A method for multi-objective collaborative optimization control of watershed ecological water resources

Abstract

The present application relates to the technical field of industrial control system, disclose a kind of basin ecological water resources multi-objective collaborative optimization control method, comprising: the state space model of controlled unit is built, disturbance parameter sequence is collected;According to model prediction solution obtains control feasible solution set under different working conditions, extract the public intersection satisfying threshold constraint to establish robust control envelope set;Real-time monitoring the state deviation rate of change of controlled main loop, according to this dynamic adjustment controlled slave loop control dead zone;According to the deviation attenuation characteristics of main loop, anti-saturation reset is implemented to the integral regulator of slave loop, and inertia filter is implemented to given value, the present application establishes online adjustment boundary using robust control envelope set, reduces the impact of random disturbance on control law stability, effectively suppresses the oscillation generated in the adjustment process, realizes the nonlinear decoupling of different time scale control loop, and guarantees the running stability of controlled object.

Description

Technical Field

[0001] This invention relates to a multi-objective collaborative optimization control method for watershed ecological water resources, belonging to the field of industrial control technology. Background Technology

[0002] Current mainstream technologies employ adaptive model predictive control or proportional-integral-derivative (PID) regulation. By establishing a mathematical model of the controlled domain, commands are issued to sluice gates, pumping stations, and aeration devices. This approach achieves water resource allocation and water quality compliance under stable hydrological conditions. However, with increasing random climate fluctuations, the controlled system faces high-dimensional non-stationary disturbances. Due to the system's large time delay, strong nonlinearity, and multivariate coupling characteristics, the overall hydraulic variables and surface water quality indicators exhibit significant time-scale asymmetry in spatial transmission and physical response speed. Existing deterministic control logic relies on historical time-series data to calibrate parameters. When dealing with non-stationary disturbances, the predictive residuals of the control loop amplify rapidly. To track static set targets, the actuators generate high-frequency oscillations, leading to a decrease in system stability.

[0003] Conventional improvement approaches involve increasing the sampling frequency or adopting a logic suspension strategy. However, in industrial settings, state freezing cannot cover the physical continuity of biochemical dissipation processes. When the master loop is in a transitional state, if the dead zone of the slave loop is mechanically widened, integral saturation occurs within the control law. When the system resumes tracking, the accumulated control potential energy drives the actuator to output at full load, causing secondary impacts on the physical riverbed and electromechanical equipment. The lack of a real-time deviation sensing mechanism in the controlled master-slave loop is a core issue leading to inaccuracies in multi-objective coordination. For example, Chinese invention patent application CN120337827A discloses an annual-scale robust scheduling logic for lake water resources-water security coordination. The method, system, equipment, and storage medium utilize Bernoulli-Gamma distribution to construct an annual-scale overall scheduling scheme. Discrete scenario clustering enhances the system's long-term adaptability to meteorological uncertainties. However, in actual industrial control, this method anchors quasi-static boundary settings and lacks a dynamic sensing mechanism for the real-time deviation change rate of the controlled master-slave loop. When the hydraulic master loop undergoes transient reconstruction due to extreme runoff, the scheduling logic struggles to modulate the control dead zone of the slave loop in real time, causing ineffective closed-loop compensation at the water quality regulation end in turbulent flow fields, inducing integral saturation. The control law design lacks dynamic residual discharge and physical rate matching mechanisms, generating potential energy impacts at the moment of system reset. Heterogeneous loops cannot be deeply decoupled under non-stationary operating conditions.

[0004] Therefore, how to construct a core control mechanism that can accommodate uncertain disturbances and eliminate time scale misalignment and integral storm impact between heterogeneous loops has become the technical problem to be solved by this invention. Summary of the Invention

[0005] To address the problems mentioned in the background art, the technical solution of the present invention is as follows: A multi-objective collaborative optimization control method for watershed ecological water resources, comprising the following steps:

[0006] Step S1: Construct a state-space model of the hydrodynamic and water quality relationship of the controlled unit of the watershed. The state-space model includes a controlled main loop composed of hydraulic scheduling and a controlled slave loop composed of water quality regulation. Obtain the perturbation parameter sequence that characterizes the perturbation features of the controlled domain.

[0007] Step S2: Input the disturbance parameter sequence into the state space model, obtain the feasible solution set of control variables under different working conditions through model prediction and calculation, and extract the common intersection of the feasible solution set according to the preset threshold constraint to establish the robust control envelope set.

[0008] Step S3: Monitor the rate of change of state deviation of the controlled master loop in real time, and dynamically adjust the control dead zone of the controlled slave loop according to the rate of change of state deviation. When the rate of change of state deviation is greater than the preset stable benchmark, increase the control dead zone and lock the online adjustment process within the robust control envelope set.

[0009] Step S4: Extract the deviation decay time constant of the controlled master circuit, perform anti-saturation reset on the integral regulator of the controlled slave circuit according to the deviation decay time constant, and perform first-order inertial filtering on the given target value of the controlled slave circuit according to the filter factor determined by the deviation decay time constant.

[0010] Preferably, the establishment of the robust control envelope set in step S2 includes: step S21, using a random sampling algorithm to discretize the disturbance parameter sequence to generate an input parameter set covering rainfall, runoff, and drought / flood conditions; step S22, calling the model predictive control algorithm to solve the state space model in each sampling period to obtain the control law family that has not touched the water level warning line and pollutant concentration limit under the disturbance of each input parameter set; step S23, extracting the common boundary of the control law family and determining the parameter space defined by the linear inequality constraint as the robust control envelope set.

