxic tank was linearly weighted, and when the nitrate concentration at the effluent outlet of the anoxic tank... greater than the upper limit threshold for nitrate concentration At that time, the increment of the carbon source acceleration rate is: ； wherein, For the increment of the carbon source feeding rate, the final carbon source feeding rate is determined by the sum of the carbon source feeding rate output by the random forest model and the increment of the carbon source feeding rate.

5. The control method according to claim 4, characterized by, In step S2, each decision tree of the random forest model independently outputs a control suggestion value, and the final output value is the weighted average of the suggestion values of each decision tree. The weight is determined by the prediction accuracy of the decision tree history.

6. The control method according to claim 5, characterized by In step S1, the real-time collected parameters further include the mixed liquor suspended solids concentration at the effluent of the aerobic tank ; In step S3, when the mixed liquor suspended solids concentration is greater than the mixed liquor suspended solids concentration upper threshold , a preset low sludge return ratio is read , the sludge return ratio output by the random forest model is covered, and the sludge return pump frequency is adjusted according to the preset low sludge return ratio ; when the mixed liquor suspended solids concentration is less than the mixed liquor suspended solids concentration lower threshold , a preset high sludge return ratio is read , the sludge return ratio output by the random forest model is covered, and the sludge return pump frequency is adjusted according to the preset high sludge return ratio . when the mixed liquor suspended solids concentration is less than a lower threshold value < mixed liquor suspended solids concentration < mixed liquor suspended solids concentration is greater than an upper threshold value then the sludge return ratio is calculated to cover the sludge return ratio output by the random forest model according to the following equation:​​ ； As the nitrate concentration at the outlet of the anoxic basin Less than the lower threshold of the nitrate concentration If so, the adjusted sludge return ratio is calculated to cover the sludge return ratio output by the random forest model according to the following formula: ； wherein, is the adjusted sludge recycle ratio, is the current real-time sludge recycle ratio.

7. The control method according to claim 6, characterized by, In step S1, the real-time collected parameters further include dissolved oxygen at the effluent outlet of the anaerobic tank In step S2, when the voting disagreement rate of the random forest output carbon source dosing rate is greater than a preset disagreement rate threshold , the optimized carbon source dosing rate is calculated according to the following formula to cover the carbon source dosing rate output by the random forest model: ​ ； wherein, is the optimized carbon source feeding rate, is the target C / N ratio, is the sodium acetate solution density, is the sodium acetate solution concentration.

Images