A desulfurization intelligent control system based on machine learning algorithm

Abstract

The present application relates to the field of desulfurization control, and discloses a desulfurization intelligent control system based on a machine learning algorithm, comprising a collection module, an analysis module, a prediction module and a decision module, wherein a multi-physical field parameter in a desulfurization tower is collected through multi-sensor fusion to generate a multi-physical field state tensor; the dynamics behavior of an intelligent feature extraction analysis system is utilized to generate a characteristic parameter tensor; a state space model is constructed based on a dynamic system theory to generate a multi-target intelligent prediction result. The present application establishes a closed-loop control system from holographic perception to intelligent decision-making, solves the problems of perception limitations and single targets existing in traditional desulfurization systems, and improves desulfurization efficiency and reduces energy consumption.

Description

Technical Field

[0001] This invention relates to the field of desulfurization control, and in particular to a desulfurization intelligent control system based on machine learning algorithms. Background Technology

[0002] With increasingly stringent national environmental protection policies and the deepening implementation of the dual-carbon strategy, flue gas desulfurization systems in coal-fired power plants, steel, and chemical industries are facing unprecedented pressure to operate efficiently, stably, and with low carbon emissions. Existing desulfurization control systems generally suffer from the following shortcomings: Existing systems mostly rely on discrete sensors at a small number of key points for measurement, which cannot comprehensively and accurately reflect the continuous and three-dimensional distribution of multiple physical fields such as SO2 concentration field, slurry pH field, and temperature field inside the tower. Traditional control is mostly based on simplified mechanism models or PID control, which makes it difficult to accurately describe and predict the complex dynamic behavior of the desulfurization tower, a large inertial, highly nonlinear, and multivariable coupled system. Multiple actuators, such as absorbent dosing, slurry spraying, and fan air volume, are usually controlled independently, lacking a coordinated linkage mechanism. When faced with abnormal operating conditions, the system lacks global self-organizing reconstruction and fault-tolerant control capabilities, which can easily lead to control mismatch, regulation oscillation, or even unplanned shutdowns.

[0003] Therefore, we propose a desulfurization intelligent control system based on machine learning algorithms to solve the above problems. Summary of the Invention

[0004] This invention provides a desulfurization intelligent control system based on machine learning algorithms, which is used to construct a closed-loop cognitive control system from multi-physics field perception to intelligent decision execution.

[0005] This invention provides a desulfurization intelligent control system based on machine learning algorithms, comprising: a data acquisition module, which acquires parameters of the flue gas treatment process in the desulfurization tower through a multi-sensor fusion system, inputs multi-source heterogeneous data into a data fusion processor, and generates a multi-physics state tensor; an analysis module, which analyzes the dynamic behavior of the system based on the multi-physics state tensor and generates a feature parameter tensor; a prediction module, which constructs geometric features and analyzes evolution laws based on the feature parameter tensor and generates a multi-objective intelligent prediction result; and a decision-making module, which generates an intelligent control parameter set based on the multi-objective intelligent prediction result and the multi-physics state tensor.

[0006] Optionally, in the first implementation of the first aspect of the present invention, a three-dimensional spatiotemporal distribution tensor of SO2 concentration is generated by analyzing the SO2 concentration distribution using a laser spectral analyzer array and processing technology; the hydrogen ion activity in the slurry is monitored by implanting a circulation pipeline and a high-density microelectrode array, and a three-dimensional dynamic response field of slurry pH value is generated by combining the analyzer data; a cooperative distribution cloud map is generated based on a high-resolution infrared thermal imaging system combined with pressure sensor data; an SO2 diffusion feature descriptor is generated based on the SO2 concentration three-dimensional spatiotemporal distribution tensor; the three-dimensional dynamic response field of slurry pH value and the cooperative distribution cloud map are input into a multi-physics field coupling analyzer to generate a multi-field coupling feature matrix; and a multi-physics field state tensor is constructed based on the SO2 diffusion feature descriptor and the multi-field coupling feature matrix.

[0007] Optionally, in a second implementation of the first aspect of the present invention, a multi-scale feature representation is constructed based on the multi-physics state tensor to generate a system multi-scale feature description; the evolution law of the calculated parameters is based on the system multi-scale feature description to generate a multi-scale feature evolution sequence; the behavior characteristics of the feature parameters are analyzed using the multi-scale feature evolution sequence to generate a stable feature set; and a feature parameter tensor is generated based on the multi-scale feature evolution sequence and the stable feature set.

[0008] Optionally, in a third implementation of the first aspect of the present invention, the method includes: constructing a state-space model based on the feature parameter tensor to generate a system state-space trajectory; analyzing the dynamic evolution characteristics of the system through the system state-space trajectory to generate a system dynamic behavior description subset; generating multi-level stability warning indicators and performance interval identifiers based on the system dynamic behavior description subsets; and generating multi-objective intelligent prediction results based on the multi-level stability warning indicators and performance interval identifiers.

[0009] Optionally, in a fourth implementation of the first aspect of the present invention, the method includes: constructing a multi-objective function and constraints for the optimization problem based on the multi-objective intelligent prediction results and the multi-physics state tensor, and generating a mathematical description of the optimization problem; inputting the mathematical description of the optimization problem into a constraint optimization solver to generate a set of candidate optimization solutions; generating a set of basic control parameters using the set of candidate optimization solutions; and obtaining a set of intelligent control parameters based on the set of basic control parameters.

[0010] Optionally, in the fifth implementation of the first aspect of the present invention, a multi-objective function and constraints for the optimization problem are constructed based on the multi-objective intelligent prediction results and the multi-physics state tensor. The multi-objective function is expressed as a comprehensive optimization objective function: in As decision variables, For desulfurization efficiency, To achieve the target desulfurization efficiency, This represents the total energy consumption of the system. This is the energy consumption baseline value. As a system stability indicator, , , The weighting coefficients and + + =1.

[0011] Optionally, in a sixth implementation of the first aspect of the present invention, an intelligent control parameter set is obtained by establishing a correspondence between the optimization target and the set value of the underlying actuator based on the basic control parameter set.

[0012] Optionally, in a seventh implementation of the first aspect of the present invention, an output module is included to output the intelligent control parameter set to a multi-actuator collaborative control system to achieve intelligent optimization control of desulfurization.

[0013] Optionally, in the eighth implementation of the first aspect of the present invention, the limestone slurry metering device is adjusted according to the absorbent dosing strategy in the intelligent control parameter set to generate a precise dosing control command; the slurry circulation optimization scheme in the intelligent control parameter set is adjusted through adaptive PID control to adjust the output of the circulating pump frequency converter to generate a flow optimization control command; the oxidation fan speed and guide vane angle are adjusted according to the oxidation air volume in the intelligent control parameter set to generate a precise air volume control signal; the precise dosing control command, the flow optimization control command, and the precise air volume control signal are input into a multi-loop collaborative controller to generate multi-actuator collaborative operation parameters; and the multi-actuator collaborative input is input into a real-time dynamic optimizer to generate a mechanism dynamic adjustment command.

