Intelligent optimization control method for aeration quantity and internal reflux quantity based on MABR process

Abstract

This invention relates to the field of intelligent optimization control technology, specifically to an intelligent optimization control method for aeration volume and internal recirculation flow based on MABR process. The method includes the following steps: Kalman filtering is used to derive the predicted mean value by substituting it into the degradation equation; the difference in dissolved oxygen is compared to determine if the scalar exceeds the limit; the gain coefficient is simultaneously calculated to obtain the biofilm mass transfer; multiplication is performed to update the network connection weights according to a 200-iteration cycle to generate air-water synergistic instructions; a genetic algorithm is used to fuse the mass transfer and synergistic instruction values; power consumption and head scalars are extracted to construct the coordinate accumulation dimension difference; the rotational speed is encoded as a voltage analog wave and the air pressure is converted to a duty cycle level to obtain the driving pulse waveform; a two-dimensional distributed coordinate matrix is ​​set as the addressing space to strip the power consumption value and head parameter to calculate the congestion distance, avoiding the local valley dilemma of conventional gradient optimization methods, thereby reducing the power consumption of electromechanical equipment by 15% and ensuring that the effluent chemical parameters are within the specified concentration boundaries.

Description

Technical Field

[0001] This invention relates to the field of intelligent optimization control technology, and in particular to an intelligent optimization control method for aeration volume and internal recirculation flow based on MABR process. Background Technology

[0002] Intelligent optimization control specifically involves using nonlinear approximation algorithms and multivariable decoupling control theory to establish a mathematical analytical mapping model for multi-input, multi-output nonlinear biochemical reaction systems. The specific operational steps include introducing biochemical mechanism equations to construct a state-space matrix, using a gradient descent optimization algorithm to solve for the convergent solution of the objective function, and then outputting quantitative control commands to the frequency converter and pneumatic valve actuators. This approach abandons the traditional static adjustment mode based on empirically set deviation thresholds and instead adopts an optimization framework that combines process dynamics model prediction and compensation with nonlinear constraint handling capabilities, enabling multiple physicochemical variable indicators to converge to the preset environmental protection standard and specification boundaries.

[0003] The intelligent optimization control method for aeration volume and internal recirculation flow based on MABR technology is a multi-variable collaborative regulation scheme that couples oxygen mass transfer on the biofilm surface, controlled logic, and nitrate circulation distribution strategy in the hydraulic retention zone. The scheme aims to break through the limitations of traditional operation with fixed oxygen supply and fixed water recirculation ratio. It predicts theoretical oxygen demand by introducing a microbial growth kinetic model and solves the mathematical constraint model of recirculation pump power consumption in the low energy consumption range using a sequential quadratic programming algorithm. The scheme maintains the effluent ammonia nitrogen concentration and the anoxic tank nitrate concentration within the specified threshold. It outputs blower frequency reduction commands and water pump opening control signals, thereby reducing aeration power consumption and internal recirculation power consumption, and ensuring that the effluent total nitrogen concentration remains stable within the preset safe concentration range.

[0004] The mathematical analytical mapping model for multi-input multi-output systems, which employs nonlinear approximation algorithms and multivariable decoupled control theory, suffers from several underlying operational defects. The state-space matrix constructed based on conventional biochemical mechanism equations is highly dependent on the assumption of a steady-state environment. It faces severe computational oscillations due to nonlinear abrupt changes in upstream influent water quality parameters. The gradient descent algorithm for solving the objective function's convergence process is heavily dependent on the accuracy of the initial parameter assignments, and the optimization iterative evolution trajectory is prone to stagnation and getting trapped in local minima. A combined sequential quadratic programming algorithm is used to solve the nonlinear constraint model, employing a fourth-order Lagrange algorithm. Multiplier calculus calculations consume a large amount of computing resources of the underlying processor. Facing the multidimensional spatiotemporal evolution of biochemical parameters superposition calculations, the system matrix singularity failure is caused by the unsolvable fault. The static mathematical analytical model continues to use the convergence solution of the previous conventional stable period and issues erroneous fan control commands, causing the oxygen supply rate of the micropores inside the biochemical reaction tank to be delayed by more than 120 seconds. The core dissolved oxygen scalar of the reaction system drops sharply, falling below the safety maintenance limit of 2.0 mg / L. Two million units of active aerobic nitrifying bacteria encounter severe hypoxia stress and enter an irreversible death cycle, which disrupts the nitrate circulation and distribution balance inside the anoxic tank and causes illegal discharge accidents. Summary of the Invention

[0005] The purpose of this invention is to overcome the shortcomings of existing technologies and propose an intelligent optimization control method for aeration volume and internal recirculation flow based on MABR process.

[0006] To achieve the above objectives, the present invention adopts the following technical solution: an intelligent optimization control method for aeration volume and internal recirculation flow based on MABR process, comprising the following steps: Step 1: Based on the sensor readings of the biochemical system monitoring nodes, read the oxygen demand and nitrate concentration of the return weir, extract the flow velocity variable of the hollow membrane tube, arrange the concentration values ​​to construct a column vector array, splice the column vector array and the flow velocity variable to merge the operation elements, and establish the spatiotemporal load tensor; Step 2: Based on the spatiotemporal load tensor, decompose the multidimensional tensor elements and extract the diagonal values ​​to precipitate discrete sampling points. After deriving the predicted mean by substituting it into the equation through Kalman filtering, compare it with the scalar of dissolved oxygen difference, determine the scalar exceeding the limit, and perform simultaneous calculation of gain coefficient and mean to obtain biofilm transfer mass. Step 3: Based on the biofilm mass transfer, splice the submersible pump speed and valve opening, sum and output the operation command and extract the return evaluation value, subtract the evaluation value from the baseline to calculate the difference scalar, determine the scalar out of bounds and update the network connection weight by multiplication, and generate the air-water synergy command. Step 4: Based on the biofilm mass transfer and the gas-water synergy command, the numerical values ​​and commands are fused using a genetic algorithm to extract power consumption and head scalars to construct coordinates, the dimensional difference is accumulated to output the distribution scalar, the lower limit scalar coordinates are removed and the outer layer points are retained to establish the frontier control parameters; Step 5: Based on the aforementioned front-end control parameters, select independent points from the outer boundary coordinates, separate the output speed and micro-hole air pressure values, encode the speed as a voltage pulse and send it to the inverter contacts, convert the air pressure into a duty cycle level and send it to the drive air valve terminal to obtain the drive pulse waveform.