[0011] Preferably, the state-space model uses the flow rate, water level, and COD concentration of the cross-section as state variables, the gate opening and the distributed interception multiple as control variables, and the non-point source pollution load and rainfall intensity as external disturbance variables.

[0012] Preferably, the step S3 of dynamically adjusting the control dead zone of the controlled slave loop includes: step S31, calculating the absolute value of the rate of change of the state deviation of the controlled master loop; step S32, when the absolute value exceeds the preset stability benchmark, increasing the control dead zone of the controlled slave loop according to the preset sensitivity coefficient; step S33, when the absolute value is lower than or equal to the preset stability benchmark, restoring the control dead zone to the static preset value.

[0013] Preferably, step S4, which involves performing anti-saturation reset on the integral regulator of the controlled slave loop, includes: step S41, observing the decay trajectory of the rate of change of the state deviation of the controlled master loop, and extracting the exponential change characteristics of the decay trajectory to determine the deviation decay time constant; step S42, converting the rate of change of the state deviation into a negative compensation gain through the deviation decay time constant, and continuously compensating it to the integral regulator of the controlled slave loop to neutralize the integral accumulation generated during the control dead zone adjustment.

[0014] Preferably, the first-order inertial filtering of the given target value of the controlled slave loop in step S4 includes: step S43, obtaining the target value step signal of the controlled slave loop at the control dead zone reset time; step S44, using the deviation decay time constant as the filtering parameter, smoothing the target value step signal, and constructing an asymptotic tracking curve that is synchronized with the stable velocity of the physical flow field of the controlled master loop.

[0015] Preferably, the control frequency of the controlled master loop is higher than that of the controlled slave loop, and the control gain of the controlled slave loop in the state-space model is limited by the real-time boundary value of the robust control envelope set.

[0016] Preferably, the method is completed by the control unit of the watershed dispatch center. The control unit performs logical operations based on the real-time hydrological data and water quality sensor signals collected by the lower-level PLC through the industrial bus, and sends the control variables as set values ​​to the execution end of the controlled object.

[0017] Preferably, in step S2, when solving the state-space model, under the premise of satisfying the linear inequality constraint of the robust control envelope set, the balance between the flood control margin of the watershed and the water quality compliance rate of the section is achieved by adjusting the weight coefficients.

[0018] Compared with the prior art, the beneficial effects of the present invention are:

[0019] 1. In the multi-objective collaborative optimization control of watershed ecological water resources, multiple climate disturbance scenario vectors are input into the established integrated state space model of the controlled domain. This drives the model to complete the traversal calculation of control quantities under various extreme climate fluctuation conditions. The control parameter subspace that does not reach the safety limit under each scenario is extracted to construct a robust envelope set. This physically constrains the online multi-objective optimization process within the parameter range with safety redundancy, reduces the impact of high-dimensional non-stationary random disturbances on the stability of the control law, and ensures that the controlled entities in the watershed remain within the safe control envelope under disturbance conditions that exceed historical experience.

[0020] 2. By utilizing the real-time residual gradient generated by the main loop of the hydraulic actuator to dynamically modulate the current control dead zone of the driven loop, the action threshold of the biochemical regulation end is adaptively widened during the transient process of hydraulic flow field reconstruction. This effectively blocks the parasitic interference of the turbulent hydrodynamic field on the surface water quality control process, enabling the driven actuator to automatically shield invalid feedback compensation actions, reducing the frequency of invalid opening and closing and mechanical losses of physical actuators during the non-steady-state transition period, and realizing the nonlinear decoupling of the overall hydrological regulation and surface biochemical control under time-scale asymmetric conditions.

[0021] 3. By introducing a feedforward integral discharge mechanism based on the residual decay time constant, the residual gradient of the dominant loop is converted into a negative compensation quantity and continuously fed into the deviation integral accumulator of the driven loop. This suppresses the accumulation of integral potential energy caused by control suspension during the dead zone widening period. Combined with the first-order inertial filtering of the target tracking command execution, this ensures that the system achieves asymptotic matching between the control state and the physical response rate at the dead zone reset moment. This eliminates the integral saturation overshoot and physical pulse impact that the traditional freeze strategy inevitably produces at the reset moment, ensuring the global convergence of the multivariable strongly coupled controlled system. Attached Figure Description

[0022] Figure 1 This is a schematic diagram of the multi-objective collaborative control and loop decoupling principle architecture of watershed water resources in this invention;

[0023] Figure 2 This is a flowchart of the adaptive logic and robust optimization process of the control system under complex working conditions in this invention.

[0024] The objectives, features, and advantages of this invention will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0025] The technical solutions of the embodiments of this application will be clearly described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, not all embodiments. All other embodiments obtained by those skilled in the art based on the embodiments of this application are within the scope of protection of this application.

[0026] A multi-objective collaborative optimization control method for watershed ecological water resources includes the following steps:

[0027] Step S1: Construct a state-space model of the hydrodynamic and water quality relationship of the controlled unit of the watershed. The state-space model includes a controlled main loop composed of hydraulic scheduling and a controlled slave loop composed of water quality regulation. Obtain the perturbation parameter sequence that characterizes the perturbation features of the controlled domain.

