Dynamic multi-objective optimization control method for sewage treatment process

Abstract

The invention relates to the technical field of sewage treatment, and discloses a dynamic multi-objective optimization control method for a sewage treatment process, which comprises the following steps: step 1, carrying out bioenergy collection on effluent of an adjusting tank through a double-cavity microbial fuel cell, and supplying output voltage to a multi-parameter sensor and an edge controller through a conversion circuit; step 2, a multi-parameter sensor collects sewage COD value, ammonia nitrogen concentration, dissolved oxygen content and oxidation reduction potential data by taking 5 min as a cycle, and transmits the data to an edge controller; and 3, the edge controller divides three working condition modes of low load, conventional load and impact load according to the ratio of the real-time water inlet flow to the design flow. The technical scheme of working condition mode dynamic division and reinforcement learning control instruction generation is adopted, the technical effect that the aeration intensity and the carbon source dosage are adaptively adjusted along with the water inlet load is achieved, and compared with a fixed period aeration and artificial experience dosing mode in the prior art, the defects that control response is lagged and energy consumption and chemical consumption are virtually high are overcome.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of sewage treatment, in particular to a dynamic multi-objective optimization control method for a sewage treatment process. BACKGROUND

[0002] Due to the development of China's economic society and the improvement of people's living standards, water resources consumption and sewage discharge are increasing. In order to reduce the impact of water pollution, the growth of the national economy and the enhancement of people's environmental awareness have brought new opportunities for the development of sewage treatment process automation technology. However, the sewage treatment process is a complex dynamic system, with long biochemical reaction period, complex pollutant composition, real-time changes in water inflow and water composition, aeration energy consumption, pumping energy consumption, and mutual coupling and influence between water quality.

[0003] For example, the Chinese invention application with publication number CN119717523A discloses a dynamic multi-objective optimization control method for a sewage treatment process based on probability and dimension joint prediction, relating to the technical field of sewage treatment. It includes: obtaining process data of the sewage treatment process, wherein; constructing an optimization objective function according to a support vector regression-based predictor and the process data; and using a dynamic multi-objective optimization algorithm based on probability and dimension joint prediction and a fast elitist multi-objective genetic algorithm to solve the objective function to obtain a set of Pareto optimal solutions.

[0004] For example, the Chinese invention application with publication number CN118778522A discloses a dynamic multi-objective optimization control method and system for a sewage treatment process. First, real-time data in the sewage treatment process are collected, and a self-adaptive fuzzy neural network is used to establish an optimization objective model for water quality and energy consumption; then, a multi-objective optimization method is used to optimize and solve the optimization solution of nitrate nitrogen concentration and dissolved oxygen concentration, and an environmental detection operator is used to detect whether the environment has changed.

[0005] The above-mentioned patents have the following shortcomings:

[0006] 1. The existing control method uses fixed cycle aeration and manual experience dosing mode, and cannot establish a dynamic coupling mechanism with water inflow fluctuation and water quality concentration change, resulting in excessive aeration energy consumption and reagent dosage under low load conditions, and out-of-specification water indicators due to control response lag during high load periods, which cannot adapt to the characteristics of significant day-night differences and significant seasonal fluctuations in rural sewage discharge.

[0007] 2. The traditional equipment lacks remote state monitoring and intelligent fault tolerance mechanism, and cannot automatically switch to a safe operation mode and trigger a remote alarm when the aeration device is abnormal or the sensor fails suddenly, which increases the risk of unplanned system shutdown, and the fault recovery period is too long due to the response time of rural area maintenance personnel, which easily causes secondary pollution.

[0008] 3. The prior art does not construct a synergistic optimization model of energy consumption and drug consumption. In the long-term operation, due to the lack of process parameter self-learning ability, the actual operation energy consumption is significantly higher than the design value, the drug consumption increases unreasonably due to equipment aging, and the contradiction between the limited operation and maintenance budget of the township sewage treatment facility is increasingly prominent.

[0009] To this end, the present application provides a sewage treatment process dynamic multi-objective optimization control method to solve the problems mentioned above. SUMMARY

[0010] In view of the deficiencies of the prior art, the present application provides a sewage treatment process dynamic multi-objective optimization control method to solve the problems mentioned in the background art.

[0011] To achieve the above purpose, the present application is implemented by the following technical scheme: a sewage treatment process dynamic multi-objective optimization control method, comprising:

[0012] Step 1, biological energy is collected from the effluent of the adjusting tank by the double-chamber microbial fuel cell, and the output voltage is supplied to the multi-parameter sensor and the edge controller through the conversion circuit;

[0013] Step 2, the multi-parameter sensor collects the COD value, ammonia nitrogen concentration, dissolved oxygen content and oxidation-reduction potential data of the sewage with a period of 5 minutes, and transmits them to the edge controller;

[0014] Step 3, the edge controller divides the low load, normal load and impact load three working condition modes according to the ratio of the real-time inflow to the design flow;

[0015] Step 4, the edge controller generates a set of collaborative control instructions of aeration intensity, carbon source dosage rate and sludge return flow based on the improved Q-learning algorithm according to the current working condition mode;

[0016] Step 5, the aeration intensity is adjusted by the variable frequency micro-aeration fan, and the collaborative control instructions are executed by synchronously controlling the opening degree of the sheet-shaped slow-release bin and the return period of the peristaltic pump;

[0017] Step 6, when the data change rate of the multi-parameter sensor exceeds the set threshold, the ring-shaped ultrasonic assembly is started to clean the sensor probe and the historical control parameters are called;

[0018] Step 7, the edge controller packs and uploads the device running state data to the remote monitoring platform through the LoRa communication module;

[0019] Step 8, the experience replay pool parameters of the Q-learning algorithm are iteratively updated based on 24h running data.

[0020] Preferably, in step 1, the anode cavity of the dual-cavity microbial fuel cell is filled with a carbon felt biofilm carrier, and the cathode cavity is provided with a polytetrafluoroethylene air diffusion layer;

[0021] The standard for dividing the working mode in step 3 is: a real-time flow ratio less than 0.3 is judged as low load, 0.3 to 1.2 is normal load, and greater than 1.2 is judged as impact load.