[0007] As a further aspect of the present invention, the spatiotemporal load tensor includes a column vector array of oxygen demand and nitrate concentration, a discrete variable of flow velocity inside the hollow membrane tube, and a time step dimension constant parameter; the biofilm mass transfer includes the predicted mean of matrix degradation, a scalar of dissolved oxygen differential distribution characteristics, and a gain coefficient; the gas-water synergistic command includes a submersible pump operating speed variable, a gas supply valve adjustment opening parameter, and a network connection weight; the front-edge control parameter includes operating power consumption value, delivery head parameter, a spatial distribution discrete scalar, and the coordinates of peripheral edge scattered points; and the drive pulse waveform includes a frequency converter contact voltage analog wave and a drive air valve terminal duty cycle level.

[0008] As a further aspect of the present invention, the specific steps for establishing the spatiotemporal load tensor are as follows: Based on the sensor readings of the biochemical system monitoring node, the oxygen demand and nitrate concentration are read in a polling manner, the discrete variable of the flow velocity inside the hollow membrane tube is extracted, the concentration values ​​are arranged into a vertical array, and the vertical array and the flow velocity variable are spliced ​​horizontally to establish a multidimensional biochemical parameter column vector. Based on the multidimensional biochemical parameter column vector, the numerical discrete elements inside the address vector are traversed, a preset time step dimension constant parameter is introduced, the numerical elements and the time dimension parameter are spliced ​​together, and the tensor boundary is expanded by horizontally filling blank nodes to establish the spatiotemporal load tensor.

[0009] As a further aspect of the present invention, the specific steps for obtaining the mass transfer of the biomembrane are as follows: Based on the spatiotemporal load tensor, the internal node elements of the multidimensional tensor are decomposed, redundant values ​​off-diagonal are removed, core numerical features of the main diagonal are extracted, a preset spatial environment distribution constant is associated, a reconstructed multidimensional feature array vector is extracted, and discrete load sampling points are established. Based on the discrete load sampling points, multidimensional values ​​inside the nodes are extracted, Kalman filtering is used to substitute the variable features into the degradation equation, multidimensional spatial distribution parameters are transmitted and node internal feature indicators are integrated, the corresponding environmental boundary is calculated, the array mean is derived, and the matrix degradation prediction mean is obtained. Based on the predicted mean of matrix degradation, the dissolved oxygen value of the microporous aeration disc is read and compared with the safety threshold to determine the state deviation. The corresponding differential distribution feature scalar is extracted. When the scalar exceeds the limit, the gain coefficient is captured. The gain coefficient and the predicted mean are combined to obtain the biofilm transfer mass.

[0010] As a further aspect of the present invention, the Kalman filter extracts multidimensional numerical parameters within discrete load sampling points to construct an initial state covariance array, substitutes the multidimensional numerical parameters into the matrix degradation equation, calculates the state transition analytical matrix vector, integrates multidimensional spatial distribution parameters to perform prior prediction deduction, replaces the initial discrete variable features within the equation, integrates matrix consumption and biological growth dynamics characteristic indicators within the nodes, calculates the state propagation evolution covariance matrix, calculates the specific environmental constraint boundary of the associated biochemical system, accumulates each state vector and calculates the mathematical expectation quotient, and derives the prior prediction mean of the generated array.

[0011] As a further aspect of the present invention, the process involves disassembling the internal node elements of the multidimensional tensor, extracting the spatial, temporal, and concentration dimension values ​​of the third-order spatiotemporal load tensor, performing a dimensionality reduction matrix expansion operation according to the modal expansion rule, and establishing multiple two-dimensional modal expansion matrices. A singular value decomposition matrix operation formula is introduced to calculate the multiple two-dimensional modal expansion matrices separately, generating multiple left singular vector matrices and separating the corresponding singular value scalars. The left singular vector matrices are correlated with the original third-order spatiotemporal load tensor to perform a modal multiplication operation, deriving and constructing an orthogonal factor array, and comparing the singular value scalars with a preset truncation threshold. Dimensional arrays corresponding to scalars below the threshold are truncated and removed to generate a condensed principal component core tensor, thereby completing the disassembly of the internal node elements of the multidimensional tensor.

[0012] As a further aspect of the present invention, the specific steps for generating the gas-water synergistic command are as follows: Based on the biofilm mass transfer, the submersible pump operating speed is read and the air supply valve adjustment opening is extracted. The associated parameters are spliced ​​to construct a vector, the matrix node accumulation elements are substituted, the trial command is sent to the control end, the underlying hardware operation indicators are extracted, and an operation return evaluation scalar is established. Based on the operational return evaluation scalar, the equipment operating baseline is retrieved, the baseline and evaluation scalar are subtracted, the corresponding numerical difference feature is calculated, the difference feature is compared with the safety boundary and the feature overstepping direction is determined, the internal node connection weights are overwritten, and the gas-water coordination command is generated.