[0014] Optionally, in the ninth implementation of the first aspect of the present invention, the system further includes self-organizing reconfiguration through system dynamic analysis, abnormal state identification, adaptive control, and multi-objective optimization: The system dynamic evolution is estimated and predicted based on the characteristic parameter tensor to generate a description of the system's dynamic behavior; the abnormal operating conditions and potential risks in system operation are identified by analyzing the deviations and trends of key parameters, generating abnormal operating condition identification results; a recovery strategy for the system under abnormal operating conditions is constructed based on the abnormal operating condition identification results, generating a set of adaptive control strategies; the set of adaptive control strategies is optimized by balancing desulfurization efficiency, energy consumption, and stability, generating system reconfiguration instructions.

[0015] The mechanism of this invention is as follows: By generating a high-dimensional spatiotemporal tensor through multi-sensor fusion, the limitations of traditional point-based measurement are overcome. By utilizing multi-scale feature extraction and dynamic behavior analysis, the inherent evolution law of the system is revealed. Based on state space construction and risk prediction, forward-looking coordination of multi-objective conflicts is achieved. Beneficial effects: By fusing multiple sensors, a three-dimensional spatiotemporally continuous multiphysics state tensor is generated, breaking through the limitations of traditional point-based measurement; Based on multi-scale feature extraction and dynamic behavior analysis, we can achieve a leap from data to understanding system behavior. By utilizing state-space construction and multi-objective risk prediction, a pre-intervention prediction and decision-making mechanism is established. Attached Figure Description

[0016] Figure 1 This is a schematic diagram of an embodiment of a desulfurization intelligent control system based on machine learning algorithms according to an embodiment of the present invention; Figure 2 This is a schematic diagram of another embodiment of a desulfurization intelligent control method based on machine learning algorithms in this invention.

[0017] Figure 3 This is a schematic diagram illustrating the application of multi-sensor fusion and tensor theory in the data acquisition and fusion process. Detailed Implementation

[0018] This invention provides a desulfurization intelligent control system based on machine learning algorithms, used to construct a closed-loop cognitive control system from multi-physics field sensing to intelligent decision execution. The terms "first," "second," "third," "fourth," etc. (if present) in the specification, claims, and accompanying drawings are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments described herein can be implemented in a sequence other than that illustrated or described herein. Furthermore, the terms "comprising" or "having" and any variations thereof are intended to cover a non-exclusive inclusion; for example, a process, method, system, product, or device that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or devices.

[0019] For ease of understanding, the specific process of the embodiments of the present invention is described below. Please refer to [link / reference]. Figure 1 One embodiment of the intelligent desulfurization control system based on machine learning algorithms in this invention includes: 101. Acquisition Module: The distributed multi-sensor fusion system collects all operating parameters of the flue gas treatment process in the desulfurization tower in real time. This includes obtaining the spatial distribution matrix of SO2 concentration using a laser spectral analyzer, obtaining the pH value of the slurry using a microelectrode array sensor, and obtaining the temperature field distribution cloud map using an infrared thermal imager. The collected multi-source heterogeneous data is input into a data fusion processor based on tensor theory to generate a multi-physics state tensor containing spatiotemporal characteristics.

[0020] It is understood that the executing entity of this invention can be a desulfurization intelligent control device based on machine learning algorithms, or it can be a terminal or a server; no specific limitation is made here. This embodiment of the invention will be described using a server as an example.

[0021] It should be noted that, taking the desulfurization tower of a 600MW unit in a large coal-fired power plant as an example, a total of 16 monitoring sections are arranged at different heights of the absorption zone from top to bottom of the desulfurization tower, below the inlet flue, between the three spray layers, and above the outlet mist separator.

[0022] Sensor Deployment and Data Acquisition: Specific sensors are installed at each monitoring section. First, a sensor network is formed using eight TDLAS (Tunable Diode Laser Absorption Spectroscopy) analyzers. These instruments emit laser beams at specific frequencies, passing through the flue. By analyzing the attenuation spectrum of the laser beam after absorption by sulfur dioxide molecules in the flue gas, the SO2 concentration along the path is retrieved in real time. Using a tomographic reconstruction algorithm, the measurements at each section are processed into a 20×20 gridded concentration distribution matrix. The data in each cell of the matrix represents the SO2 concentration value of that tiny region, expressed in milligrams per cubic meter.

[0023] In the corresponding slurry spraying layer area, an array sensor consisting of 24 miniature pH electrodes was deployed. These electrodes were immersed in the flowing slurry in a grid pattern, simultaneously measuring the slurry pH value at different spatial locations at a frequency of once per second, with a measurement accuracy of 0.01 pH units. These pH values ​​directly reflect the acidity and alkalinity distribution of the slurry. Simultaneously, three high-resolution infrared thermal imagers were installed at key locations on the outer side of the tower wall, capturing temperature field distribution cloud maps of the tower surface at a rate of 10 frames per second. Through inversion calculations using a heat conduction model, the approximate temperature field distribution inside the tower was obtained, with a temperature measurement range of 40 to 80 degrees Celsius and an accuracy of ±0.5℃.

[0024] The three types of sensors (laser spectrometer, pH microelectrode array, and infrared thermal imager) synchronously acquire data in 100-millisecond cycles. The acquired raw data is heterogeneous: SO2 concentration is a spatial matrix, pH value is a discrete point array, and temperature is a two-dimensional image pixel value. All of these data are timestamped and labeled with spatial coordinates before being sent to a central data fusion processor. This processor is built based on tensor theory. It integrates all the data within an acquisition cycle into a four-dimensional tensor. The four dimensions of this tensor are: time (continuous acquisition moments), spatial location (defined by three-dimensional coordinates X, Y, Z), physical field type (SO2 concentration, pH value, temperature), and the numerical value of the physical quantity itself.

[0025] The processor discretizes the tower's internal space into a 50 (height) × 20 (radial direction 1) × 20 (radial direction 2) grid. For each grid cell, a spatial interpolation algorithm is used to convert the discrete data measured by the sensors into SO2 concentration, pH value, and temperature value at the center of that cell. Finally, a multiphysics state tensor with dimensions of [1 (time step) × 50000 (spatial grid points) × 3 (physical field type) × 1 (numerical value)] is generated.

[0026] 102. Analysis Module: The generated multiphysics state tensor is input into the intelligent feature extraction module based on multi-scale feature extraction theory. By constructing a multi-scale analysis framework and calculating the feature evolution flow, the dynamic behavior of the system is analyzed at multiple spatial scales, generating a feature parameter tensor containing the evolution sequence of key parameters and system stability indicators.

[0027] It should be noted that the multiphysics state tensor generated from the previous data is a four-dimensional tensor with dimensions [1, 50000, 3, 1], which contains the values ​​of three physical fields: SO2 concentration, slurry pH, and temperature at 50,000 spatial grid points inside the desulfurization tower at a certain moment.

[0028] This multiphysics state tensor can be viewed as an instantaneous snapshot of a physical system at different scales. The module defines a set of key parameters characterizing the system state: SO2-pH correlation strength (α1): characterizing the correlation strength between the SO2 concentration field and the slurry pH field; initial values ​​can be calculated based on data; pH-temperature correlation strength (α2): characterizing the correlation strength between the slurry pH field and the temperature field; initial values ​​can be calculated based on data; SO2-temperature correlation strength (α3): characterizing the correlation strength between the SO2 concentration field and the temperature field; initial values ​​can be calculated based on data.