[0022] Preferably, the improved Q-learning algorithm in step 4 constructs a three-dimensional state space including flow ratio, pollutant concentration gradient, and equipment health, and defines a reward function with processing efficiency, energy consumption cost, and equipment loss as dimensions.

[0023] Preferably, in step 1, collecting bioenergy from the effluent of the regulating tank by using a dual-chamber microbial fuel cell further comprises:

[0024] Sub-step 1.1: Fill the anode cavity with carbon felt biofilm carriers. When sewage flows, organic matter oxidation reaction occurs to produce electrons and protons. The open circuit voltage V oc :

[0025] V oc =E cathode -E anode ,

[0026] Among them, E cathode is the cathode oxygen reduction potential, E anode is the oxidation potential of the organic substrate at the anode;

[0027] In sub-step 1.2, a polytetrafluoroethylene air diffusion layer is set in the cathode chamber. Protons pass through the proton exchange membrane to react with oxygen to produce a reduction reaction. The output current density i is:

[0028]

[0029] Among them, V operate is the operating voltage, R internal is the internal resistance of the battery;

[0030] Sub-step 1.3, when V operate When the voltage is lower than 3.6V, it switches to supercapacitor power supply mode. The switching judgment conditions are:

[0031] V operate <V threshold and

[0032] Among them, V threshold is the preset voltage threshold, is the voltage change rate;

[0033] Sub-step 1.4, output voltage is stabilized to 5V±0.2V by DC-DC converter, conversion efficiency η:

[0034]

[0035] Where, P out is output power, P in is input power, P loss is circuit loss power.

[0036] Preferably, in step 2, the multi-parameter sensor collects sewage parameters at a cycle of 5min further comprises:

[0037] Sub-step 2.1, start sensor self-calibration mode, calculate zero point drift compensation value of each channel:

[0038] ΔC i = C raw,i -C ref,i ,

[0039] Where, ΔC i is the zero point drift compensation value of the i channel, C raw,i is the original reading of the i channel, C ref,i is the reference value of the standard solution;

[0040] Sub-step 2.2, after the solid-state ion sieve membrane captures the target ions, calculate the ammonia nitrogen concentration C NH3 :

[0041] C NH3 = K (E m -E0) / S,

[0042] Where, K is the membrane screening coefficient, E m is the measured potential, E0 is the reference potential, and S is the Nernst slope;

[0043] Sub-step 2.3, microelectrode array detects dissolved oxygen content, and temperature compensation is carried out:

[0044] DO adj = DO raw / [1+α(T-T ref )],

[0045] Where, DO adj is the corrected value of dissolved oxygen concentration, DO raw is the original value of dissolved oxygen, α is the temperature coefficient, T is the real-time water temperature, and T ref is the standard temperature;

[0046] Sub-step 2.4, when the parameter mutation rate δ is detected to be greater than 10%, trigger abnormal processing:

[0047] δ = |(C t -C t-1 )| / (Δt·C t-1 ),

[0048] If δ exceeds the threshold, discard the current data,

[0049] wherein, δ is the data mutation rate, C t is the pollutant concentration measurement value of the current sampling period, C t-1 is the pollutant concentration history value of the last sampling period, and Δt is the time interval of adjacent sampling periods.

[0050] Sub-step 2.5, encapsulate the data by CRC-16 check and format transmission:

[0051] Data packet = [Header] [C NH3 ] [D adj ] [ORP] [CRC],

[0052] wherein, Data packet is the data frame structure, Header is the data frame header information, ORP is the oxidation-reduction potential measurement value, and CRC is the cyclic redundancy check code.

[0053] Preferably, in step 3, the edge controller divides the working condition mode further comprises:

[0054] Sub-step 3.1, calculate the real-time flow ratio η:

[0055] η = Q actual / Q design ,

[0056] wherein, Q actual is the current inflow, and Q design is the design treatment flow;

[0057] Sub-step 3.2, sliding average filtering processing is performed on η to update the effective flow ratio η eff :

[0058] η eff = (η t + η t-1 + η t-2 ) / 3,

[0059] When |η t - η t-1 | > 0.5, trigger the mutation flag:

[0060] Flag abrupt = 1,

[0061] wherein, η effη t is the real-time flow ratio of the current sampling period t-1 is the historical flow ratio of the last sampling period t-2 is the historical flow ratio of the last-but-one sampling period abrupt is the flow mutation flag

[0062] Sub-step 3.3, dividing the working condition mode:

[0063]

[0064] wherein Mode is the working condition mode classification result, η low is the low load determination threshold, η high is the impact load determination threshold

[0065] Sub-step 3.4, when Flag abrupt = 1, revising the working condition mode:

[0066] Mode = argmin(|η eff -η hist |),

[0067] wherein argmin(·) is a mathematical operator, η hist is the historical same-period flow ratio mean value

[0068] Sub-step 3.5, generating the working condition identifier code:

[0069] CaseID = [Mode][Flag abrupt ][Q actual / Q design × 100],

[0070] wherein CaseID is the working condition code

[0071] Preferably, in the step 4, the edge controller runs the improved Q-learning algorithm to generate the control instruction set further comprises:

[0072] Sub-step 4.1, constructing a three-dimensional state space M = (η eff , ΔC, H), calculating the pollutant concentration gradient ΔC:

[0073] ΔC = (C t - C t-1 ) / ΔT,

[0074] wherein C t is the current pollutant concentration, C t-1 is the last-period concentration, and ΔT is the sampling interval

[0075] Sub-step 4.2, define the device health H:

[0076] H = 1 -∑(β k ·F k ) / F max ,

[0077] where β k is the kth fault weight, F k is the number of failures, F max is the maximum allowed number of failures;

[0078] Sub-step 4.3, design the action space A = (Δf, Δα, Δq),

[0079] where Δf is the change in aeration frequency, Δα is the increase in slow-release bin opening, and Δq is the change in return flow rate;