[0013] As a further aspect of the present invention, the specific steps for establishing the aforementioned frontier control parameters are as follows: Based on the biofilm mass transfer and the gas-water synergy command, the features and command variables are fused, the operating power consumption value is extracted and the conveying head parameter is captured, the power consumption value and the head parameter are associated to construct a planar coordinate, the bottom discrete node elements are aggregated, and a two-dimensional distributed coordinate matrix is ​​established. Based on the dual-dimensional distributed coordinate matrix, a genetic algorithm is used to traverse and address to read the position coordinates of the bottom node, the features of the horizontal and vertical axes are stripped, the values ​​of adjacent nodes in the same dimension are subtracted and the absolute difference is extracted, and the absolute difference of each item is accumulated to output the crowding representation, thus establishing a spatially distributed discrete scalar. Based on the spatially distributed discrete scalar, the coordinates of the scalar-bound nodes are retrieved, the discrete scalar is compared with the lower limit truncation threshold and overlapping nodes below the lower limit threshold are removed, the remaining peripheral edge scattered points of the remaining level are truncated, the effective coordinate elements are summarized and retained, and the frontier control parameters are established.

[0014] As a further aspect of the present invention, the genetic algorithm sets a two-dimensional distributed coordinate matrix as the address space, constructs a traversal pointer array that maps the two-dimensional coordinate plane, triggers step-by-step traversal actions according to the index order of the matrix rows and columns, locates the underlying mapping storage location corresponding to the discrete node, reads the planar position coordinate value of the individual pointed to by the current pointer, converts the read and extracted node coordinates into the spatial gene sequence encoding of the initial population individual, and outputs an address sequence array containing multiple node position features.

[0015] As a further aspect of the present invention, the specific steps for obtaining the driving pulse waveform are as follows: Based on the aforementioned frontier control parameters, the two-dimensional coordinate point matrix of the outer boundary is screened, the independent node position mapping elements are extracted, the node-bound hardware operation parameters are retrieved, the target speed variable of the submersible pump is split, the air pressure control value of the microporous air valve is separated, and the electromechanical drive parameter values ​​are established. Based on the electromechanical drive parameters, the rotational speed and micro-hole air pressure parameters are read, the rotational speed characteristic is encoded as a voltage analog wave, a pulse is sent to the inverter electrical contacts, the air pressure value is converted into a duty cycle level, and a level command is sent to the hardware driver to obtain the drive pulse waveform.

[0016] Compared with the prior art, the advantages and positive effects of the present invention are as follows: In this invention, the predicted mean is derived by substituting Kalman filtering into the degradation equation, the scalar value of the dissolved oxygen difference is compared to determine the scalar value exceeding the limit, the gain coefficient is combined to obtain the biofilm transfer mass, the submersible pump speed and valve opening are spliced ​​together to sum and output the operation command, and the evaluation value is subtracted from the baseline to calculate the difference scalar value, and the network connection weight is updated by multiplication according to a two-hundred-iteration cycle to generate air-water synergistic command. In this invention, the transmission quality and cooperative command values ​​are fused by a genetic algorithm, the power consumption and head scalar are extracted to construct the coordinate accumulation dimension difference, the outer boundary coordinates are screened to extract independent points to split the rotation speed and micro-hole air pressure values, the rotation speed is encoded as a voltage analog wave and the air pressure is converted into a duty cycle level to obtain the driving pulse waveform. In this invention, by setting a two-dimensional distributed coordinate matrix as the addressing space to separate the power consumption value and the head parameter to calculate the congestion distance, the conventional gradient optimization method avoids the dilemma of falling into local valleys. It outputs multiple driving waveforms located at the lowest energy consumption Pareto front, thereby reducing the power consumption of electromechanical equipment by 15% and ensuring that the chemical parameters of the effluent are within the specified concentration boundaries. Attached Figure Description

[0017] Figure 1 This is a schematic diagram of the main steps of the present invention. Detailed Implementation

[0018] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention.

[0019] Example 1 Please see Figure 1 This invention provides a technical solution: an intelligent optimization control method for aeration volume and internal recirculation flow based on MABR process, comprising the following steps: Step 1: Based on the sensor readings of the biochemical system monitoring nodes, read the oxygen demand and nitrate concentration of the return weir, extract the flow velocity variable of the hollow membrane tube, arrange the concentration values ​​to construct a column vector array, splice the column vector array and the flow velocity variable to merge the operation elements, and establish the spatiotemporal load tensor; Step 2: Based on the spatiotemporal load tensor, decompose the multidimensional tensor elements and extract the diagonal values ​​to precipitate discrete sampling points. After deriving the predicted mean by substituting it into the equation through Kalman filtering, compare it with the scalar of dissolved oxygen difference, determine the scalar exceeding the limit, and perform simultaneous calculation of gain coefficient and mean to obtain the biofilm transfer mass. Step 3: Based on the biofilm mass transfer, splice the submersible pump speed and valve opening, sum and output the operation command and extract the return evaluation value, subtract the evaluation value from the baseline to calculate the difference scalar, determine the scalar out of bounds and update the network connection weight by multiplication, and generate the air-water synergy command. Step 4: Based on the biofilm mass transfer and gas-water synergy command, the numerical values ​​and commands are fused through a genetic algorithm to extract the power consumption and head scalars to construct coordinates. The dimensional difference is accumulated to output the distribution scalar. The lower limit scalar coordinates are removed and the outer layer points are retained to establish the frontier control parameters. Step 5: Based on the front-end control parameters, select independent points from the outer boundary coordinates, separate the output speed and micro-hole air pressure values, encode the speed as a voltage pulse and send it to the inverter contacts, convert the air pressure into a duty cycle level and send it to the drive air valve terminal to obtain the drive pulse waveform.