[0028] Step S2: Input the disturbance parameter sequence into the state space model, obtain the feasible solution set of control variables under different working conditions through model prediction and calculation, and extract the common intersection of the feasible solution set according to the preset threshold constraint to establish the robust control envelope set.

[0029] Step S3: Monitor the rate of change of state deviation of the controlled master loop in real time, and dynamically adjust the control dead zone of the controlled slave loop according to the rate of change of state deviation. When the rate of change of state deviation is greater than the preset stable benchmark, increase the control dead zone and lock the online adjustment process within the robust control envelope set.

[0030] Step S4: Extract the deviation decay time constant of the controlled master circuit, perform anti-saturation reset on the integral regulator of the controlled slave circuit according to the deviation decay time constant, and perform first-order inertial filtering on the given target value of the controlled slave circuit according to the filter factor determined by the deviation decay time constant.

[0031] Preferably, the establishment of the robust control envelope set in step S2 includes: step S21, using a random sampling algorithm to discretize the disturbance parameter sequence to generate an input parameter set covering rainfall, runoff, and drought / flood conditions; step S22, calling the model predictive control algorithm to solve the state space model in each sampling period to obtain the control law family that has not touched the water level warning line and pollutant concentration limit under the disturbance of each input parameter set; step S23, extracting the common boundary of the control law family and determining the parameter space defined by the linear inequality constraint as the robust control envelope set.

[0032] Preferably, the state-space model uses the flow rate, water level, and COD concentration of the cross-section as state variables, the gate opening and the distributed interception multiple as control variables, and the non-point source pollution load and rainfall intensity as external disturbance variables.

[0033] Preferably, the step S3 of dynamically adjusting the control dead zone of the controlled slave loop includes: step S31, calculating the absolute value of the rate of change of the state deviation of the controlled master loop; step S32, when the absolute value exceeds the preset stability benchmark, increasing the control dead zone of the controlled slave loop according to the preset sensitivity coefficient; step S33, when the absolute value is lower than or equal to the preset stability benchmark, restoring the control dead zone to the static preset value.

[0034] Preferably, the calculation logic for controlling the dead zone follows the following formula: Where δ(t) is the dynamically adjusted control dead zone at time t. η is the preset minimum dead zone width, and η is the preset sensitivity coefficient. Let be the rate of change of the state deviation of the controlled main circuit at time t.

[0035] Preferably, step S4, which involves performing anti-saturation reset on the integral regulator of the controlled slave loop, includes: step S41, observing the decay trajectory of the rate of change of the state deviation of the controlled master loop, and extracting the exponential change characteristics of the decay trajectory to determine the deviation decay time constant; step S42, converting the rate of change of the state deviation into a negative compensation gain through the deviation decay time constant, and continuously compensating it to the integral regulator of the controlled slave loop to neutralize the integral accumulation generated during the control dead zone adjustment.

[0036] Preferably, the first-order inertial filtering of the given target value of the controlled slave loop in step S4 includes: step S43, obtaining the target value step signal of the controlled slave loop at the control dead zone reset time; step S44, using the deviation decay time constant as the filtering parameter, smoothing the target value step signal, and constructing an asymptotic tracking curve that is synchronized with the stable velocity of the physical flow field of the controlled master loop.

[0037] Preferably, the control frequency of the controlled master loop is higher than that of the controlled slave loop, and the control gain of the controlled slave loop in the state-space model is limited by the real-time boundary value of the robust control envelope set.

[0038] Preferably, the method is completed by the control unit of the watershed dispatch center. The control unit performs logical operations based on the real-time hydrological data and water quality sensor signals collected by the lower-level PLC through the industrial bus, and sends the control variables as set values ​​to the execution end of the controlled object.

[0039] Preferably, in step S2, when solving the state-space model, under the premise of satisfying the linear inequality constraint of the robust control envelope set, the balance between the flood control margin of the watershed and the water quality compliance rate of the section is achieved by adjusting the weight coefficients.

[0040] Example 1: When the system faces the dual severe disturbances of sudden non-point source pollution load and rainfall intensity during the flood season, the multi-node water conservancy hub linkage system under the basin dispatch control center needs to maintain the cross-sectional chemical oxygen demand concentration at the standard without touching the water level warning line. At this time, the controlled main loop composed of hydraulic dispatch adjusts the opening of the downstream sluice gates significantly in response to the surge in rainfall runoff, resulting in a severe reconstruction of the physical flow field of water quality regulation. Due to the time scale asymmetry between the spatial transmission lag of hydraulic variables and the temporal continuity of biochemical dissipation processes, the controlled slave loop composed of water quality regulation forcibly tracks the given cross-sectional water quality target in this turbulent hydraulic transition state based on fixed control parameters calibrated according to historical time series data. Its underlying control law generates nonlinear parasitic interference under the drive of rapidly fluctuating state deviation signals, frequently outputting limit full-load commands to the downstream distributed interception and aeration execution ends, directly causing high-frequency opening and closing oscillations of electromechanical equipment and secondary physical impacts on the riverbed.