[0080] Sub-step 4.4, construct the reward function R:

[0081] R = ω1(1 - C out / C in ) + ω2(1 - P / P max ) + ω3·H,

[0082] where ω1, ω2, ω3 are dynamic weights, C out is the effluent concentration, C in is the influent pollutant concentration, P is the real-time power, and P max is the rated power;

[0083] Sub-step 4.5, use the double-ε-greedy strategy to select actions:

[0084] ε = ε0·exp(-k·N episode ),

[0085] When rand(·) < ε, explore randomly, otherwise select the action with the maximum Q value, ε0 is the initial exploration rate, k is the decay coefficient, N episode is the cumulative number of algorithm training periods, ε is the exploration rate parameter, and rand(·) is an arbitrary number generation function;

[0086] Sub-step 4.6, apply the constraint condition to modify the action:

[0087] f current + Δf > f max , Δf = f max - f current ,

[0088] where f current is the current actual operating frequency of the aeration fan, and f max is the maximum safe frequency of the aeration fan.

[0089] Preferably, in step 5, adjusting aeration intensity by a variable frequency micro-aeration blower and executing control instructions further comprises:

[0090] Sub-step 5.1, parsing the cooperative control instruction set, calculating the target value f of aeration frequency target :

[0091] f target =f current +Δf·sgn(C out -C std ),

[0092] wherein f current is the current frequency, Δf is the aeration frequency change amount, C out is the effluent concentration, C std is the discharge standard limit value, and sgn is the sign function;

[0093] Sub-step 5.2, controlling the opening degree α of the sheet-shaped slow-release bin according to the following formula:

[0094] α=α0+k p ·(H target -H actual ),

[0095] wherein α0 is the reference opening degree, k p is the proportional coefficient, H target is the target carbon-nitrogen ratio, and H actual is the measured carbon-nitrogen ratio;

[0096] Sub-step 5.3, adjusting the peristaltic pump backflow period according to the following formula:

[0097] T pump =T base ·exp(-λ·ΔC),

[0098] wherein T pump is the peristaltic pump working period, T base is the reference period, λ is the attenuation coefficient, and ΔC is the pollutant concentration gradient;

[0099] Sub-step 5.4, the actuator linkage constraint condition is determined according to the following formula:

[0100] f target >0.8f max , α>50°,

[0101] α=50°-0.2(f target -0.8f max ),

[0102] to prevent high-frequency aeration and high-opening degree carbon injection from occurring simultaneously, wherein ftarget f is the aeration frequency target value max f is the maximum safe frequency of the aeration fan, a is the opening degree of the sheet-shaped slow-release bin;

[0103] Sub-step 5.5, update the execution state flag according to the following formula:

[0104] Flag exec = round(f target / f max ×100)∣∣round(α)×1000∣∣round(T pump ),

[0105] Wherein, Flag exec is the execution mechanism state combination flag, and round is the rounding operation.

[0106] Preferably, the step 6, when the multi-parameter sensor data change rate exceeds the set threshold, the abnormal processing further comprises:

[0107] Sub-step 6.1, calculate the data mutation rate δ according to the following formula:

[0108]

[0109] Wherein, δ is the data mutation rate, C t is the current pollutant concentration, C t-1 is the last period concentration, and ΔT is the sampling interval;

[0110] Sub-step 6.2, when δ> δ threshold , start ultrasonic cleaning according to the following formula:

[0111] T clean = min(120,30+15·[δ / 5]),

[0112] Wherein, δ threshold is the mutation threshold, and T clean is the cleaning time;

[0113] Sub-step 6.3, call the historical control parameter according to the following formula:

[0114]

[0115] Wherein, H param is the historical optimal parameter index value, C t-i is the pollutant concentration measurement value of i sampling periods before the current time t, argmin k is the mathematical operator, and C hist,k-i is the water quality data corresponding to the k-i group of parameters in the historical database;

[0116] Sub-step 6.4, data recovery is performed according to the following formula:

[0117]

[0118] wherein, C valid is the effective data output value, C hist,Hparam is the H param th parameter corresponding to the pollutant concentration data set;

[0119] Sub-step 6.5, the fault code encoding is generated according to the following formula:

[0120]

[0121] wherein, FaultCode is the fault code, round is the rounding operation, sensorID is the sensor identity code, clean count is the ultrasonic cleaning cumulative number counter.

[0122] Preferably, in step 7, the edge controller further comprises uploading device running state data through the communication module.

[0123] Sub-step 7.1, the device state matrix L is constructed according to the following formula:

[0124]

[0125] wherein, f current is the current frequency, a is the opening degree of the sheet-shaped slow-release bin, T pump is the working period of the peristaltic pump, δ is the data mutation rate, C out is the effluent concentration, C in is the influent pollutant concentration, H is the device health, FaultCode is the fault code, and CaseID is the working condition characteristic code identifier.

[0126] Sub-step 7.2, the data packet is compressed according to the following formula:

[0127]

[0128] wherein, Payload is the actual transmission data packet structure, Zigzag(L) is the device state matrix compression encoding method, is the XOR encryption, and CRC32 is the check polynomial.

[0129] Sub-step 7.3, the transmission power is dynamically adjusted according to the following formula:

[0130] P TX = min(P max , P base + 3·[RSSI / 10]),

[0131] Among them, P TX is the actual transmission power control amount, P max is the rated power, P base is the starting value of the benchmark transmit power, and RSSI is the received signal strength;

[0132] Sub-step 7.4, retransmission control is performed according to the following formula:

[0133] N retry It controls the number of retransmissions of the data packet, and ACK is the data packet confirmation flag;

[0134] When the confirmation frame ACK is not received, it is adaptively retransmitted according to the signal strength;

[0135] In sub-step 7.5, update the communication status flag according to the following formula:

[0136]

[0137] Among them, CommFlag is the communication status composite flag, and round is the rounding operation.