[0029] The module performs a series of multi-scale analysis transformations, an iterative "coarsening" process designed to extract macroscopic patterns from microscopic details. The specific steps are as follows: Select the analysis scale. The module presets four different spatial scales for analysis: Scale S0 (original scale, corresponding to 50,000 grid points), Scale S1 (relatively coarse scale), Scale S2 (even coarser scale), and Scale S3 (the coarsest scale, corresponding to only a few hundred macroscopic regions).

[0030] First transformation (S0→S1): The module performs block averaging of the physical field data (SO2 concentration, pH, temperature) from eight adjacent (2×2×2) spatial grid points in the original tensor and recalculates the key parameters. During this process, the system parameters (α1, α2, α3) "evolve" according to their data distribution at the new scale. After calculation, at the new scale S1, the parameters evolve as follows: α1=A1, α2=A2, α3=A3 (A1, A2, A3 are specific calculated values). Iterative transformation (S1→S2, S2→S3): The above process is repeated. At scale S2, the parameters further evolve as follows: α1=B1, α2=B2, α3=B3. At scale S3, the parameters become: α1=C1, α2=C2, α3=C3. The module generates a feature parameter tensor and, through the above analysis, obtains the evolution behavior of the system's key parameters at different observation scales, extracting key features: the key parameter evolution sequence, which is the trajectory of parameter changes with scale. The evolution sequence of α1 is [α1_S0,A1,B1,C1]. This sequence reveals the scale dependence of interactions between different physical fields.

[0031] By analyzing the changing trends of key parameters under multi-scale transformations, a set of system stability indices can be calculated, including the variance of parameter changes, trend slope, etc., forming a stability index vector I=[I1,I2,...]. These indices quantify the consistency and changing trend of the system state.

[0032] This information is encapsulated into a new feature parameter tensor. This tensor contains two main parts: one is a matrix consisting of the evolution sequence of each key parameter; the other is a vector containing stability index values.

[0033] 103. Prediction Module: The generated feature parameter tensor is input into the intelligent prediction module based on dynamic system and stability analysis theory. By constructing the geometric features of the system state space and analyzing the dynamic evolution law of the system, a multi-objective intelligent prediction result is generated, which includes stability early warning indicators, efficiency optimization intervals and fault risk levels.

[0034] It should be noted that the feature parameter tensor transmitted from the feature extraction module (step 102) contains key parameters describing the dynamics of the desulfurization system, the evolution sequence of the three key parameters (α1, α2, α3) at four different spatial scales, and a set of stability indices (I1, I2, ...).

[0035] The desulfurization system is modeled as an abstract state space. Each point in this state space represents a complete state of the system at a certain moment. This state is defined by all the parameters in the characteristic parameter tensor. The coordinates of a point can be (the value of α1 at scale S3, the value of α2 at scale S3, the value of α3 at scale S3, the stability index I1, ...). In this way, the continuous operation of the system over time traces a trajectory in this state space.

[0036] The entire system is considered a "dynamic system," where: the state space is the phase space composed of macroscopic average parameters (key parameters at the coarsest scale S3). It describes the overall, global state of the system; dynamic behavior is described by the state equations, showing how the system state evolves over time. The core task of the module is to analyze the stability of this dynamic system in order to predict its behavior, by calculating stability indices and analyzing evolutionary patterns. The module calculates Lyapunov exponents or stability criteria at various points in the state space. In a certain region of the state space, a higher calculated stability index indicates that the system dynamics are stable and the trajectory tends to converge; while in another region, a lower or negative stability index indicates that the system is in a metastable or critical state, sensitive to external disturbances, and the trajectory is prone to bifurcation or instability.

[0037] Based on the above stability analysis, the abstract mathematical quantities are transformed into specific engineering prediction indicators: when the system trajectory enters a region where the stability indicator is below the threshold, the module will trigger an advanced warning, and the prediction result indicates: "Within the next 5 minutes, due to the decoupling of the pH field and the temperature field, there is a risk of scaling at the bottom of the tower, and the stability level drops to level C (high risk)."

[0038] The system identifies regions in the state space with high stability indices and stable parameters, corresponding to efficient and stable system operation. The prediction results suggest: "Under current conditions, maintaining the macroscopic key parameter α1 between [A1_low, A1_high] and α2 between [A2_low, A2_high] will allow the system to operate within an optimized desulfurization efficiency range exceeding 95%." By analyzing the speed and direction of the trajectory towards the stability critical point, the module assesses the risk of failure. The prediction result is: "Insufficient oxidation airflow is causing the system to slide along a highly unstable path towards the 'limestone blockage' failure point; the risk level is expected to rise from yellow to red within 15 minutes." These prediction results are integrated into a structured multi-objective intelligent prediction result. This is no longer raw data but a deep insight into the system's future behavior, stability boundaries, and risk probabilities, providing a direct basis for the final intelligent decision-making.

[0039] 104. Decision Module: The generated multi-objective intelligent prediction results and the multi-physics state tensor are input into the intelligent decision module based on constraint optimization theory. By solving the multi-objective constraint optimization problem, a set of intelligent control parameters is generated, including the absorbent precise dosing strategy, the slurry circulation optimization scheme, and the oxidation air volume adjustment command.

[0040] It should be noted that at the current moment, the intelligent prediction module (step 103) outputs multi-objective intelligent prediction results, indicating that: "The system is currently in a metastable state, with a desulfurization efficiency of 92%, which is lower than the lower limit of the optimization range of 95%. The main risk is the uneven distribution of the slurry pH field, and a tendency for scaling is predicted in the next 5 minutes. The stability warning level is yellow." At the same time, the current multiphysics state tensor provided by the data fusion module (step 101) shows that: the average pH value in the bottom area of ​​the tower is 5.1, but the local area is as low as 4.8; the inlet SO2 concentration instantaneously rises to 2800 mg / m³.

[0041] The entire desulfurization tower is considered as a multi-input, multi-output controlled system. The system's overall performance indicators (desulfurization efficiency, energy consumption, and stability) are defined as the core optimization objective for decision-making. The module's goal is to find a set of control parameters that, under specific constraints, maximize the system's overall performance indicators.

[0042] The decision variables are three adjustable operating quantities: limestone slurry dosage (Fslurry), currently 12.5 m³ / h; slurry circulation pump frequency (Fpump), currently 38 Hz; and oxidation blower air volume (Fair), currently 85,000 standard m³ / h. The constraints on the decision are directly derived from the predicted results and real-time status: efficiency constraint, the desulfurization efficiency must be increased from 92% to no less than 95%; stability constraint, the pH value must be prevented from decreasing further, and the pH value at the bottom of the tower must be ensured to be no less than 5.0 to eliminate scaling warnings. Physical limit constraints: each operating quantity has its upper and lower limits: Fslurry ∈ [10,20], Fpump ∈ [30,45], Fair ∈ [70,000,100,000].