[0020] The spatiotemporal load tensor includes a column vector array of oxygen demand and nitrate concentration, a discrete variable of flow velocity inside the hollow membrane tube, and a time step dimension constant parameter. The biofilm mass transfer includes the predicted mean of matrix degradation, a scalar of the differential distribution characteristics of dissolved oxygen, and a gain coefficient. The gas-water coordination command includes the submersible pump operating speed variable, the air supply valve adjustment opening parameter, and the network connection weight. The front-end control parameters include the operating power consumption value, the delivery head parameter, the spatial distribution discrete scalar, and the coordinates of the peripheral edge scattered points. The drive pulse waveform includes the inverter contact voltage analog wave and the duty cycle level of the drive air valve terminal.

[0021] The specific steps for establishing the spatiotemporal load tensor are as follows: Based on the sensor readings of the biochemical system monitoring node, a multidimensional array recombination and splicing method is adopted. The underlying hardware communication protocol is called to poll and read the oxygen demand and nitrate concentration. The baud rate parameter of the programmable logic controller is set to 9600 bits per second and the data bit length parameter is specified to 8 bits. The oxygen demand and nitrate concentration values ​​are captured at a period of 500 milliseconds and written to the continuous address area of ​​memory. The floating-point register is accessed to read the discrete variable of the flow velocity inside the hollow membrane tube and forcibly converted to a 32-bit single-precision floating-point value. The dimension deformation command is called to operate the concentration value inside the continuous address area of ​​memory. The tensor dimension recombination parameter is set to 1024 rows and 1 column to construct a vertical numerical array model. The column cascading binding instruction is called to horizontally splice the above vertical numerical array model and the single-precision flow velocity floating-point value. The spatial cascading coordinate axis index constant is set to 1 and the memory broadcast alignment mechanism is enabled. The flow velocity floating-point value is copied bit by bit in the row index increment order to fill the dimension extension bit at the end of the concentration array, generating a multidimensional biochemical parameter column vector. Based on a multidimensional biochemical parameter column vector, a tensor dimension broadcast filling method is adopted. The memory step traversal pointer is called to address the discrete numerical elements inside the multidimensional biochemical parameter column vector. The initial offset of the traversal pointer is set to 0 bytes and the loop jump step size parameter is set to 4 bytes. A preset time step dimension constant parameter is introduced. The preset time step dimension constant parameter is obtained by reading the accumulated value of the main frequency clock register of the programmable logic controller system and dividing it by the set total number of sampling times of 2000. The specific scalar is specified as 1.5 seconds. The three-dimensional space expansion instruction is called to concatenate the discrete numerical elements with the 1.5-second time dimension parameter. The target expanded tensor rank parameter is set to 3 and the depth dimension stacking mode is enabled. A three-dimensional structure containing 512 blank memory blocks is allocated. The edge minimum value filling instruction is called to fill the blank nodes of the three-dimensional structure horizontally. The edge filling minimum value constant is specified as 0.001 and the filling boundary truncation threshold is set to the maximum index number plus 1. The memory area of ​​the expanded tensor boundary is overwritten bit by bit to generate the spatiotemporal load tensor.

[0022] The specific steps for obtaining biomembrane mass transfer are as follows: Based on the spatiotemporal load tensor, the singular value decomposition algorithm is used to decompose the internal node elements of the multidimensional tensor. The tensor modulus expansion command is invoked to flatten the third-order mathematical calculation array along the modal direction, reconstructing a two-dimensional operation matrix array with a set matrix row number parameter of 1024 and a column number parameter of 512. Factorization operation commands are then invoked to process the internal elements of the two-dimensional operation matrix array, deriving and establishing an orthogonal basis matrix and a singular value diagonal matrix. Eigenvalue filtering and truncation commands are invoked to remove redundant values ​​off-diagonal, setting the truncation tolerance scalar to 0.001, discarding scalars below the threshold, and extracting the main diagonal. The core numerical features of the line are associated with the preset spatial environment distribution constant. The preset spatial environment distribution constant is obtained by reading the physical structure parameters of the microporous aeration component from the underlying hardware initialization configuration file and assigning the specific value of 300 square meters per cubic meter. The Kronecker product multiplication operation instruction is called to combine the core numerical features of the main diagonal with the constant scalar of 300 square meters per cubic meter to perform a multiplication operation. 200 megabytes of memory space are allocated to store the product parameters. The one-dimensional continuous array is reconstructed by arranging them in row-major order, and the reconstructed multi-dimensional feature array vector is extracted to establish discrete load sampling points. Based on discrete load sampling points, a memory step-by-step read instruction is invoked to extract multidimensional values ​​within the nodes. The address offset row spacing constant is set to 4 bytes. An extended Kalman filter algorithm combined with a Monod biochemical kinetic differential derivation model is used to substitute variable features into the degradation equation. The maximum specific degradation rate parameter is set to 2.5 mg / day, and the half-saturation constant parameter is set to 0.8 mg / L. A Jacobi matrix differentiation instruction is invoked to perform linearized partial derivative calculations on the multidimensional values, transmitting multidimensional spatial distribution parameters and integrating node internal characteristic indicators to calculate the corresponding environmental boundary. The calculated corresponding environmental boundary... By introducing Neumann boundary constraints, the substrate concentration gradient scalar on the reactor wall is set to 0, and a forced boundary assignment is performed. The state transition operation command is called to multiply the partial derivative matrix and the environmental boundary parameter array. The scalar of the diagonal element of the process noise covariance matrix is ​​set to 0.05. The numerical addition accumulation command is used to merge the noise matrix and the transition product parameter, and the array prior evolution covariance value is calculated. The arithmetic summation operation combined with the division operation command is called to perform a summation calculation on the 1024 prior evolution covariance values ​​and divide by the total number of element parameters to derive the array mean and obtain the matrix degradation prediction mean. Based on the predicted mean of matrix degradation, a matrix inversion and state gain compensation algorithm is used to poll the dissolved oxygen value of the microporous aeration disc via an industrial field communication protocol. The communication baud rate is set to 19200 bits per second, and the read / write polling period is specified as 100 milliseconds. The state deviation is determined by comparing it with a safety threshold. The safety threshold is input into the non-volatile memory area of ​​the programmable logic controller via the human-machine interface control panel, and a specific value of 2.0 mg / L is assigned. The dissolved oxygen value is subtracted from the constant scalar of 2.0 mg / L by a subtraction operation instruction to extract the corresponding differential distribution feature scalar. When the scalar exceeds the limit, the gain coefficient is captured. Specifically, the limit refers to the absolute difference exceeding the upper limit of the 0.5 mg / L tolerance range. The Gaussian-Jordanian elimination inversion instruction is invoked to perform a flip-inversion calculation on the sum array of the measurement noise covariance matrix and the prior error covariance matrix. Combined with floating-point multiplication, the output of the flip-inversion is multiplied with the prior error covariance matrix to extract the Kalman gain coefficient. The gain coefficient and the predicted mean are then combined. A single-precision floating-point multiplication operation is then invoked to multiply the Kalman gain coefficient with the differential distribution characteristic scalar to obtain the dynamic compensation adjustment product. Numerical addition instructions are used to accumulate the dynamic compensation adjustment product and the matrix degradation prediction mean. The internal parameters of the hardware memory base address block are overwritten to obtain the biofilm transfer mass.