[0138] The present invention provides a dynamic multi-objective optimization control method for a sewage treatment process. It has the following beneficial effects:

[0139] 1. The present invention adopts the technical solution of dynamic division of working mode and reinforcement learning control instruction generation to achieve the technical effect of adaptive adjustment of aeration intensity and carbon source dosage according to the influent load. Compared with the fixed-cycle aeration and manual experience dosing mode in the existing technology, it solves the problems of control response lag and inflated energy and drug consumption.

[0140] 2. The present invention adopts a multi-parameter anomaly detection and remote status monitoring linkage technology solution to achieve the technical effects of equipment fault self-diagnosis and intelligent switching of safety modes. Compared with the existing technology, which lacks a remote fault-tolerant mechanism, it solves the problem of sudden failures causing system shutdown and secondary pollution.

[0141] 3. The present invention adopts a multi-objective optimization reward function and experience replay parameter iteration technical solution to achieve the technical effect of coordinated optimization of energy consumption and drug consumption and long-term operation self-adaptation. Compared with the existing technology that lacks the self-learning ability of process parameters, it solves the problem of irrational growth of operating costs and the contradiction between operation and maintenance budget. BRIEF DESCRIPTION OF THE DRAWINGS

[0142] Figure 1 Flowchart of the present invention. DETAILED DESCRIPTION

[0143] For those skilled in the technical field, the technical solutions in the embodiments of the present application will be described clearly and completely below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, other embodiments obtained by those of ordinary skill in the art without creative labor should be within the scope of protection of the present application.

[0144] The present application will be described in detail below in conjunction with the drawings:

[0145] Embodiments:

[0146] Please refer to the accompanying Figure 1 , the embodiment of the present application provides a sewage treatment process dynamic multi-objective optimization control method, comprising:

[0147] Step 1, the output voltage is collected by the double-cavity microbial fuel cell to the adjusting pool effluent, and the output voltage is supplied to the multi-parameter sensor and the edge controller by the conversion circuit;

[0148] Substep 1.1, fill the carbon felt biofilm carrier in the anode cavity, and the oxidation reaction of organic matter occurs when the sewage flows to produce electrons and protons, and the open circuit voltage V oc :

[0149] V oc =E cathode -E anode ,

[0150] Wherein, E cathode is the cathode oxygen reduction potential, and E anode is the anode organic substrate oxidation potential;

[0151] Substep 1.2, the cathode cavity is provided with a polytetrafluoroethylene air diffusion layer, and the protons pass through the proton exchange membrane and react with oxygen to output the current density i:

[0152]

[0153] Wherein, V operate is the working voltage, and R internal is the internal resistance of the battery;

[0154] Substep 1.3, when V operate is lower than 3.6V, switch to super capacitor power supply mode, and the switching judgment condition is:

[0155] V operate <V threshold And

[0156] Wherein, V threshold is a preset voltage threshold, is the voltage change rate;

[0157] In sub-step 1.4, stabilize the output voltage to 5V±0.2V through the DC-DC converter. The conversion efficiency η is:

[0158]

[0159] Among them, P out is the output power, P in is the input power, P loss is the circuit loss power;

[0160] Step 2: The multi-parameter sensor collects sewage COD value, ammonia nitrogen concentration, dissolved oxygen content, and redox potential data in a 5-minute cycle and transmits it to the edge controller;

[0161] Sub-step 2.1: Start the sensor self-calibration mode and calculate the zero drift compensation value of each channel:

[0162] ΔC i =C raw,i -C ref,i ,

[0163] Where, ΔC i is the zero drift compensation value of channel i, C raw,i is the original reading of channel i, C ref,i is the reference value of the standard solution;

[0164] Sub-step 2.2: After the solid ion screening membrane captures the target ions, calculate the ammonia nitrogen concentration C NH3 :

[0165] C NH3 =K(E m -E0) / S,

[0166] Among them, K is the membrane screening coefficient, E m is the measured potential, E0 is the reference potential, S is the Nernst slope;

[0167] Sub-step 2.3: Detect dissolved oxygen content using a microelectrode array and perform temperature compensation:

[0168] DO adj =DO raw / [1+α(TT ref ]),

[0169] Among them, DO adj is the dissolved oxygen concentration correction value, DO raw is the original value of dissolved oxygen, α is the temperature coefficient, T is the real-time water temperature, T ref is the standard temperature;

[0170] Sub-step 2.4, when the parameter mutation rate δ>10% is detected, triggering abnormal processing:

[0171] δ = |(C t -C t-1 )| / (Δt·C t-1 ),

[0172] If δ exceeds the threshold, discard the current data,

[0173] wherein δ is the data mutation rate, C t is the pollutant concentration measurement value of the current sampling period, C t-1 is the pollutant concentration history value of the last sampling period, and Δt is the time interval of adjacent sampling periods.

[0174] Sub-step 2.5, encapsulate the data by CRC-16 check and format transmission:

[0175] Data packet = [Header] [C NH3 ] [D adj ] [ORP] [CRC],

[0176] wherein Data packet is the data frame structure, Header is the data frame header information, ORP is the oxidation-reduction potential measurement value, and CRC is the cyclic redundancy check code.

[0177] Step 3, the edge controller divides the low load, normal load and impact load three working condition modes according to the real-time water inflow and design flow ratio;

[0178] Sub-step 3.1, calculate the real-time flow ratio η:

[0179] η = Q actual / Q design ,

[0180] wherein Q actual is the current water inflow, and Q design is the design treatment flow;

[0181] Sub-step 3.2, sliding average filtering processing is performed on η, and the effective flow ratio η eff is updated:

[0182] η eff = (η t + η t-1 + η t-2 ) / 3,

[0183] When |η t -η t-1 |>0.5, trigger mutation flag:

[0184] Flag abrupt = 1,

[0185] wherein η eff is the effective flow rate ratio, η t is the real-time flow rate ratio of the current sampling period, η t-1 is the historical flow rate ratio of the last sampling period, η t-2 is the historical flow rate ratio of the second last sampling period, and Flag abrupt is the flow rate mutation flag;

[0186] Sub-step 3.3, dividing the working condition mode:

[0187]

[0188] wherein Mode is the working condition mode classification result, η low is the low load determination threshold, and η high is the impact load determination threshold;

[0189] Sub-step 3.4, when Flag abrupt = 1, correcting the working condition mode:

[0190] Mode = argmin(|η eff - η hist |),

[0191] wherein argmin(·) is a mathematical operator, and η hist is the historical same period flow rate ratio mean value;

[0192] Sub-step 3.5, generating the working condition identifier code:

[0193] CaseID = [Mode][Flag abrupt ][Q actual / Q design x 100],

[0194] wherein CaseID is the working condition code;

[0195] Step 4, the edge controller runs the improved Q-learning algorithm based on the current working condition mode to generate the aeration intensity, carbon source dosage rate, and sludge return flow collaborative control instruction set;

[0196] Sub-step 4.1, constructing a three-dimensional state space M = (η eff , ΔC, H) and calculating the pollutant concentration gradient ΔC:

[0197] ΔC = (C t - C t-1 ) / ΔT,

[0198] wherein Ct C is the current pollutant concentration t-1 C is the concentration of the last cycle, ΔT is the sampling interval

[0199] Sub-step 4.2, define the device health H:

[0200] H = 1 -∑(β k ·F k ) / F max ,

[0201] Where β k is the kth fault weight, F k is the number of failures, F max is the maximum allowed number of failures

[0202] Sub-step 4.3, design the action space A = (Δf, Δα, Δq),

[0203] Where Δf is the change in aeration frequency, Δα is the increase in slow-release bin opening, and Δq is the change in return flow rate

[0204] Sub-step 4.4, construct the reward function R:

[0205] R = ω1(1-C out / C in )+ω2(1-P / P max )+ω3·H,

[0206] Where ω1, ω2, ω3 are dynamic weights, C out is the effluent concentration, C in is the influent pollutant concentration, P is the real-time power, and P max is the rated power

[0207] Sub-step 4.5, use the double-ε-greedy strategy to select actions:

[0208] ε = ε0·exp(-k·N episode ),

[0209] When rand(·) < ε, explore randomly, otherwise select the action with the maximum Q value, ε0 is the initial exploration rate, k is the decay coefficient, N episode is the cumulative number of algorithm training cycles, ε is the exploration rate parameter, and rand(·) is an arbitrary number generation function

[0210] Sub-step 4.6, apply the constraint condition to modify the action:

[0211] f current + Δf > f max , Δf = f max - f current ,

[0212] wherein f current is the current actual operating frequency of the aeration blower, f max is the maximum safe frequency of the aeration blower;

[0213] Step 5, adjust the aeration intensity by the variable frequency micro-aeration blower, and execute the cooperative control instruction synchronously by controlling the opening degree of the sheet-shaped slow-release bin and the reflux cycle of the peristaltic pump;

[0214] Sub-step 5.1, analyze the cooperative control instruction set, and calculate the target value f target of the aeration frequency:

[0215] f target = f current + Δf· sgn(C out -C std ),

[0216] wherein f current is the current frequency, Δf is the aeration frequency change amount, C out is the effluent concentration, C std is the discharge standard limit value, and sgn is the sign function;

[0217] Sub-step 5.2, control the opening degree α of the sheet-shaped slow-release bin according to the following formula:

[0218] α = α0+ k p ·(H target -H actual ),

[0219] wherein α0is the reference opening degree, k p is the proportional coefficient, H target is the target carbon-nitrogen ratio, and H actual is the measured carbon-nitrogen ratio;

[0220] Sub-step 5.3, adjust the reflux cycle of the peristaltic pump according to the following formula:

[0221] T pump = T base · exp(-λ· ΔC),

[0222] wherein T pump is the working cycle of the peristaltic pump, T base is the reference cycle, λ is the attenuation coefficient, and ΔC is the pollutant concentration gradient;

[0223] Sub-step 5.4, the linkage constraint condition of the actuator is determined according to the following formula:

[0224] f target > 0.8f max , α > 50°,

[0225] α = 50°- 0.2(ftarget -0.8f max ),

[0226] Prevent high-frequency aeration and high-opening carbon injection from occurring simultaneously, where f target is the target value of aeration frequency, f max is the maximum safe frequency of the aeration fan, and a is the opening of the sheet-shaped slow-release bin;

[0227] Sub-step 5.5, update the execution state flag according to the following formula:

[0228] Flag exec = round(f target / f max ×100)∣∣round(α)×1000∣∣round(T pump ),

[0229] where Flag exec is the execution mechanism state combination flag, and round is the rounding operation;

[0230] Step 6, when the multi-parameter sensor data change rate exceeds the set threshold, start the ring ultrasonic assembly cleaning sensor probe and call the historical control parameters;

[0231] Sub-step 6.1, calculate the data mutation rate δ according to the following formula:

[0232]

[0233] where δ is the data mutation rate, C t is the current pollutant concentration, C t-1 is the previous period concentration, and ΔT is the sampling interval;

[0234] Sub-step 6.2, when δ > δ threshold , start ultrasonic cleaning according to the following formula:

[0235] T clean = min(120, 30 + 15·[δ / 5]),

[0236] where δ threshold is the mutation threshold, and T clean is the cleaning duration;

[0237] Sub-step 6.3, call the historical control parameters according to the following formula:

[0238]

[0239] where H param is the historical optimal parameter index value, and C t-iis the pollutant concentration measurement value of the sampling period i before the current time t, argmin k For mathematical operators, C hist,k-i is the water quality data corresponding to the kith group of parameters in the historical database;

[0240] In sub-step 6.4, perform data recovery according to the following formula:

[0241]

[0242] Among them, C valid is the valid data output value, For H param The pollutant concentration data set corresponding to the parameter;

[0243] In sub-step 6.5, generate the fault code according to the following formula:

[0244]

[0245] Among them, FaultCode is the fault code, round is the rounding operation, sensorID is the sensor identification code, clean count A counter for cumulative ultrasonic cleaning times;