[0043] The above objectives and constraints are formalized into a mathematical problem: Given all constraints, find a set of (Fslurry, Fpump, Fair) that optimizes the overall performance index J. The module solves this problem through the following steps: establishing a mapping relationship. The module internally stores a mapping relationship constructed from historical data and a mechanistic model, which describes how changes in control parameters affect performance indicators such as desulfurization efficiency, energy consumption, and stability.

[0044] The search was conducted within the feasible region defined by the constraints. Multiple combinations were attempted. Option A: Fslurry was increased to 14.0, Fpump to 40.0, and Fair remained unchanged. Calculations showed this option could improve efficiency to 95.5%, but it led to supersaturation of the slurry at the bottom of the column. While the overall performance index J improved, it was not significant, and the risk of scaling was not completely eliminated. Option B: Fslurry was slightly adjusted to 13.2, Fpump was significantly increased to 42.5, and Fair was increased to 89000. Calculations showed that stronger circulation improved the uniformity of slurry distribution, and moderate oxidation airflow improved oxidation efficiency. This option not only improved efficiency to 96.2%, but also improved the flow field, resulting in a more uniform pH distribution (the lowest pH at the bottom increased to 5.2). The overall performance index J reached its optimum, and the system tended towards a more stable state.

[0045] After evaluation, Scheme B was determined to be the optimal solution because it optimized the overall performance index J while satisfying all constraints. Based on this, the module generated the final set of intelligent control parameters: a precise absorbent dosing strategy, setting the limestone slurry dosage to 13.2 cubic meters per hour; an optimized slurry circulation scheme, setting the circulation pump frequency to 42.5 Hz; and an oxidation air volume adjustment command, setting the oxidation air volume to 89,000 standard cubic meters per hour. This set of control parameters aims to drive the system efficiently and stably from a metastable state to the target performance range with optimal performance indicators, demonstrating the advantages of intelligent decision-making based on optimization theory.

[0046] 105. Output module: Outputs the generated set of intelligent control parameters to the multi-actuator collaborative control system, including a model-predictive limestone slurry precision metering device, an adaptive slurry circulation frequency conversion speed regulation system, and an intelligent pH dynamic compensation device, to achieve intelligent optimization control of the desulfurization chemical reaction process, gas-liquid mass transfer process, and thermodynamic process.

[0047] It should be noted that the intelligent decision-making module (step 104) generates the following precise set of control parameters: limestone slurry dosage: 13.2 cubic meters / hour; slurry circulation pump frequency: 42.5 Hz; oxidation blower air volume: 89,000 standard cubic meters / hour. The server sends these three setpoint instructions to the corresponding intelligent actuators via the industrial communication network.

[0048] The execution process of the limestone slurry precision metering device based on model prediction shows that the slurry dosing control system received a setpoint of 13.2 cubic meters per hour. This is not just a simple flow control command. The device incorporates a dynamic model of the slurry delivery pipeline, capable of predicting the time required for the flow rate to stabilize after adjusting the valve opening and the potential overshoot. The device does not directly linearly adjust the control valve opening to the corresponding position. It calculates an optimal adjustment trajectory: first, it quickly opens the control valve to a predicted opening (corresponding to 13.0 cubic meters per hour); then, as the flow sensor feedback value approaches the target, it slowly approaches the setpoint of 13.2 cubic meters per hour in small steps (adjusting the opening by 0.5% each time). The entire adjustment process is completed smoothly within 45 seconds, avoiding flow surges and pipeline vibrations, ensuring the accuracy and stability of the chemical absorbent dosing, thereby optimizing the desulfurization chemical reaction process.

[0049] The circulating pump frequency converter received a 42.5 Hz frequency command. This system not only changes the pump speed but also focuses on its overall impact on the fluid dynamics within the column. While increasing the pump speed, the frequency converter works in real-time with the pressure and pH distribution sensors within the column. As the frequency increases from the current 38.0 Hz, the system detects a gradual increase in the pressure differential within the column, indicating a more uniform droplet distribution. The system fine-tunes the acceleration rate to ensure a smooth increase in pressure differential, avoiding excessive acceleration that could overload the pump current. Ultimately, the pump speed stabilizes at 42.5 Hz, corresponding to a circulating pump current of 285 amperes. This operation significantly enhances the contact area and turbulence intensity between the slurry and flue gas, optimizing the gas-liquid mass transfer process.

[0050] An oxidation airflow command of 89,000 standard cubic meters per hour is sent to the inlet guide vane adjustment mechanism of the oxidation blower. The core task of this equipment is to ensure the stability of the chemical reaction environment within the slurry tank, particularly the uniformity of pH. Increasing the oxidation airflow accelerates the oxidation of sulfites in the slurry tank, an exothermic process that leads to localized temperature increases and pH fluctuations. Based on real-time data from multiple pH and temperature sensors within the slurry tank, the equipment dynamically fine-tunes the airflow command. It initially commands a rapid increase in airflow to 90,000 to quickly initiate oxidation. When a downward trend in the central pH is detected due to accelerated reaction, the airflow is immediately fine-tuned to 88,500 for buffering, ultimately stabilizing around the target value of 89,000. This dynamic compensation effectively balances oxidation efficiency with the acid-base and thermal balance of the slurry tank, optimizing the thermodynamic process and preventing scaling or equipment corrosion caused by localized over-oxidation or excessively high temperatures.

[0051] In this embodiment of the invention, the beneficial effects are as follows: by generating a three-dimensional spatiotemporal tensor through multi-sensor fusion, the cognitive limitations of traditional point-based measurement are broken; by utilizing multi-scale feature extraction and dynamic system theory, a paradigm upgrade from data monitoring to behavior prediction is achieved; by using constraint optimization solution, multi-objective conflicts are transformed into collaborative control strategies; and finally, through model prediction and dynamic compensation of the actuator, collaborative optimization of desulfurization chemical reaction, gas-liquid mass transfer and thermodynamic processes is achieved.

[0052] Please see Figure 2 , 3 One embodiment of the intelligent desulfurization control method based on machine learning algorithms in this invention includes: 201. Real-time acquisition of full-condition parameters of flue gas treatment process in desulfurization tower through distributed multi-sensor fusion system, including obtaining SO2 concentration spatial distribution matrix by laser spectroscopy analyzer, obtaining slurry pH value by microelectrode array sensor, and obtaining temperature field distribution cloud map by infrared thermal imager. The acquired multi-source heterogeneous data is input into data fusion processor based on tensor theory to generate multi-physics state tensor containing spatiotemporal characteristics.