[0023] The specific steps for generating gas-water synergistic commands are as follows: Based on biofilm mass transfer, a Markov decision state space construction method is adopted. The programmable logic controller (PLC) communication protocol is used to read the submersible pump's operating speed and extract the air supply valve's opening. The communication cycle is set to 50 milliseconds, and the speed data type is specified as a 16-bit unsigned integer. The valve opening percentage is read from the corresponding register. A multidimensional array horizontal concatenation instruction is used to construct a vector by concatenating related parameters. The vector dimension is set to 1 row and 3 columns, and the mass transfer values, speed values, and valve opening values ​​are arranged sequentially. The tensor addition module is called to substitute the matrix node accumulation elements, setting the initial bias constant to 0.5, and numerical addition instructions are used to perform the operation step by step. The system accumulates vector elements and bias constants, sends a probe command to the control terminal, calls the pseudo-random number generator module to generate Gaussian white noise disturbance values ​​with a mean parameter of 0 and a variance parameter of 0.1, uses a single-precision addition command to add the Gaussian white noise disturbance values ​​and the accumulated vector elements to perform the action exploration operation, outputs 4 to 20 mA analog current signals to the relay pins of the underlying actuator through the digital-to-analog conversion interface, calls the energy consumption power acquisition command to extract the underlying hardware operation indicators, sets the sampling time window to 10 seconds and reads the active power kilowatt-hour consumption value inside the energy meter, uses normalized division operation to divide by the rated power base value to scale the variable to the 0 to 1 range, and establishes an operation reward evaluation scalar; Based on the operational reward evaluation scalar, a deep deterministic gradient reinforcement learning algorithm is employed. The internal flash file system is invoked to retrieve the equipment operating baseline. This baseline is obtained by collecting the average power consumption of internal electromechanical operations over 30 historical days, writing it to a read-only sector of non-volatile memory, and performing hard-coded preset assignment, specifying a value of 45 kilowatts. A single-precision floating-point subtraction operator is used to subtract the baseline from the evaluation scalar, calculating the corresponding numerical difference features. An absolute value comparison instruction is then used to compare the difference features with the safety boundary and determine the direction of feature boundary violations. The safety boundary is determined by manually inputting threshold parameters from the maintenance panel and sending them to the control core verification area, setting an upper limit parameter of +5 kilowatts and a lower limit parameter. The value is obtained as negative 5 kW. The logical judgment branch statement is called to identify whether the differential feature is in the positive interval of the upper limit parameter or the negative interval of the lower limit parameter. The adaptive moment estimation optimizer is called to start the backpropagation differentiation process to overwrite the internal node connection weights. The global learning rate hyperparameter is set to 0.001 and the first moment estimation exponential decay rate is specified to be 0.9. The loss function is obtained by using the chain rule of calculus for the partial derivative matrix of the hidden layer neurons of the multilayer perceptron. The floating-point multiply-add operation instruction is called to multiply and combine the partial derivative matrix with the learning rate parameter to update the weight parameter of the multinomial matrix array of the evaluation network. The latest forward propagation tensor output term of the action network is extracted and associated with the frequency control hardware interface of the frequency converter to generate the gas-water coordinated instruction.