[0246] Step 7: The edge controller packages the device operation status data and uploads it to the remote monitoring platform through the LoRa communication module;

[0247] In sub-step 7.1, construct the device state matrix L according to the following formula:

[0248]

[0249] Among them, f current is the current frequency, α is the opening of the sheet slow-release chamber, T pump is the working cycle of the peristaltic pump, δ is the data mutation rate, C out is the effluent concentration, C in is the influent pollutant concentration, H is the equipment health, FaultCode is the fault code, and CaseID is the operating condition characteristic code identifier;

[0250] In sub-step 7.2, compress the data packet according to the following formula:

[0251]

[0252] Among them, Payload is the actual transmission data packet structure, Zigzag (L) is the device state matrix compression encoding method, It is XOR encryption, and CRC32 is the check polynomial;

[0253] Sub-step 7.3, dynamically adjust the transmission power as follows:

[0254] P TX = min(P max ,P base + 3·[RSSI / 10]),

[0255] Wherein, P TX is the actual transmission power control, P max is the rated power, P base is the reference transmission power starting value, RSSI is the received signal strength;

[0256] Sub-step 7.4, retransmission control as follows:

[0257] N retry is the number of data packet retransmission control, ACK is the data packet confirmation flag;

[0258] When the confirmation frame ACK is not received, the signal strength is adaptively retransmitted;

[0259] Sub-step 7.5, update the communication status flag as follows:

[0260]

[0261] Wherein, CommFlag is the communication state composite flag, round is the rounding operation;

[0262] Step 8, based on 24h operation data iterative update Q-learning algorithm experience playback pool parameters.

[0263] In step 1, the anode cavity of the double-cavity microbial fuel cell is filled with carbon felt biofilm carrier, and the cathode cavity is provided with polytetrafluoroethylene air diffusion layer.

[0264] In step 3, the working condition mode division standard is: the real-time flow ratio is less than 0.3, which is determined as low load, 0.3 to 1.2 is the conventional load, and greater than 1.2 is determined as impact load.

[0265] In step 4, the improved Q-learning algorithm constructs a three-dimensional state space containing flow ratio, pollutant concentration gradient and equipment health degree, and defines a reward function with treatment efficiency, energy consumption cost and equipment loss as dimensions.

[0266] Through the bioenergy self-supply system to realize the equipment power supply closed loop, using microbial fuel cell to directly convert organic chemical energy into electrical energy, break through the limitation of traditional sewage treatment system relying on external power supply, significantly improve the system self-operation ability when deployed in remote areas, at the same time through intelligent power supply switching mechanism to ensure the continuous and stable operation of key equipment.

[0267] A multi-dimensional water quality perception network is constructed, dynamic calibration and abnormal data filtering technology are adopted, sensor drift and transient interference problems are effectively overcome, and accurate continuous monitoring of water quality parameters is realized, providing a high-reliability data foundation for subsequent intelligent decision-making, and greatly reducing the risk of control misalignment caused by measurement errors.

[0268] An adaptive operating condition recognition model based on flow characteristics is established, transient flow fluctuation interference is eliminated through sliding filtering and historical data comparison, and the operating load level is accurately divided, providing a classification benchmark for process parameter optimization under different inflow conditions, and solving the control response disorder problem caused by the lag of traditional method operating condition judgment.

[0269] Innovative design of multi-objective reinforcement learning control architecture, the health status of equipment is considered in decision-making, the dynamic reward mechanism is used to balance processing efficiency and operating cost, and the collaborative optimization of aeration, dosing, reflux and other execution units is realized, breaking through the limitations of isolated adjustment of each process parameter under the control mode of artificial experience.

[0270] Step 5: Develop intelligent linkage control strategy for actuator, introduce physical constraint condition to prevent equipment overload operation, through frequency adjustment and proportional opening control guided by sign function, ensure that process adjustment is quickly and accurately implemented, while maintaining treatment effect, and maximize reduce mechanical loss and energy waste.

[0271] Build an abnormal operating condition self-recovery system, use mutation rate threshold to trigger ultrasonic cleaning and historical parameter calling mechanism, effectively deal with sudden conditions such as sensor contamination failure, and ensure continuous and stable operation of the system through multi-source data fusion repair technology, significantly reduce the ecological risk caused by unplanned downtime.

[0272] Design an extensible Internet of Things communication architecture, realize efficient transmission in low-bandwidth environment through matrix data organization and adaptive power control, establish a panoramic visualization channel for equipment operating status, provide real-time data support for remote intelligent operation and maintenance, and solve the problem of facility supervision in rural areas.

[0273] Create an experience knowledge continuous evolution model, update algorithm parameters through historical data driving, make the control system have process adaptability and equipment aging compensation ability, ensure the continuous optimization of treatment efficiency and cost control indicators in long-term operation, and break through the technical bottleneck of performance degradation of static control model.

[0274] Although embodiments of the present application have been shown and described, it will be understood by those of ordinary skill in the art that various changes, modifications, substitutions and alterations can be made without departing from the principles and spirit of the present application, and the scope of the present application is defined by the appended claims and their equivalents.

Claims

1. A dynamic multi-objective optimization control method for a sewage treatment process, characterized in that: include: Step 1: bioenergy is collected from the effluent of the regulating pool by a dual-chamber microbial fuel cell, and the output voltage is supplied to the multi-parameter sensor and the edge controller by a conversion circuit; Step 2: The multi-parameter sensor collects sewage COD value, ammonia nitrogen concentration, dissolved oxygen content, and redox potential data in a 5-minute cycle and transmits it to the edge controller; Step 3: The edge controller divides the working mode into three modes: low load, normal load, and impact load according to the ratio of the real-time water inflow rate to the design flow rate; Step 4: The edge controller runs the improved Q-learning algorithm based on the current operating mode to generate a coordinated control instruction set for aeration intensity, carbon source injection rate, and sludge return flow; Step 5: Adjust the aeration intensity through the variable frequency micro-aeration fan, and synchronously control the opening of the sheet-shaped slow-release chamber and the reflux cycle of the peristaltic pump to execute the coordinated control instructions; Step 6: When the rate of change of the multi-parameter sensor data exceeds a set threshold, the annular ultrasonic component is started to clean the sensor probe and the historical control parameters are called; Step 7: The edge controller packages the device operation status data and uploads it to the remote monitoring platform through the LoRa communication module; Step 8: Iteratively update the experience replay pool parameters of the Q-learning algorithm based on the 24-hour running data.