[0041] To address physical control interference caused by time scale misalignment between heterogeneous actuators, the system control unit invokes a robust control envelope set, generated by solving the input parameter set covering extreme rainfall and runoff and defined by linear inequalities, to establish online adjustment boundaries. Real-time hydrological data acquired by the lower-level programmable logic controller is used to monitor the rate of change of the controlled main loop's state deviation. When the absolute value of the rate of change of the controlled main loop's state deviation is detected... When the value exceeds the preset stability benchmark, the control unit dynamically increases the control dead zone δ(t) of the controlled slave loop according to the preset sensitivity coefficient η. The dynamic adjustment logic follows the formula... Where δ(t) is the control dead zone of the controlled slave loop after dynamic widening at time t, and its dimensions are consistent with the target setpoint of the slave loop. Here, η is the preset minimum dead zone width parameter, and η is a dimensionless sensitivity scaling factor in positive real form. The parameter representing the rate of change of state deviation acquired by the controlled master loop at time t, with dimensions representing the increment of state deviation per unit time, is used to forcibly lock the current online adjustment process of the slave loop within the robust control envelope set. The broadened action threshold directly shields the ineffective high-frequency feedback compensation signal generated by the turbulent hydrodynamic field on the water quality control process. At the physical mechanism evolution level, when the action threshold of the controlled slave loop is broadened by the aforementioned adaptive logic, the bottom-level aeration device and distributed shut-off valve enter a mechanically suspended state for anti-impact protection, maintaining the state in the event of... The steady-state output baseline before disturbance is blocked, and the high-frequency mechanical agitation induced by the frequent full-load opening and closing of heterogeneous actuators is blocked, effectively preventing the bottom sludge of the biochemical reaction tank from being suspended over a large area due to secondary shear force during this transient process. In this way, the system gives way to the huge physical dilution and convection scouring effect of the overall surge water body itself to absorb the non-point source pollution load during the transition period of flood wave advection. It abandons the ineffective attempt to artificially inject interference energy into the system to accelerate the dissipation of surface biochemical substances, and cuts off the chaotic physical superposition response between the overall turbulent hydrodynamic reconstruction and the surface aeration shear force.

[0042] During the transition period when the control dead zone of the controlled slave loop is in a dynamically increasing state, the control unit synchronously converts the extracted state deviation change rate into a negative compensation gain through the deviation decay time constant, and continuously compensates it to the integral regulator of the controlled slave loop. This neutralizes the biochemical integral accumulation that occurs continuously during the period when the control module is suspended. The control unit continuously observes the decay trajectory of the state deviation change rate of the controlled master loop and extracts its exponential change characteristics to update and determine the deviation decay time constant. When the absolute value... When the convergence reaches a value lower than or equal to the preset stable benchmark, the control unit restores the control dead zone to the static preset value. Based on the previously determined deviation decay time constant as the filtering parameter, it performs first-order inertial filtering on the step signal of the given target value received by the controlled slave loop at the dead zone reset time, constructing an asymptotic tracking curve that keeps in sync with the stable velocity of the physical flow field of the controlled main loop. The control variables sent to the execution end of the controlled object smoothly approach the set target along the asymptotic tracking curve. Under the architecture of physical flow field reconstruction and internal residual dynamic modulation of the controller, the system avoids the impact of integral storms and mechanical broadening derivative pulses between heterogeneous control loops. The controlled entity in the watershed is smoothly positioned within the safe control boundary that satisfies multi-objective collaborative optimization under the strong coupling condition of multiple variables. In step S2, the model prediction solution depends on the preset quadratic objective function. The system comprehensively calculates the water quality target tracking error in the prediction time domain and the incremental penalty of the sluice gate opening and interception multiple in the control time domain. The control unit runs the optimization algorithm to minimize the objective function and solve the feasible solution set of the control variables.

[0043] To address the computational challenges posed by the Navier-Stokes partial differential equations within extremely short online sampling periods, the control unit, during online closed-loop operation, did not directly solve the original equations grid by grid. Instead, relying on pre-determined robust envelope constraints, it used Taylor expansion and Jacobian matrix linearization to map the original high-dimensional nonlinear model in situ into a lightweight autoregressive exogenous (ARX) linear time-invariant reduced-order surrogate model near the operating point. The computation matrix register directly loaded this reduced-order surrogate model for matrix multiplication prediction and forward time-domain extrapolation. Combined with the control unit's built-in preprocessed conjugate gradient (PCG) quadratic programming solver, the complex calculus of fluid dynamics was reconstructed into engineering-grade industrial algebraic computation, ensuring that each round of model prediction converged in hard real-time within sub-millisecond timeframes. Regarding the calculation process of the system's underlying state variables and dead zones, the controlled main loop state deviation change rate in step S3... Limited to the time derivative of the water level state deviation in the controlled main loop, based on the rate of change of state deviation. The formula for dynamically adjusting the dead zone δ(t) of the controlled slave loop. In this context, the sensitivity coefficient η is defined as a dimensional transformation function constant, directly converting the spatial physical quantity of hydraulic fluctuations into the widening of the dead zone for biochemical concentration control, thus eliminating dimensional contradictions in the equation transfer of heterogeneous physical quantities. For the controlled slave loop integral controller in step S4, an anti-saturation reset procedure is implemented, with internal operational logic relying on physical variable compensation formulas. The control unit calls and extracts the deviation decay time constant in real time. According to the formula Calculate the output negative compensation gain ΔI. The proportional gain coefficient is preset for the loop integral regulator, and t is the duration of the control dead zone widening state. During each sampling period of the dead zone widening, the control unit directly writes the negative compensation gain ΔI as a bias term into the historical integral accumulation register of the integral regulator. By relying on unidirectional algebraic sum operation to forcibly reduce the control potential energy accumulation caused by the error state retention, the output set value of the controlled entity at the control dead zone reset trigger time is consistent with the steady-state baseline after the actual physical flow field reconstruction.