[0024] The specific steps for establishing frontier control parameters are as follows: Based on biofilm mass transfer and gas-water co-operation commands, a multi-dimensional variable feature fusion and coordinate space mapping method is used to fuse features and command variables. The underlying application programming interface communication function is called to capture the floating-point mass transfer value and the 16-bit unsigned integer co-operation command code within the memory address block. The memory read offset parameter is set to 8 bytes. A one-dimensional array concatenation command is called to generate an initial 1-row, 4-column fusion vector according to the continuous memory arrangement rules. A protocol parsing and unpacking command is called to extract the operating power consumption value and capture the conveyor head parameter. The absolute physical address of the industrial bus addressing message is specified to locate the register and extract the kilowatt-hour power consumption single-precision floating-point value. A specific function code command is then sent. The system captures the meter-dimensional head single-precision floating-point parameter, associates the power consumption value with the head parameter to construct a planar coordinate system, calls the memory space allocation function to delineate a blank addressing area containing 500 floating-point node structures, calls the coordinate axis variable mapping write instruction to force the power consumption floating-point value to be assigned to the horizontal dimension attribute block of the structure, forces the head floating-point parameter to be assigned to the vertical dimension attribute block of the structure, calls the pointer auto-increment command to gather the underlying discrete node elements, sets the pointer auto-increment step constant to 1 and enables the direct memory access channel, captures the underlying continuously updated hardware status parameters at a 50-millisecond polling cycle to cover the blank index area of ​​the above structure bit by bit, and establishes a two-dimensional distributed coordinate matrix. Based on a two-dimensional distributed coordinate matrix, a non-dominated sorting genetic algorithm crowding calculation model is adopted. The system calls a memory-based step-by-step search pointer to traverse and address the coordinates of the underlying nodes. The starting address of the search pointer is bound to index 0 of the matrix array, and the search depth parameter is specified as the extreme values ​​of 500 nodes. Data structure decomposition instructions are used to extract the horizontal and vertical axis features. The horizontal axis decomposition mask constant is set to hexadecimal FFFF0000, and the vertical axis decomposition mask constant is set to hexadecimal 0000FFFF. Bitwise AND logical operations are used to separate the power consumption horizontal axis value and the head vertical axis parameter, storing them in two independent one-dimensional cache arrays. The floating-point subtraction unit subtracts the values ​​of adjacent nodes in the same dimension and extracts the absolute difference. It performs the subtraction operation on adjacent elements within the two sets of one-dimensional cache arrays. It calls the absolute value conversion function in the system's math library to forcibly convert all negative differences into positive floating-point scalars. It calls the floating-point addition accumulator register to accumulate the absolute differences of each item and output the congestion characterization. It sets the initial accumulator base constant to 0.0. It uses a loop summation instruction to superimpose the absolute difference parameters of the horizontal axis and the vertical axis. It calls the division operation instruction to divide by the corresponding dimension range constant to perform a normalization operation and output a single-precision congestion floating-point value, establishing a spatially distributed discrete scalar. Based on spatially distributed discrete scalars, a Pareto front optimization and nonlinear threshold truncation screening method is used to retrieve the coordinates of scalar-bound nodes. A hash mapping lookup table instruction is called, inputting a congestion degree floating-point value as the primary key parameter. This is used to match the coordinates of the corresponding original two-dimensional power-consuming head structure within the underlying physical storage block. The discrete scalars are compared with a lower truncation threshold, and overlapping nodes below the threshold are removed. This lower truncation threshold is obtained by extracting the median of the distribution density of invalid convergent solutions from 50 historical algorithm optimization iterations, writing it into the system's global read-only variable area, and specifying a single-precision value of 0.55. A single-precision floating-point comparison instruction is then called to compare each discrete scalar with 0.5. 5. Constant size relationship: When the hardware condition jump branch is triggered, if the scalar is less than 0.55 constant, the array element erasure instruction is called to release the corresponding memory block and clean up overlapping nodes. The memory block copy instruction is called to extract the scattered points on the outer edge of the remaining level. For valid node addresses greater than or equal to 0.55 constant, the block transfer command is triggered. The single transfer block size parameter is set to 16 bytes. The discrete scattered points distributed on the first Pareto front in the multidimensional space are extracted. The dynamic array recombination and splicing command is called to summarize and retain valid coordinate elements. The upper limit parameter of the target recombination array dimension is set to 100 items, and the bubble sort operation is performed according to the ascending order rule of the vertical axis to reshape the memory array order and establish the front control parameter.

[0025] The specific steps to obtain the driving pulse waveform are as follows: Based on frontier control parameters, a multi-dimensional addressing mapping and data decomposition and separation method is adopted. The memory block extreme value addressing command is called to filter the outer boundary two-dimensional coordinate matrix. The addressing boundary offset constant is set to 2 bytes. A loop comparison instruction is called to traverse the structure array and match the extreme value scalars of the horizontal and vertical axes. A hash lookup table read command is called to extract the mapping elements of independent node positions. The mapping primary key type parameter is set to a 32-bit unsigned integer, and the hash collision detection step size constant is specified as 1. The system's internal static random access memory cache is accessed. A register indirect addressing instruction is called to retrieve the node-bound hardware operation parameters, pointing to the starting address of the underlying device parameter library storage and setting the read span parameter to 16 bytes. Bitwise AND mask operations are then called. The target speed variable of the submersible pump is split, the high-order mask constant is set to hexadecimal FFFF0000, the right shift logic operation instruction is used to shift 16 bits to the right to extract the 16-bit integer speed value inside the high-order storage register, and the speed forced truncation upper limit parameter is set to 1500 revolutions per minute. The low-order mask interception instruction is called to separate the micro-orifice air valve air pressure control value, the low-order mask constant is set to hexadecimal 0000FFFF, the bitwise AND logic operation is used to extract the single-precision air pressure floating-point parameter inside the low-order storage register, and the air pressure truncation protection upper limit scalar is specified as 2.0 MPa. The memory splicing and packing command is called to combine the above separated speed and air pressure parameters and store them into an independent physical address block to establish the electromechanical drive parameter value. Based on electromechanical drive parameters, a digital-to-analog converter (DAC) and pulse width modulation (PWM) coding algorithm are used. The DAC channel is accessed to read the rotational speed and micro-orifice pressure parameters. The DAC transfer mode constant is set to cyclic burst mode, and the single transfer word length is specified as 32 bits. The DAC peripheral interface library function is used to encode the rotational speed characteristic as a voltage analog wave. The DAC hardware resolution parameter is set to 12 bits, the lower limit of the output analog voltage range is specified as 0 volts, and the upper limit as 10 volts. A linear interpolation instruction is used to linearly map the 1500 RPM upper limit integer rotational speed value to a discrete digital scale range of 0 to 4095, and a hardware bus write operation is used to store the data in the DAC data holding register. Finally, a general-purpose input / output pin level toggling command is invoked. The pulse is sent to the inverter's electrical contacts, the pin's electrical output speed parameter is set to 50 MHz, and the hardware pin is configured as a push-pull output mode. The advanced timer module is called to convert the air pressure value to a duty cycle level. The timer hardware prescaler parameter is set to 71, and the constant value of 999 is written into the auto-reload register to establish a 10 kHz basic hardware carrier frequency. The CPU multiplier instruction is used to multiply the above 2.0 MPa air pressure value mapping ratio constant and the constant value of 999. The timer capture comparator register comparison parameter is calculated and a forced assignment input operation is performed. The pulse width modulation output enable instruction is called to send a level instruction to the hardware driver, triggering the hardware complementary channel dead time generator and setting the dead time delay safety constant to 100 nanoseconds to obtain the drive pulse waveform.