[0053] Specifically, a high-precision laser spectral analyzer array deployed in the inlet flue, absorption tower body, and outlet flue of the desulfurization tower, combined with advanced signal processing technology, is used to analyze the SO2 concentration distribution in real time. Combined with flue gas velocity sensor data, a three-dimensional spatiotemporal distribution tensor of SO2 concentration containing dynamic distribution characteristics is generated. A high-density microelectrode array embedded in the slurry circulation pipeline and spray system is used to monitor the hydrogen ion activity in the slurry in real time using electrochemical measurement technology. Combined with online slurry density analyzer data, a three-dimensional dynamic response field of slurry pH value containing ion distribution characteristics is generated. A high-resolution infrared thermal imaging system deployed at key thermal measurement points in the desulfurization tower accurately captures the temperature distribution. Combined with pressure sensor data, a temperature-pressure correlation coefficient is generated. The SO2 concentration three-dimensional spatiotemporal distribution tensor is input into a field feature analysis module based on spatial statistical theory. Through spatial correlation calculation and variogram analysis, the statistical distribution characteristics of the concentration field are revealed, and an SO2 diffusion feature descriptor is generated. At the same time, the three-dimensional dynamic response field of slurry pH value and the temperature-pressure co-distribution cloud map are input into a multi-physics field coupling analyzer. Through multivariate statistical analysis, a multi-field coupling feature matrix is ​​generated. The SO2 diffusion feature descriptor and the multi-field coupling feature matrix are input into a tensor fusion system based on data fusion theory. Through feature-level fusion algorithm, a unified multi-physics field state tensor containing multi-physics field co-evolution information of the desulfurization process is constructed. This tensor fully characterizes the key state features of the desulfurization system.

[0054] It should be noted that in the industrial plant with a desulfurization tower 12 meters high and 8 meters in diameter, four sets of high-precision laser spectrometers were installed at the inlet flue, the tower body (at heights of 2 meters, 6 meters, and 10 meters from the bottom), and the outlet flue. Each instrument acquired SO2 concentration data through direct measurement and spatial interpolation at a sampling interval of 100 milliseconds. At t=0 seconds, the SO2 concentration measured at the inlet flue was 2800 mg / m³. 3 After the reaction inside the tower, the outlet concentration dropped to 35 mg / m³. 3 By combining flow velocity sensor data (flow velocity v = 12 m / s), a three-dimensional spatiotemporal distribution tensor T1(x, y, z, t) is generated, where x, y, and z are spatial coordinates (resolution 0.1 meters), and t is time. This tensor can be represented as a three-dimensional matrix at a specific moment, with each point containing the concentration value and flow velocity.

[0055] A high-density microelectrode array (2 mm spacing) was implanted in the slurry circulation pipeline, and hydrogen ion activity was monitored using electrochemical measurement technology. The pH value in the spray layer area was measured to be 5.6, while the online slurry density analyzer showed a density of 1120 kg / m³. 3 Through interpolation calculation, a three-dimensional dynamic response field T2 for the pH value of the slurry is generated. Its dimension is consistent with T1, and each grid point contains the pH value and density parameters.

[0056] A high-resolution infrared thermal imaging system was deployed at the gas-liquid interface of the absorber to acquire temperature data at 500-millisecond intervals. The measured temperature gradient at the interface was 35℃ / m, while a pressure sensor recorded a pressure value of 105 kPa. A temperature-pressure co-distribution cloud map T3 was generated.

[0057] Inputting T1 into the field characteristic analysis module based on spatial statistical theory calculates the spatial autocorrelation and variogram of the concentration field. A spatial autocorrelation coefficient of 0.85 was measured in the central region of the tower, indicating good concentration uniformity at this location. An SO2 diffusion characteristic descriptor D1 is generated, containing statistical characteristic values ​​and distribution uniformity indices. Inputting T2 and T3 into the multiphysics coupling analyzer, through multivariate statistical analysis (calculating the correlation coefficient between pH and temperature), generates a multi-field coupling characteristic matrix M1 with dimensions [time × spatial point × coupling parameter].

[0058] Finally, D1 and M1 are input into a tensor fusion system based on data fusion theory. Through feature-level fusion algorithms (principal component analysis or tensor splicing), a unified multiphysics state tensor Tfinal is constructed, with dimensions [time × spatial point × physical quantity × statistical and coupling features]. At t=1 second, the tensor value of a certain spatial point is [SO2 concentration = 120 mg / m³]. 3 [pH=5.9, temperature=52℃, spatial autocorrelation index=0.82, coupling strength=0.15], which fully characterizes the cooperative state of the multiphysics field.

[0059] 202. Input the generated multiphysics state tensor into the intelligent feature extraction module based on multi-scale feature extraction theory. By constructing a multi-scale analysis framework and calculating the feature evolution flow, analyze the dynamic behavior of the system at multiple spatial scales and generate a feature parameter tensor containing the evolution sequence of key parameters and system stability index.

[0060] Specifically, the multi-physics state tensor is input into the multi-scale analysis modeling unit, and a multi-scale feature representation describing the desulfurization process is constructed using multi-scale analysis methods such as spatial pyramids or wavelet transforms, generating a multi-scale feature description of the system. The multi-scale feature description of the system is input into the feature evolution flow calculation unit, and the evolution of key statistical features (mean, variance, gradient) at different scales is calculated to determine the evolution law of system parameters with the observation scale, generating a multi-scale feature evolution sequence containing reaction intensity, distribution uniformity, and inter-field correlation strength. The multi-scale feature evolution sequence is input into the stability analysis module, and the system stability index is calculated by analyzing the convergence, mutation points, and other behavioral characteristics of feature parameters with scale changes, generating a stability feature set. The multi-scale feature evolution sequence and the stability feature set are input into the feature integration unit, and a unified multi-scale feature description is constructed by tensor splicing or fusion, generating a feature parameter tensor containing a summary information of the system's multi-scale dynamic behavior.

[0061] It should be noted that the multiphysics state tensor obtained from step 201 contains monitoring data of multiple physical quantities at a large number of spatiotemporal points. This tensor is input into the multiscale analysis modeling unit for multiscale feature construction. Taking the state of the desulfurization system at a certain moment as an example, the system contains three main physical fields: flue gas SO2 concentration field (average value is 125 mg / m³). 3 The spatial variation coefficient was 0.15, the slurry pH field (pH value 5.8), and the temperature field (average temperature 54℃). These fields were analyzed at four scales: the original scale (S0), the half-scale (S1), the quarter-scale (S2), and the eighth-scale (S3), using the spatial pyramid method. At each scale, the mean, standard deviation, and correlation coefficient between fields for each physical field were calculated as key parameters for that scale.

[0062] The variation of key parameters with scale was analyzed. The average SO2 concentration field at the S0 scale is 125 mg / m³. 3 At the S1 scale (coarser), it evolves to 130 mg / m³. 3 At the S2 scale, it is 128 mg / m³. 3 At the S3 scale, it is 126 mg / m³. 3The correlation coefficient between the SO2 field and the pH field was -0.15 at the S0 scale, evolved to -0.18 at the S1 scale, -0.20 at the S2 scale, and -0.22 at the S3 scale. This evolution reveals that the strength of the inter-field correlation slightly increases as the observation scale becomes coarser (the region of interest becomes larger).

[0063] Stability indices are calculated by analyzing the smoothness and trend (monotonicity) of the parameter series. The variance of parameter variation across different scales is calculated; a smaller variance indicates greater consistency of the parameter across different scales and a more stable system. The calculated scale consistency index is 0.85 (close to 1 indicates stability), and the trend stability index is 0.78.