[0044] Example 2: When the system faces the dual severe disturbances of sudden non-point source pollution load and rainfall intensity during the flood season, a computational fluid dynamics simulation test platform is constructed based on historical data acquisition and monitoring of the control system operation, and internally solves the Navier-Stokes control equations to reproduce the physical flow field reconstruction process of water quality regulation. Gaussian white noise with a signal-to-noise ratio of 20dB and power frequency interference harmonics with a frequency of 50Hz are introduced and superimposed on the real-time hydrological data acquisition port of the controlled main loop to simulate physical control interference under industrial electromagnetic environment. The sampling period is selected as a key control parameter, balancing the real-time acquisition of high-frequency state deviation signals of the turbulent hydrodynamic field with the computational load of the programmable logic controller. The judgment logic follows the principle that when the main frequency of the water level fluctuation in the controlled main loop is greater than 0.5Hz, signal aliasing under the Nyquist sampling theorem is suppressed, and the sampling period tends to reach its lower limit. Based on this judgment logic, the sampling period is set to 100ms, and a preset minimum dead zone width parameter is also set. The rainfall intensity was set as 2.0 mg / L, and the rainfall intensity was used as the core parameter to characterize the system disturbance intensity. Three gradient test conditions were set: 15.5 mm / h, 32.4 mm / h, and 58.6 mm / h. A control group 1 was set up, which only used the model predictive control algorithm with a fixed dead zone. A control group 2 was set up, which started the dead zone dynamic adjustment logic and set the sensitivity scaling factor η to deviate from the preset envelope boundary to 15.0. An experimental group was set up, which adopted the controlled slave loop control of the dead zone δ(t) dynamic adjustment logic and set the sensitivity scaling factor η to 3.5.

[0045] When the input rainfall intensity was 32.4 mm / h, the platform monitored that the peak value of the unfiltered controlled master loop state deviation change rate reached 0.42 m / min. The experimental control unit established the online adjustment boundary in real time according to the formula, widening the control dead zone δ(t) of the controlled slave loop from 2.0 mg / L to 3.47 mg / L. For the 2.0 mg / L static minimum dead zone width parameter called by the controlled slave loop, this boundary value was pre-calculated offline by the system based on steady-state dry season baseline data. The inference was based on the inherent optical white noise amplitude of the optical ultraviolet water quality sensor under the clear water calibration state, and superimposed the long-term statistical root mean square variance envelope of small concentration fluctuations in the biochemical reaction flow field under natural conditions. The system used this inferred boundary to establish the physical noise shielding threshold. This was to ensure that the water quality feedback control loop would not generate ineffective frequent fine-tuning integral actions when subjected to routine harmless random measurement disturbances. During the transition period when the control dead zone was in a dynamically widening state, the experimental group synchronously extracted the state deviation change rate and converted it into a negative compensation gain, which was continuously compensated to the integral regulator of the controlled slave loop. The measured data showed that the biochemical integral accumulation of the experimental group converged to 12.4 mg / L·s. The control group 1 lacked this adaptive dead zone widening action, which caused its integral accumulation to climb to 85.6 mg / L·s, causing the lower sluice gate opening to experience 14 mechanical oscillations under full load commands within 10 minutes. Under the drive of the extreme value sensitivity coefficient, the control dead zone of the control group 2 widened to 8.30 mg / L, causing the controlled slave loop to lose its ability to track the cross-sectional chemical oxygen demand concentration.

[0046] Extract the measured physical characterization values ​​of the test group under a rainfall intensity of 58.6 mm / h, and the absolute value of the rate of change of the controlled main loop state deviation. When the convergence rate drops to below or equal to the preset stable baseline of 0.05 m / min, the control unit restores the control dead zone to the static preset value of 2.0 mg / L. The control unit uses the previously determined deviation decay time constant as a filtering parameter to perform first-order inertial filtering on the target value step signal. The test data exhibits a non-linear pattern. When the rainfall intensity increases from 15.5 mm / h to 32.4 mm / h, the number of mechanical oscillations of the sluice gate in the test group remains within 2, and the steady-state tracking error of the cross-sectional chemical oxygen demand concentration is 3.1%. When the rainfall intensity climbs to 58.6 mm / h... When a secondary physical impact is induced in the riverbed, the steady-state tracking error is 3.5%, indicating that the robust control envelope set has a stable performance plateau region when resisting extreme disturbances; the peak chemical oxygen demand concentration of control group 2 under the corresponding working condition exceeds 45.2 mg / L; the above data comparison shows that the two technical actions of dynamically increasing the control dead zone and compensating for negative gain have a unidirectional causal relationship. By widening the action threshold to shield the ineffective high-frequency feedback compensation signal, and by relying on the negative compensation gain to neutralize the accumulation of biochemical integral, a physical decoupling path for the time scale between heterogeneous control loops is constructed.