[0026] The above are merely preferred embodiments of the present invention and are not intended to limit the present invention in any other way. Any person skilled in the art may make changes or modifications to the above-disclosed technical content to create equivalent embodiments that can be applied to other fields. However, any simple modifications, equivalent changes, and modifications made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention shall still fall within the protection scope of the present invention.

Claims

1. A method for intelligent optimization and control of aeration rate and internal recirculation rate based on MABR process, characterized in that, Includes the following steps: Step 1: Based on the sensor readings of the biochemical system monitoring nodes, read the oxygen demand and nitrate concentration of the return weir, extract the flow velocity variable of the hollow membrane tube, arrange the concentration values ​​to construct a column vector array, splice the column vector array and the flow velocity variable to merge the operation elements, and establish the spatiotemporal load tensor; Step 2: Based on the spatiotemporal load tensor, decompose the multidimensional tensor elements and extract the diagonal values ​​to precipitate discrete sampling points. After deriving the predicted mean by substituting it into the equation through Kalman filtering, compare it with the scalar of dissolved oxygen difference, determine the scalar exceeding the limit, and perform simultaneous calculation of gain coefficient and mean to obtain biofilm transfer mass. Step 3: Based on the biofilm mass transfer, splice the submersible pump speed and valve opening, sum and output the operation command and extract the return evaluation value, subtract the evaluation value from the baseline to calculate the difference scalar, determine the scalar out of bounds and update the network connection weight by multiplication, and generate the air-water synergy command. Step 4: Based on the biofilm mass transfer and the gas-water synergy command, the numerical values ​​and commands are fused using a genetic algorithm to extract power consumption and head scalars to construct coordinates, the dimensional difference is accumulated to output the distribution scalar, the lower limit scalar coordinates are removed and the outer layer points are retained to establish the frontier control parameters; Step 5: Based on the aforementioned front-end control parameters, select independent points from the outer boundary coordinates, separate the output speed and micro-hole air pressure values, encode the speed as a voltage pulse and send it to the inverter contacts, convert the air pressure into a duty cycle level and send it to the drive air valve terminal to obtain the drive pulse waveform.

2. The intelligent optimization control method for aeration volume and internal recirculation flow based on MABR process according to claim 1, characterized in that, The spatiotemporal load tensor includes a column vector array of oxygen demand and nitrate concentration, a discrete variable of flow velocity inside the hollow membrane tube, and a time step dimension constant parameter. The biofilm mass transfer includes the predicted mean of matrix degradation, a scalar of the differential distribution characteristics of dissolved oxygen, and a gain coefficient. The gas-water synergistic command includes a variable of submersible pump operating speed, a parameter of the air supply valve adjustment opening, and a network connection weight. The front-edge control parameters include operating power consumption, a parameter of delivery head, a discrete scalar of spatial distribution, and the coordinates of scattered points on the outer edge. The drive pulse waveform includes a simulated waveform of inverter contact voltage and a duty cycle level of the drive air valve terminal.

3. The intelligent optimization control method for aeration volume and internal recirculation flow based on MABR process according to claim 1, characterized in that, The specific steps for establishing the spatiotemporal load tensor are as follows: Based on the sensor readings of the biochemical system monitoring nodes, the oxygen demand and nitrate concentrations are read in a polling manner, the discrete variable of the flow velocity inside the hollow membrane tube is extracted, the concentration values ​​are arranged into a vertical array, and the vertical array and the flow velocity variable are spliced ​​horizontally to establish a multidimensional biochemical parameter column vector. Based on the multidimensional biochemical parameter column vector, the numerical discrete elements inside the address vector are traversed, a preset time step dimension constant parameter is introduced, the numerical elements and the time dimension parameter are spliced ​​together, and the tensor boundary is expanded by horizontally filling blank nodes to establish the spatiotemporal load tensor.

4. The intelligent optimization control method for aeration rate and internal recirculation rate based on MABR process according to claim 1, characterized in that, The specific steps for obtaining the mass transfer of the biomembrane are as follows: Based on the spatiotemporal load tensor, the internal node elements of the multidimensional tensor are decomposed, redundant values ​​off-diagonal are removed, core numerical features of the main diagonal are extracted, a preset spatial environment distribution constant is associated, a reconstructed multidimensional feature array vector is extracted, and discrete load sampling points are established. Based on the discrete load sampling points, multidimensional values ​​inside the nodes are extracted, Kalman filtering is used to substitute the variable features into the degradation equation, multidimensional spatial distribution parameters are transmitted and node internal feature indicators are integrated, the corresponding environmental boundary is calculated, the array mean is derived, and the matrix degradation prediction mean is obtained. Based on the predicted mean of matrix degradation, the dissolved oxygen value of the microporous aeration disc is read and compared with the safety threshold to determine the state deviation. The corresponding differential distribution feature scalar is extracted. When the scalar exceeds the limit, the gain coefficient is captured. The gain coefficient and the predicted mean are combined to obtain the biofilm transfer mass.