[0064] The generated feature parameter tensor consists of two main parts: first, the evolution sequence of key parameters (mean values ​​of each field, correlation coefficients) at four feature scales, forming a parameter matrix; and second, a vector composed of stability indices. By concatenating these two parts of the feature tensor, they are integrated into a unified feature parameter tensor, which describes the statistical behavior and stability of the desulfurization system at different scales, providing a multi-scale feature foundation for subsequent intelligent prediction.

[0065] 203. Input the generated feature parameter tensor into the intelligent prediction module based on dynamic system and stability analysis theory. By constructing the geometric features of the system state space and analyzing the dynamic evolution law of the system, generate multi-objective intelligent prediction results including stability early warning indicators, efficiency optimization intervals and fault risk levels.

[0066] Specifically, the characteristic parameter tensor is input into the state space construction unit. By defining system state variables (key parameters, stability indicators) and the state space, a state space model describing the dynamic evolution of the desulfurization system is constructed, generating the system state space trajectory. The system state space trajectory is input into the dynamic behavior analysis unit. By calculating the dynamic characteristics of the trajectory, such as derivatives, curvature, and Lyapunov exponents, the dynamic evolution characteristics of the system are analyzed, generating a subset of system dynamic behavior descriptions. The subset of system dynamic behavior descriptions is input into the stability and performance evaluation module. By monitoring abnormal changes in dynamic characteristics (Lyapunov exponent turning from negative to positive, curvature abrupt change), the system stability boundary and performance optimization region are identified, generating multi-level stability early warning indicators and performance interval markers. By inputting multi-level stability early warning indicators and performance range identifiers into the risk prediction unit, and by associating the mapping relationship between dynamic feature changes and historical failure modes, a multi-objective intelligent prediction result containing efficiency optimization range and failure risk level is generated.

[0067] It should be noted that the feature parameter tensor obtained from step 202 includes multi-scale feature parameters and stability indices. This tensor is input into the state space construction unit to construct the system state space model. The system state variables are defined as a three-dimensional state space consisting of three key variables: SO2 removal efficiency (currently 95.2%), slurry circulation rate (8500 cubic meters per hour), and system comprehensive stability index (currently 0.82). The system's trajectory over time is formed within this space.

[0068] The local curvature (characterizing the rate of change in direction) and short-term Lyapunov exponent (characterizing sensitivity to initial conditions, i.e., local stability) of the state trajectory were calculated. Near the standard operating point (desulfurization efficiency 95%, stability index 0.85), the measured trajectory curvature was relatively small (0.02), and the Lyapunov exponent was negative (-0.15), indicating system stability. When the system deviated due to inlet SO2 concentration disturbances, at the point (efficiency 94%, stability index 0.65), the trajectory curvature increased (0.25), and the Lyapunov exponent approached zero (-0.02), indicating deteriorating dynamic characteristics and a tendency towards instability.

[0069] A yellow alert is triggered when the Lyapunov index is greater than -0.05, and an orange alert is triggered when it is greater than 0. Simultaneously, regions where the Lyapunov index is less than -0.1 and the curvature is less than 0.1 are identified, corresponding to performance optimization zones with high desulfurization efficiency (>96%) and low energy consumption.

[0070] Establish a correlation between dynamic features (rapid increase in Lyapunov exponent, abrupt change in curvature) and specific faults (scaling, sudden drop in efficiency). When the similarity between the current trajectory features and the feature patterns before the "scaling" fault exceeds 85%, a "moderate risk of scaling" is predicted.

[0071] The generated multi-objective intelligent prediction results include: stability warning level (based on the current Lyapunov index = -0.02, triggering a yellow warning), and efficiency optimization range (slurry circulation rate 8000-9000 m³ / h). 3 / h, corresponding to the region with a Lyapunov index < -0.1, and fault risk level (medium risk of scaling in the absorber tower, low risk of pump overload). These predictions provide operational boundary guidance based on the dynamic behavior of the system for subsequent intelligent decision-making.

[0072] 204. Input the generated multi-objective intelligent prediction results and the multi-physics state tensor into the intelligent decision-making module based on constraint optimization theory. By solving the multi-objective constraint optimization problem, generate a set of intelligent control parameters including the absorbent precise dosing strategy, the slurry circulation optimization scheme and the oxidation air volume adjustment command.

[0073] Specifically, the multi-objective intelligent prediction results and multi-physics state tensors are input into the objective and constraint definition unit. By analyzing the prediction results and real-time states, a multi-objective function (maximizing efficiency, minimizing energy consumption, maximizing stability) and constraints (equipment limits, safety boundaries) for the optimization problem are constructed, generating a mathematical description of the optimization problem. The mathematical description of the optimization problem is input into the constraint optimization solver, which uses multi-objective optimization algorithms (NSGA-II, weighted summation) to solve for the constrained Pareto optimal solution set, generating a candidate optimization solution set. The candidate optimization solution set is input into the decision evaluation unit, which selects the final decision solution from the candidate optimization solution set by weighing the satisfaction of each objective, generating a basic control parameter set. The basic control parameter set is input into the setpoint mapping unit, which derives the final intelligent control parameter set by establishing the correspondence between the optimization objective and the setpoint of the underlying actuator.

[0074] It should be noted that a multi-objective function can be expressed as a comprehensive optimization objective function: in These are decision variables (such as slurry dosage, circulation volume, and oxidation air volume). For desulfurization efficiency, The target desulfurization efficiency is (e.g., 95%). This represents the total energy consumption of the system. This is a baseline value for energy consumption (e.g., 900 kW·h). This is a system stability index (range 0-1). , , The weighting coefficients and + + =1, dynamically adjusted according to the current warning level (e.g., set during a yellow warning). > > (Stability is the priority). Constraints include limiting the slurry circulation rate to 6000-10000 m³. 3 / h, pH value safety range is 5.0-6.5, oxidation air volume range is 8000-15000m³ / h. 3 / h, equipment power limit, and must meet the desulfurization efficiency η(u) ≥ (Efficiency lower limit required by environmental protection standards).

[0075] The mathematical description of the optimization problem is input into the constrained optimization solver. A multi-objective optimization algorithm (NSGA-II, weighted summation, etc.) is used to solve for the constrained Pareto optimal solution set, generating a candidate optimization solution set. This candidate solution set is then input into the decision evaluation unit. By weighing the satisfaction of each objective, the final decision solution is selected from the candidate solution set, generating the basic control parameter set. Finally, the basic control parameter set is input into the setpoint mapping unit. By establishing the correspondence between the optimization objective and the setpoints of the underlying actuators, the final control setpoint set is derived.

[0076] The multi-objective intelligent prediction results obtained from step 203 show that the system is in a yellow alert state, and the efficiency optimization range is slurry circulation rate of 8000-9000 m³ / h. 3 / h, the system needs to move towards a more stable region, and the multiphysics state tensor shows that the current SO2 concentration is 2850 mg / m³. 3 The slurry pH value was 5.3 and the temperature was 52℃.