[0047] Example 3: This example combines Figures 1 to 2 This paper describes a multi-objective collaborative optimization control method for watershed ecological water resources, such as... Figure 1 As shown, in the overall architecture, the disturbance parameter sequence is first used to obtain the disturbance characteristics of the controlled domain and input them into the state space model, which contains the controlled main loop and the controlled slave loop. Two control solution paths are derived in parallel from this state space model. The first path obtains the feasible solution set of the control variables through model prediction and imports the robust control envelope set downward to extract the common intersection of the feasible solution sets. The second path monitors the controlled main loop in real time through the rate of change of state deviation and extends into two branches: one is to pass it downward to the control dead zone to dynamically adjust according to the state of the main loop, and the control dead zone is pointed to the robust control envelope set through the dashed line unidirectional feedback to implement boundary locking; the other is to pass it to the deviation decay time constant path to extract the decay characteristics of the main loop, and then transmit it to the integral regulator for anti-saturation reset and supplemented by the first-order inertial filter target value.

[0048] like Figure 2 As shown, the system control program starts from the standby / initialization state. After receiving the system start command, it enters the parameter identification and model building stage to identify the system matrix and disturbances. During this building process, if a system fault / forced stop is triggered, the system will be directly guided to the shutdown state via the dashed path. Under normal circumstances, when the state space model is completed, the process proceeds to the prediction of the feasible control solution set to calculate the control solution that satisfies the constraints. This prediction step obtains multiple feasible solution sets at the output and injects them into the robust control envelope set locking to extract the common intersection as a constraint. At the same time, the prediction output also triggers the main loop monitoring and dead zone adjustment downward to monitor the deviation, dynamically adjust the dead zone, and feed back upward to the robust control envelope set locking stage under the action of dead zone adjustment triggered by the rate of change of state deviation. Subsequently, under the condition constraint of the control variable rate of change locking, the system executes the regulator anti-saturation reset, adjusts the integral term, updates the target value, and finally performs a state determination. If the system shutdown condition is not met, it will return to the parameter identification and model building stage for re-execution. If the shutdown condition is met, it will finally switch to the system shutdown state.

[0049] Example 4: When the system faces the initial parameter calibration condition of a multi-node hydraulic hub linkage system, the control unit establishes the quantized values ​​of the preset stability benchmark and deviation decay time constant, initializes the dynamic adjustment mechanism for the control dead zone and the anti-saturation reset mechanism for the integral regulator, and avoids the response delay differences and control divergence caused by relying on empirical preset parameters. To meet the technical requirements of the control parameter setting, the control unit inputs a standard step hydraulic load disturbance to the controlled main loop composed of hydraulic scheduling; the level sensor continuously collects the water level deviation response time series of the controlled main loop; the control unit extracts the first... In the secondary decay oscillation interval, the peak deviation and steady-state error band parameters of this interval are extracted. The control unit multiplies the steady-state error band parameters by the natural oscillation frequency of the controlled main circuit to calculate the reference rate parameter, which is used as the preset stable reference. In one engineering configuration, the measured steady-state error band value is 0.02m, and the natural oscillation frequency of the controlled main circuit is 2.5 times per minute. Based on this, the control unit calculates the preset stable reference as 0.05m / min. During the transition period when the control dead zone is in a dynamically increased state, the control unit initiates the online extraction logic of the deviation decay time constant, and uses the real-time acquired rate of change of the state deviation of the controlled main circuit. Store in a sliding sampling window containing 50 sampling periods.

[0050] The control unit calculates the natural logarithm of the state deviation change rate sequence within the sliding sampling window, and uses the least squares method to solve the linear fitting slope of the sequence with the natural logarithm. The control unit takes the reciprocal of the absolute value of the fitting slope as the deviation decay time constant. The control unit writes this deviation decay time constant into the integral gain compensation register of the lower-level programmable logic controller, and drives the programmable logic controller to calculate the negative compensation gain based on the deviation decay time constant. The programmable logic controller writes the negative compensation gain into the historical accumulation storage address of the integral regulator, and neutralizes the corresponding biochemical integral accumulation based on the numerical scale of the negative compensation gain. The control unit measures the preset stability benchmark through the standard step response, and extracts the deviation decay time constant by combining the sliding sampling window and the logarithmic linear fitting algorithm, binding the parameter setting of the water quality regulation control law to the dynamic decay attribute of the hydraulic scheduling physical flow field. The controlled slave loop composed of water quality regulation obtains the boundary judgment threshold and anti-saturation reset decay factor based on the physical flow field reconstruction rate. The system relies on the underlying data flow logic to reduce the random pulse interference caused by the empirical parameter setting, and constrains the online regulation process under the multivariable strong coupling condition to be located within the robust control envelope set.