2. A dynamic multi-objective optimization control method for a sewage treatment process according to claim 1, characterized in that: In the step 1, the anode cavity of the dual-cavity microbial fuel cell is filled with a carbon felt biofilm carrier, and the cathode cavity is provided with a polytetrafluoroethylene air diffusion layer; The standard for dividing the working mode in step 3 is: a real-time flow ratio less than 0.3 is judged as low load, 0.3 to 1.2 is normal load, and greater than 1.2 is judged as impact load.

3. A dynamic multi-objective optimization control method for a sewage treatment process according to claim 1, characterized in that: In step 4, the improved Q-learning algorithm constructs a three-dimensional state space including flow ratio, pollutant concentration gradient, and equipment health, and defines a reward function with processing efficiency, energy consumption cost, and equipment loss as dimensions.

4. A dynamic multi-objective optimization control method for a sewage treatment process according to claim 1, characterized in that: In step 1, collecting bioenergy from the effluent of the regulating tank by using a dual-chamber microbial fuel cell further includes: Sub-step 1.1: Fill the anode cavity with carbon felt biofilm carriers. When sewage flows, organic matter oxidation reaction occurs to produce electrons and protons. The open circuit voltage V oc : V oc =E cathode -E anode , Among them, E cathode is the cathode oxygen reduction potential, E anode is the oxidation potential of the organic substrate at the anode; In sub-step 1.2, a polytetrafluoroethylene air diffusion layer is set in the cathode chamber. Protons pass through the proton exchange membrane to react with oxygen to produce a reduction reaction. The output current density i is: Among them, V operate is the operating voltage, R internal is the internal resistance of the battery; Sub-step 1.3, when V operate When the voltage is lower than 3.6V, it switches to supercapacitor power supply mode. The switching judgment conditions are: V operate <V threshold and Among them, V threshold is the preset voltage threshold, is the voltage change rate; In sub-step 1.4, stabilize the output voltage to 5V±0.2V through the DC-DC converter. The conversion efficiency η is: Among them, P out is the output power, P in is the input power, P loss is the power loss of the circuit.

5. A dynamic multi-objective optimization control method for a sewage treatment process according to claim 1, characterized in that: In step 2, the multi-parameter sensor collects sewage parameters in a 5-minute cycle, further comprising: Sub-step 2.1: Start the sensor self-calibration mode and calculate the zero drift compensation value of each channel: ΔC i =C raw,i -C ref,i , Where, ΔC i is the zero drift compensation value of channel i, C raw,i is the original reading of channel i, C ref,i is the reference value of the standard solution; Sub-step 2.2: After the solid ion screening membrane captures the target ions, calculate the ammonia nitrogen concentration C NH3 : C NH3 =K(E m -E0) / S, Among them, K is the membrane screening coefficient, E m is the measured potential, E0 is the reference potential, S is the Nernst slope; Sub-step 2.3: Detect dissolved oxygen content using a microelectrode array and perform temperature compensation: DO adj =DO raw / [1+α(T-T ref )], Among them, DO adj is the dissolved oxygen concentration correction value, DO raw is the original value of dissolved oxygen, α is the temperature coefficient, T is the real-time water temperature, T ref is the standard temperature; Sub-step 2.4: When the parameter mutation rate δ > 10% is detected, the exception handling is triggered: δ=|(C t -C t-1 )| / (Δt·C t-1 ), If δ exceeds the threshold, the current data is discarded. Among them, δ is the data mutation rate, C t is the pollutant concentration measurement value of the current sampling period, C t-1 is the historical value of the pollutant concentration in the previous sampling period, and Δt is the time interval between adjacent sampling periods; Sub-step 2.5, encapsulate the data through CRC-16 check and transmit in the following format: Data packet =[Header][C NH3 ][DO adj ][ORP][CRC], Among them, Data packet It is a data frame structure, Header is the data frame header information, ORP is the oxidation-reduction potential measurement value, and CRC is the cyclic redundancy check code.

6. A dynamic multi-objective optimization control method for a sewage treatment process according to claim 1, characterized in that: In step 3, the edge controller divides the working mode further including: Sub-step 3.1, calculate the real-time flow ratio η: η=Q actual / Q design , Among them, Q actual is the current water inlet flow, Q design To handle traffic for design; Sub-step 3.2, perform sliding average filtering on η and update the effective flow ratio η eff : or eff =(the t +n t-1 +n t-2 ) / 3, When |η t -η t-1 When |>0.5, the mutation flag is triggered: Flag abrupt =1, Among them, η eff is the effective flow ratio, η t is the real-time flow ratio of the current sampling period, η t-1 is the historical traffic ratio of the previous sampling period, η t-2 is the historical traffic ratio of the previous sampling period, Flag abrupt It is a sign of traffic mutation; Sub-step 3.3, divide the working mode: Among them, Mode is the classification result of the working mode, η low is the low load judgment threshold, η high is the shock load determination threshold; Sub-step 3.4, when Flag abrupt =1, the working mode is modified: Mode=argmin(|η eff -or hist |), Among them, argmin(·) is a mathematical operator, η hist is the average of the traffic ratio in the same period of history; Sub-step 3.5, generate the working condition identifier code: CaseID=[Mode][Flag abrupt ][Q actual / Q design ×100], Among them, CaseID is the working condition code.