[0077] After defining the above comprehensive optimization objective function and constraints, the constrained optimization solver uses a multi-objective optimization algorithm to solve the problem, obtaining a set of Pareto optimal solutions. Each solution represents a compromise solution to the objective function. Solution A: efficiency 95.5%, energy consumption 890 kWh, stability 0.75; Solution B: efficiency 96.2%, energy consumption 910 kWh, stability 0.82; Solution C: efficiency 94.8%, energy consumption 870 kWh, stability 0.85.

[0078] Based on the current priority of eliminating early warnings and improving stability (i.e., the weighting coefficient is biased), We choose solution B as the final decision solution because it has the best stability while maintaining high efficiency.

[0079] Based on the system state requirements corresponding to solution B and combined with the process model, the setpoint mapping unit calculates the required actuator setpoints: limestone slurry addition rate is 12.8 cubic meters per hour, slurry circulation rate is 8500 cubic meters per hour, and oxidation air volume is 11000 cubic meters per hour. This scheme is expected to increase desulfurization efficiency to 96.2%, system stability index to 0.82, and simultaneously lower the stability warning level from yellow to green.

[0080] 205. The generated set of intelligent control parameters is output to the multi-actuator collaborative control system, including a model-predictive limestone slurry precision metering device, an adaptive slurry circulation frequency conversion speed regulation system, and an intelligent pH dynamic compensation device, to achieve intelligent optimization control of the desulfurization chemical reaction process, gas-liquid mass transfer process, and thermodynamic process.

[0081] Specifically, the absorbent dosing strategy from the intelligent control parameter set is input into the advanced slurry dosing control unit. The limestone slurry metering device is adjusted through the model predictive control (MPC) algorithm to generate precise dosing control commands. The slurry circulation optimization scheme from the intelligent control parameter set is input into the variable frequency speed control module of the circulating pump. The output of the circulating pump frequency converter is adjusted through adaptive PID control to generate flow optimization control commands. The oxidation air volume adjustment command from the intelligent control parameter set is input into the air volume intelligent control unit. The speed of the oxidation fan and the guide vane angle are adjusted through feedforward-feedback composite control to generate precise air volume control signals. The slurry dosing control command, slurry flow field optimization command, and oxidation air volume control signal are input into the multi-loop collaborative controller. Through decoupling control or coordinated optimization algorithms, multi-actuator collaborative operation parameters are generated. The multi-actuator collaborative operation parameters are input into the real-time dynamic optimizer. By monitoring the desulfurization efficiency, system pressure drop, and energy consumption indicators online, the final dynamic adjustment commands for each actuator are generated and output to the corresponding control equipment.

[0082] It should be noted that the obtained set of intelligent control parameters includes: limestone slurry addition rate of 12.8 cubic meters per hour, slurry circulation rate of 8500 cubic meters per hour, and oxidation air volume of 11000 cubic meters per hour.

[0083] Slurry dosing control: After receiving the dosing strategy, the advanced slurry dosing control unit uses the model predictive control (MPC) algorithm based on the slurry pipeline and reaction kinetic model to calculate the optimal valve opening sequence, so as to achieve smooth and precise flow control and avoid overshoot.

[0084] The circulating pump variable frequency speed control module uses the circulation volume command as the set value and adopts an adaptive PID controller. Based on the pump's flow-speed characteristic curve and pipeline characteristics, it dynamically adjusts the PID parameters and outputs the frequency command of the frequency converter to ensure that the flow rate is stable at the target value.

[0085] The intelligent air volume control unit, based on air volume commands, employs a composite control strategy of feedforward (based on changes in inlet SO2 concentration) and feedback (based on actual air volume or oxygen content in the tower) to adjust the fan speed and guide vane opening, thereby achieving rapid and precise air volume adjustment.

[0086] The multi-loop coordinated controller takes into account the coupling relationship between the three loops of slurry addition, circulation, and oxidation (increasing the slurry will affect the load of the circulation pump, and increasing the air volume may affect the flow field inside the tower). It adopts decoupling control or coordination strategy to fine-tune the control commands of each loop to ensure optimal overall coordination.

[0087] The real-time dynamic optimizer continuously monitors the system response (desulfurization efficiency, pressure drop, energy consumption), compares it with the expected target, and makes small-range dynamic adjustments to the setpoint (fine-tuning the circulation rate to 8520m). 3 / h), and issue the final instructions to each implementing agency to achieve precise and coordinated control of the desulfurization process.

[0088] 206. Achieve system self-organization and reconfiguration through system dynamic analysis, abnormal state identification, adaptive control, and multi-objective optimization.

[0089] Specifically, the characteristic parameter tensors are input into the system dynamic analysis module. By constructing a system dynamic model and a state observer, the module calculates state estimates and predictions describing the dynamic evolution of the system, generating a description of the system's dynamic behavior. This description is then input into the abnormal state identification unit. By analyzing the deviations and trends of key parameters, the module identifies abnormal operating conditions and potential risks in the system, generating abnormal operating condition identification results. These results are then input into the adjustment unit based on adaptive control theory. By adjusting controller parameters or setpoints, the unit constructs a recovery strategy for the system under abnormal operating conditions, generating a set of adaptive control strategies. This set of adaptive control strategies is input into the multi-objective optimization module. By balancing multiple objectives such as desulfurization efficiency, energy consumption, and stability, the module generates system reconfiguration instructions. Finally, the system reconfiguration instructions are input into the control execution coordinator. By mapping the optimization instructions to the actions of specific actuators, the module generates self-organizing reconfiguration instructions for the system and feeds them back to each control execution unit in real time.

[0090] It should be noted that the feature parameter tensor obtained from step 202 includes key parameters and stability indices. This tensor is then input into the system dynamic analysis module.

[0091] Construct a state-space model or input-output model of the system. Estimate state variables that cannot be directly measured at present using a state observer, and predict the system's behavior trend in the near future, predicting that the desulfurization efficiency may decrease by 1.5% within the next 5 minutes.

[0092] Analyze the deviations and trends of key parameters (SO2 removal efficiency and stability indicators) from their normal baseline values. When the predicted efficiency decreases by more than 1% and the stability indicator is below 0.7, it is identified as an abnormal operating condition of "efficiency decline and stability degradation".

[0093] The adaptive control unit adjusts the parameters of the underlying controller (the circulating pump controller in step 205) based on the identified anomaly type (increasing the integral action to eliminate steady-state error more quickly), or fine-tunes its setpoint (temporarily increasing the circulation rate setpoint by 50m). 3 / h), to initially suppress the abnormality.

[0094] Under the new abnormal operating conditions, the multi-objective optimization module re-runs the optimization process of step 204, but considers the new constraints (equipment has reached its limit, new efficiency targets) to find the optimal operating point that maintains performance as much as possible under the current conditions. The new optimized solution is: slurry dosage 13.5 m³. 3 / h, circulation volume 8650m 3 / h, oxidation air volume 10800m³ 3 / h.

[0095] The generated self-organizing reconfiguration instruction adjusts the setpoints of each actuator based on the new optimized solution. Through real-time feedback control, the system gradually recovers from the abnormal state of "decreased efficiency and deteriorated stability" to the optimal operating range of safety and efficiency.