[0051] Example 5: When the system faces initial operating conditions such as the first network deployment of multi-node water conservancy hubs or long-period offset of external disturbance baselines, the control unit executes offline calibration procedures to construct a robust control envelope set, solving the optimization divergence problem of a multivariable strongly coupled system in the absence of prior boundary conditions; the dispatch center extracts historical time-series data of rainfall intensity and non-point source pollution load from the target watershed over the past ten years with hourly time resolution, and uses the Monte Carlo random sampling algorithm to discretize and extract extreme rainfall intervals, generating an extreme value input parameter set containing 1000 samples; the control unit injects the extreme value input parameter set into a system based on the Navier-Stokes equations. In solving the hydrodynamic state-space model of the kernel, within a simulation cycle of 5 minutes, the gradient descent optimization algorithm iteratively solves the feasible solution sequence for the sluice gate opening and distributed interception multiple that satisfy the dual hard constraints of water level warning line constraint and cross-sectional chemical oxygen demand concentration compliance. In the specific construction process of this hydrodynamic state-space model, the control unit uses the eigenorthogonal decomposition method to reduce the order of the high-dimensional Navier-Stokes partial differential equation flow field, extracts the preceding core energy modes, and establishes a standard continuous linearized matrix differential equation jointly characterized by the state vector x(t), control vector u(t), and external disturbance vector d(t). The state vector x(t) is composed of the overall cross-sectional flow rate, water level parameters, and surface cross-sectional chemical oxygen demand concentration. The system state matrix is... The system incorporates a physical spatial gradient mapping of the water flow convection diffusion coefficient. The control input matrix B represents the mechanical transmission gain boundary of the lower-level sluice gate opening and distributed interception multiple. The disturbance mapping matrix E is associated with the extreme rainfall input. This achieves physical dimensionality reduction and numerical coupling of the overall hydraulic parameters and biochemical surface parameters within a unified underlying matrix calculation framework. The control unit extracts the intersection region of all feasible solution sequences in the multidimensional control variable space, collects the coordinates of the outer envelope vertices of this intersection region, and uses a hyperplane fitting logic based on the convex hull algorithm to transform these discrete vertex coordinates into a multidimensional parameter space composed of linear inequalities. The control unit then stores the linear inequality coefficient matrix of this multidimensional parameter space in the non-volatile memory of the lower-level programmable logic controller, completing the offline calibration of the robust control envelope set.

[0052] During the pre-deployment calibration phase of the control unit entering the controlled main loop, the control unit sends a frequency sweep test excitation signal with a frequency range of 0.1Hz to 2.0Hz to the downstream sluice gate actuator, simultaneously acquiring the residual sequence of chemical oxygen demand (COD) concentration fluctuations in the corresponding frequency band of the biochemical reaction tank. The control unit calculates the power spectral density of the concentration fluctuation residual sequence and extracts the center bandwidth parameter with an attenuation amplitude exceeding -3dB. The control unit takes the reciprocal of this center bandwidth parameter and multiplies it by the maximum deviation change rate of the controlled main loop obtained in the previous testing phase. The product is set as the sensitivity coefficient to establish the calibration mapping relationship. In this calibration calculation, the reciprocal of the center bandwidth parameter objectively characterizes the natural physical relaxation time of the surface flow field of the biochemical tank in response to disturbances during the frequency-to-time domain conversion. The control unit performs a product operation on this natural physical relaxation time constant and the maximum deviation change rate characterizing the overall hydraulic space displacement rate. Essentially, the limiting characteristic convective mass transfer length scale of a specific controlled unit under extreme operating conditions is derived in the physical space scale. Under the assumption of sufficient mixing based on the Reynolds transport equation, this physical mass transfer length scale shows a highly positive linear correlation with the equivalent fluctuation amplitude of the cross-sectional chemical oxygen demand concentration. This formula constructs a cross-physical domain bridge connecting time-frequency parameters and biochemical response thresholds, providing parameter support that conforms to the underlying transport logic of fluid mechanics for subsequent dead zone widening calculations. The control unit writes the calibrated sensitivity coefficient η into the underlying register. This coefficient is called in actual operation to limit the dynamic widening logic of the control dead zone δ(t), transforming the robust control envelope set calculated offline into a safety constraint for online physical execution. The system relies on parameter calibration procedures to cut off the high-frequency oscillation feedback link caused by empirical presets. The controlled entity is positioned inside the stable control boundary that satisfies the multi-objective collaborative optimization of watershed ecological water resources in the initial state of commissioning.

[0053] Example 6: When the system faces the initial operating condition of connecting a newly built water conservancy hub node to the control network and lacking a priori dynamic response model, the control unit initiates a pre-deployment calibration procedure to construct a multi-objective collaborative optimization baseline model and establish a quantitative evolution path for control parameter settings; the dispatch center extracts historical water level fluctuation time series data and historical chemical oxygen demand (COD) concentration change sequences within a specific hydrological cycle; the control unit combines the peak variance of the historical water level fluctuation time series data and the compliance rate of the historical COD concentration change sequences into a dual evaluation function; the control unit uses a gradient descent algorithm to traverse and optimize different combinations of model prediction time-domain parameters and control time-domain parameters in the parameter space; the control unit collects the steady-state deviation amplitude of the main loop and the steady-state deviation amplitude of the slave loop generated by the derivation of each parameter combination in the state-space model; the control unit extracts the target parameter combination that minimizes the weighted sum of the dual evaluation function and writes the target parameter combination into the operation matrix register of the programmable logic controller.