5. The intelligent optimization control method for aeration volume and internal recirculation flow based on MABR process according to claim 1, characterized in that, The Kalman filter extracts multidimensional numerical parameters within discrete load sampling points to construct an initial state covariance array. These multidimensional numerical parameters are then substituted into the matrix degradation equation. The state transition analytical matrix vector is calculated, and multidimensional spatial distribution parameters are fused to perform prior prediction deduction. The initial discrete variable features within the equation are replaced, and the matrix consumption and biological growth kinetics characteristics within the nodes are integrated. The state propagation evolution covariance matrix is ​​calculated, and the specific environmental constraints of the associated biochemical system are converted. The state vectors are accumulated, and the mathematical expectation quotient is calculated to derive the prior prediction mean of the generated array.

6. The intelligent optimization control method for aeration rate and internal recirculation rate based on MABR process according to claim 4, characterized in that, The process involves disassembling the internal node elements of the multidimensional tensor, extracting the spatial, temporal, and concentration dimension values ​​of the third-order spatiotemporal load tensor, performing a dimension reduction matrix expansion operation according to the modal expansion rule, and establishing multiple two-dimensional modal expansion matrices. A singular value decomposition matrix operation formula is introduced to calculate the multiple two-dimensional modal expansion matrices separately, generating multiple left singular vector matrices and separating the corresponding singular value scalars. The left singular vector matrices are then correlated with the original third-order spatiotemporal load tensor to perform modal multiplication, deriving and constructing an orthogonal factor array. The singular value scalars are compared with a preset truncation threshold, and the dimensional arrays corresponding to scalars below the threshold are truncated to generate a condensed principal component core tensor, thus completing the disassembly of the internal node elements of the multidimensional tensor.

7. The intelligent optimization control method for aeration rate and internal recirculation rate based on MABR process according to claim 1, characterized in that, The specific steps for generating the gas-water synergistic command are as follows: Based on the biofilm mass transfer, the submersible pump operating speed is read and the air supply valve adjustment opening is extracted. The associated parameters are spliced ​​to construct a vector, the matrix node accumulation elements are substituted, the trial command is sent to the control end, the underlying hardware operation indicators are extracted, and an operation return evaluation scalar is established. Based on the operational return evaluation scalar, the equipment operating baseline is retrieved, the baseline and evaluation scalar are subtracted, the corresponding numerical difference feature is calculated, the difference feature is compared with the safety boundary and the feature overstepping direction is determined, the internal node connection weights are overwritten, and the gas-water coordination command is generated.

8. The intelligent optimization control method for aeration volume and internal recirculation flow based on MABR process according to claim 1, characterized in that, The specific steps for establishing the aforementioned frontier control parameters are as follows: Based on the biofilm mass transfer and the gas-water synergy command, the features and command variables are fused, the operating power consumption value is extracted and the conveying head parameter is captured, the power consumption value and the head parameter are associated to construct a planar coordinate, the bottom discrete node elements are aggregated, and a two-dimensional distributed coordinate matrix is ​​established. Based on the dual-dimensional distributed coordinate matrix, a genetic algorithm is used to traverse and address to read the position coordinates of the bottom node, the features of the horizontal and vertical axes are stripped, the values ​​of adjacent nodes in the same dimension are subtracted and the absolute difference is extracted, and the absolute difference of each item is accumulated to output the crowding representation, thus establishing a spatially distributed discrete scalar. Based on the spatially distributed discrete scalar, the coordinates of the scalar-bound nodes are retrieved, the discrete scalar is compared with the lower limit truncation threshold and overlapping nodes below the lower limit threshold are removed, the remaining peripheral edge scattered points of the remaining level are truncated, the effective coordinate elements are summarized and retained, and the frontier control parameters are established.

9. The intelligent optimization control method for aeration volume and internal recirculation flow based on MABR process according to claim 1, characterized in that, The genetic algorithm sets a two-dimensional distributed coordinate matrix as the address space, constructs a traversal pointer array that maps the two-dimensional coordinate plane, triggers step-by-step traversal actions according to the index order of the matrix rows and columns, locates the underlying mapping storage location corresponding to the discrete node, reads the planar position coordinate value of the individual pointed to by the current pointer, converts the read and extracted node coordinates into the spatial gene sequence encoding of the initial population individual, and outputs an address sequence array containing multiple node position features.

10. The intelligent optimization control method for aeration rate and internal recirculation rate based on MABR process according to claim 1, characterized in that, The specific steps to obtain the driving pulse waveform are as follows: Based on the aforementioned frontier control parameters, the two-dimensional coordinate point matrix of the outer boundary is screened, the independent node position mapping elements are extracted, the node-bound hardware operation parameters are retrieved, the target speed variable of the submersible pump is split, the air pressure control value of the microporous air valve is separated, and the electromechanical drive parameter values ​​are established. Based on the electromechanical drive parameters, the rotational speed and micro-hole air pressure parameters are read, the rotational speed characteristic is encoded as a voltage analog wave, a pulse is sent to the inverter electrical contacts, the air pressure value is converted into a duty cycle level, and a level command is sent to the hardware driver to obtain the drive pulse waveform.

Images