7. A dynamic multi-objective optimization control method for a sewage treatment process according to claim 1, characterized in that: In step 4, the edge controller runs the improved Q-learning algorithm to generate a control instruction set further comprising: Sub-step 4.1, construct a three-dimensional state space M = (η eff ,ΔC,H), calculate the pollutant concentration gradient ΔC: ΔC=(C t -C t-1 ) / ΔT, Among them, C t is the current pollutant concentration, C t-1 is the concentration of the previous cycle, ΔT is the sampling interval; Sub-step 4.2, define the device health H: H=1-Σ(β k ·F k ) / F max , Among them, β k is the k-th fault weight, F k is the number of failures, F max is the maximum allowable number of failures; Sub-step 4.3, design the action space A = (Δf, Δα, Δq), Among them, Δf is the change of aeration frequency, Δα is the increment of slow-release chamber opening, and Δq is the change rate of reflux flow; Sub-step 4.4, construct the reward function R: R=ω1(1-C out / C in )+ω2(1-P / P max )+ω3·H, Among them, ω1, ω2, ω3 are dynamic weights, C out is the effluent concentration, C in is the influent pollutant concentration, P is the real-time power, P max is the rated power; Sub-step 4.5, use the double ε-greedy strategy to select actions: ε=ε0·exp(-k·N episode ), When rand(·)<ε, arbitrary exploration is performed, otherwise the action with the maximum Q value is selected, ε0 is the initial exploration rate, k is the attenuation coefficient, N episode is the cumulative number of algorithm training cycles, ε is the exploration rate parameter, and rand(·) is an arbitrary number generating function; Sub-step 4.6, apply the constraints to modify the action: f current +Δf>f max ,Δf=f max -f current , Among them, f current is the actual operating frequency of the aeration fan, f max It is the maximum safe frequency of the aeration fan.

8. A dynamic multi-objective optimization control method for a sewage treatment process according to claim 1, characterized in that: In step 5, adjusting the aeration intensity by the variable frequency micro-aeration fan and executing the control instruction further includes: Sub-step 5.1: Analyze the collaborative control instruction set and calculate the aeration frequency target value f target : f target =f current +Δf·sgn(C out -C std ), Among them, f current is the current frequency, Δf is the change in aeration frequency, C out is the effluent concentration, C std is the emission standard limit, sgn is the sign function; Sub-step 5.2: Control the opening degree α of the sheet-like slow-release chamber according to the following formula: α=α0+k p ·(H target -H actual ), Among them, α0 is the reference opening, k p is the proportional coefficient, H target is the target carbon-nitrogen ratio, H actual is the measured carbon-nitrogen ratio; In sub-step 5.3, adjust the peristaltic pump reflux cycle according to the following formula: T pump =T base ·exp(-λ·ΔC), Among them, T pump is the working cycle of the peristaltic pump, T base is the reference period, λ is the attenuation coefficient, and ΔC is the pollutant concentration gradient; In sub-step 5.4, the actuator linkage constraint condition is determined according to the following formula: f target >0.8f max ,α>50°, α=50°-0.2(f target -0.8f max ), Prevent high-frequency aeration and high-opening carbon injection from occurring simultaneously, where f target is the target value of aeration frequency, f max is the maximum safe frequency of the aeration fan, and α is the opening of the sheet slow-release chamber; In sub-step 5.5, update the execution status flag according to the following formula: Flag exec =round(f target / f max ×100)∣∣round(α)×1000∣∣round(T pump ), Among them, Flag exec It is the combined flag of the actuator status, and round is the rounding operation.

9. A dynamic multi-objective optimization control method for a sewage treatment process according to claim 1, characterized in that: In step 6, when the rate of change of the multi-parameter sensor data exceeds a set threshold, performing abnormal processing further includes: In sub-step 6.1, calculate the data mutation rate δ according to the following formula: Among them, δ is the data mutation rate, C t is the current pollutant concentration, C t-1 is the concentration of the previous cycle, ΔT is the sampling interval; Sub-step 6.2, when δ>δ threshold When the ultrasonic cleaning is started, the following formula is used: T clean =min(120,30+15·[δ / 5]), Among them, δ threshold is the mutation threshold, T clean For cleaning time; In sub-step 6.3, call the historical control parameters according to the following formula: Among them, H param is the historical optimal parameter index value, C t-i is the pollutant concentration measurement value of the sampling period i before the current time t, argmin k For mathematical operators, C hist,k-i is the water quality data corresponding to the kith group of parameters in the historical database; In sub-step 6.4, perform data recovery according to the following formula: Among them, C valid is the valid data output value, For H param The pollutant concentration data set corresponding to the parameter; In sub-step 6.5, generate the fault code according to the following formula: Among them, FaultCode is the fault code, round is the rounding operation, sensorID is the sensor identification code, clean count It is a counter for the cumulative number of ultrasonic cleanings.

10. A dynamic multi-objective optimization control method for a sewage treatment process according to claim 1, characterized in that: In step 7, the edge controller uploading the device operation status data through the communication module further includes: In sub-step 7.1, construct the device state matrix L according to the following formula: Among them, f current is the current frequency, α is the opening of the sheet slow-release chamber, T pump is the working cycle of the peristaltic pump, δ is the data mutation rate, C out is the effluent concentration, C in is the influent pollutant concentration, H is the equipment health, FaultCode is the fault code, and CaseID is the operating condition characteristic code identifier; In sub-step 7.2, compress the data packet according to the following formula: Payload=Zigzag(L)⊕CRC32, Where Payload is the actual transmission data packet structure, Zigzag(L) is the device state matrix compression encoding method, ⊕ is XOR encryption, and CRC32 is the check polynomial; In sub-step 7.3, dynamically adjust the transmit power according to the following formula: P TX =min(P max ,P base +3·[RSSI / 10]), Among them, P TX is the actual transmission power control amount, P max is the rated power, P base is the starting value of the benchmark transmit power, and RSSI is the received signal strength; Sub-step 7.4, retransmission control is performed according to the following formula: N retry It controls the number of retransmissions of the data packet, and ACK is the data packet confirmation flag; When the confirmation frame ACK is not received, it is adaptively retransmitted according to the signal strength; In sub-step 7.5, update the communication status flag according to the following formula: Among them, CommFlag is the communication status composite flag, and round is the rounding operation.

Images