[0054] The control unit sends step pulse commands with an amplitude of 10% of the full range to the downstream sluice gate and distributed intercepting valve via the communication bus, and simultaneously collects the actual displacement stroke of the relevant mechanisms and the position feedback time constant upon reaching the steady state. The control unit extracts the position feedback time constant as a hardware delay parameter and directly inputs the hardware delay parameter into the boundary constraint matrix of the multi-objective collaborative optimization baseline model. The system establishes the underlying model parameters based on the offline calibration process that includes traversal optimization and hardware response feedback measurement, and binds the initial boundary of the multi-objective collaborative optimization control law to the hydrological statistical characteristics and the inherent mechanical delay properties of the controlled entity, constraining the hydraulic scheduling command and water quality regulation command to be positioned within the preset optimization control boundary at the initial moment of commissioning.

[0055] It will be apparent to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above, and that the present invention can be implemented in other specific forms without departing from the spirit or essential characteristics of the present invention.

[0056] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention.

Claims

1. A multi-objective collaborative optimization control method for watershed ecological water resources, characterized in that, Includes the following steps: Step S1: Construct a state-space model of the hydrodynamic and water quality relationship of the controlled unit of the watershed. The state-space model includes a controlled main loop composed of hydraulic scheduling and a controlled slave loop composed of water quality regulation. Obtain the perturbation parameter sequence that characterizes the perturbation features of the controlled domain. Step S2: Input the disturbance parameter sequence into the state space model, obtain the feasible solution set of control variables under different working conditions through model prediction and calculation, and extract the common intersection of the feasible solution set according to the preset threshold constraint to establish the robust control envelope set. Step S3: Monitor the rate of change of state deviation of the controlled master loop in real time, and dynamically adjust the control dead zone of the controlled slave loop according to the rate of change of state deviation. When the rate of change of state deviation is greater than the preset stable benchmark, increase the control dead zone and lock the online adjustment process within the robust control envelope set. Step S4: Extract the deviation decay time constant of the controlled master circuit, perform anti-saturation reset on the integral regulator of the controlled slave circuit according to the deviation decay time constant, and perform first-order inertial filtering on the given target value of the controlled slave circuit according to the filter factor determined by the deviation decay time constant.

2. The method for multi-objective collaborative optimization control of watershed ecological water resources according to claim 1, characterized in that, Step S2, establishing the robust control envelope set, includes: Step S21, using a random sampling algorithm to discretize the disturbance parameter sequence to generate an input parameter set covering rainfall, runoff, and drought / flood conditions; Step S22, calling the model predictive control algorithm to solve the state-space model in each sampling period to obtain the control law family that has not touched the water level warning line and pollutant concentration limit under the disturbance of each input parameter set; Step S23, extracting the common boundary of the control law family and determining the parameter space defined by the linear inequality constraint as the robust control envelope set.

3. The method for multi-objective collaborative optimization control of watershed ecological water resources according to claim 1, characterized in that, The state-space model uses the flow rate, water level, and COD concentration at the cross section of the watershed as state variables, the gate opening and the distributed interception ratio as control variables, and the non-point source pollution load and rainfall intensity as external disturbance variables.

4. The method for multi-objective collaborative optimization control of watershed ecological water resources according to claim 1, characterized in that, The step S3 of dynamically adjusting the control dead zone of the controlled slave loop includes: step S31, calculating the absolute value of the rate of change of the state deviation of the controlled master loop; step S32, when the absolute value exceeds the preset stability benchmark, increasing the control dead zone of the controlled slave loop according to the preset sensitivity coefficient; step S33, when the absolute value is lower than or equal to the preset stability benchmark, restoring the control dead zone to the static preset value.

5. The method for multi-objective collaborative optimization control of watershed ecological water resources according to claim 1, characterized in that, Step S4 involves performing an anti-saturation reset on the integral regulator of the controlled slave loop, which includes: Step S41, observing the decay trajectory of the rate of change of the state deviation of the controlled master loop, extracting the exponential change characteristics of the decay trajectory to determine the deviation decay time constant; Step S42, converting the rate of change of the state deviation into a negative compensation gain through the deviation decay time constant, and continuously compensating it to the integral regulator of the controlled slave loop to neutralize the integral accumulation generated during the control dead zone adjustment.

6. The method for multi-objective collaborative optimization control of watershed ecological water resources according to claim 1, characterized in that, Step S4 involves performing first-order inertial filtering on the given target value of the controlled slave loop, which includes: step S43, obtaining the target value step signal of the controlled slave loop at the control dead zone reset time; and step S44, using the deviation decay time constant as a filtering parameter to smoothly transform the target value step signal and construct an asymptotic tracking curve that is synchronized with the stable velocity of the physical flow field of the controlled master loop.

7. The method for multi-objective collaborative optimization control of watershed ecological water resources according to claim 1, characterized in that, The control frequency of the controlled master loop is higher than that of the controlled slave loop, and the control gain of the controlled slave loop in the state-space model is limited by the real-time boundary value of the robust control envelope set.

8. The method for multi-objective collaborative optimization control of watershed ecological water resources according to claim 1, characterized in that, The method is accomplished through the control unit of the watershed dispatch center. The control unit performs logical operations based on the real-time hydrological data and water quality sensor signals collected by the lower-level PLC through the industrial bus, and sends the control variables as set values ​​to the execution end of the controlled object.

9. A multi-objective collaborative optimization control method for watershed ecological water resources according to claim 1, characterized in that, In step S2, when solving the state-space model, under the premise of satisfying the linear inequality constraint of the robust control envelope set, the balance between the flood control margin of the watershed and the water quality compliance rate of the cross section is achieved by adjusting the weight coefficients.

Images