[0096] In this embodiment of the invention, a breakthrough from point-based measurement to three-dimensional field-state perception is achieved through multi-sensor fusion and tensor theory. A cognitive leap from data monitoring to system behavior prediction is completed based on multi-scale feature extraction and dynamic system theory. Constraint optimization and multi-objective decision-making solve the synergistic challenges of efficiency, energy consumption, and stability. Dynamic optimization and fault self-healing of the desulfurization process are achieved through actuator collaborative control and system self-organization and reconfiguration. This reduces SO2 emission concentration fluctuations, lowers system energy consumption, and shortens response time to abnormal operating conditions, solving problems such as perception limitations, control lag, target conflicts, and insufficient adaptability inherent in traditional desulfurization systems, thus achieving a balance between environmental compliance and economic operation.

[0097] The present invention also provides a desulfurization intelligent control device based on machine learning algorithms. The desulfurization intelligent control device based on machine learning algorithms includes a memory and a processor. The memory stores computer-readable instructions. When the computer-readable instructions are executed by the processor, the processor performs the steps of the desulfurization intelligent control method based on machine learning algorithms in the above embodiments.

[0098] The present invention also provides a computer-readable storage medium, which can be a non-volatile computer-readable storage medium or a volatile computer-readable storage medium, wherein the computer-readable storage medium stores instructions that, when the instructions are executed on a computer, cause the computer to perform the steps of the desulfurization intelligent control method based on machine learning algorithms.

[0099] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working processes of the systems, devices, and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0100] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0101] The above-described 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 the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A desulfurization intelligent control system based on machine learning algorithms, characterized in that, include: The acquisition module collects parameters of the flue gas treatment process in the desulfurization tower through a multi-sensor fusion system, inputs the multi-source heterogeneous data into the data fusion processor, and generates a multi-physics state tensor. The analysis module analyzes the dynamic behavior of the system based on the multiphysics state tensor and generates characteristic parameter tensors. The prediction module constructs geometric features and analyzes evolution laws based on the feature parameter tensor to generate multi-objective intelligent prediction results; The decision-making module generates a set of intelligent control parameters based on the multi-objective intelligent prediction results and the multi-physics state tensor.

2. The intelligent desulfurization control system based on machine learning algorithms according to claim 1, characterized in that, By using a laser spectral analyzer array and processing technology, the SO2 concentration distribution is analyzed to generate a three-dimensional spatiotemporal distribution tensor of SO2 concentration. By implanting circulation tubing and a high-density microelectrode array, the hydrogen ion activity in the slurry is monitored, and combined with the analyzer data, a three-dimensional dynamic response field of the slurry pH value is generated. A collaborative distribution cloud map is generated based on high-resolution infrared thermal imaging system combined with pressure sensor data; Based on the three-dimensional spatiotemporal distribution tensor of SO2 concentration, an SO2 diffusion feature descriptor is generated. The three-dimensional dynamic response field of slurry pH value and the cooperative distribution cloud map are input into a multi-physics field coupling analyzer to generate a multi-field coupling feature matrix. A multiphysics state tensor is constructed based on the SO2 diffusion characteristic descriptor and the multi-field coupling characteristic matrix.

3. The intelligent desulfurization control system based on machine learning algorithms according to claim 2, characterized in that, Based on the multi-physics state tensor, a multi-scale feature representation is constructed to generate a multi-scale feature description of the system; Based on the evolution law of the calculation parameters described by the multi-scale feature description of the system, a multi-scale feature evolution sequence is generated; The behavioral characteristics of feature parameters are analyzed using the multi-scale feature evolution sequence to generate a set of stable features; A feature parameter tensor is generated based on the multi-scale feature evolution sequence and the stability feature set.

4. The intelligent desulfurization control system based on machine learning algorithms according to claim 3, characterized in that, include: A state-space model is constructed based on the aforementioned feature parameter tensors, and the system state-space trajectory is generated. By analyzing the system's state-space trajectory, the dynamic evolution characteristics of the system are analyzed, and a subset of descriptions of the system's dynamic behavior is generated. Based on the aforementioned subset of system dynamic behavior descriptions, multi-level stability warning indicators and performance range identifiers are generated; Multi-objective intelligent prediction results are generated based on the multi-level stability early warning indicators and performance range identifiers.

5. The intelligent desulfurization control system based on machine learning algorithms according to claim 4, characterized in that, include: Based on the multi-objective intelligent prediction results and the multi-physics state tensor, construct the multi-objective function and constraints of the optimization problem, and generate the mathematical description of the optimization problem; The mathematical description of the optimization problem is input into the constraint optimization solver to generate a set of candidate optimization solutions. The candidate optimization solution set is used to generate a set of basic control parameters; The intelligent control parameter set is obtained based on the aforementioned basic control parameter set.

6. The intelligent desulfurization control system based on machine learning algorithms according to claim 5, characterized in that, Based on the multi-objective intelligent prediction results and the multiphysics state tensor, a multi-objective function and constraints for the optimization problem are constructed. The multi-objective function is expressed as a comprehensive optimization objective function: in As decision variables, For desulfurization efficiency, To achieve the target desulfurization efficiency, This represents the total energy consumption of the system. This is the energy consumption baseline value. As a system stability indicator, , , The weighting coefficients and + + =1.

7. The intelligent desulfurization control system based on machine learning algorithms according to claim 5, characterized in that, Based on the aforementioned set of basic control parameters, a set of intelligent control parameters is obtained by establishing a correspondence between the optimization target and the set values ​​of the underlying actuator.

8. The intelligent desulfurization control system based on machine learning algorithms according to claim 1, characterized in that, It includes an output module that outputs the set of intelligent control parameters to a multi-actuator collaborative control system to achieve intelligent optimization control of desulfurization.

9. The intelligent desulfurization control system based on machine learning algorithms according to claim 7, characterized in that, The limestone slurry metering device is adjusted according to the absorbent dosing strategy in the intelligent control parameter set to generate precise dosing control commands. The slurry circulation optimization scheme in the intelligent control parameter set is adjusted by adaptive PID control to regulate the output of the circulation pump frequency converter and generate flow optimization control commands. Based on the oxidation air volume in the intelligent control parameter set, the speed of the oxidation fan and the angle of the guide vanes are adjusted to generate a precise air volume control signal; The precise dosing control command, the flow optimization control command, and the air volume precise control signal are input into the multi-loop collaborative controller to generate collaborative operation parameters for multiple actuators. The multiple actuators are collaboratively input into a real-time dynamic optimizer to generate dynamic adjustment commands for the actuators.

10. The intelligent desulfurization control system based on machine learning algorithms according to claim 1, characterized in that, It also includes achieving system self-organization and reconfiguration through system dynamic analysis, abnormal state identification, adaptive control, and multi-objective optimization: Based on the aforementioned feature parameter tensors, state estimation and prediction of the dynamic evolution of the system are calculated to generate a description of the system's dynamic behavior. The description of the system's dynamic behavior involves analyzing the deviations and trends of key parameters to identify abnormal operating conditions and potential risks in the system's operation, and generating abnormal operating condition identification results. Based on the abnormal operating condition identification results, a recovery strategy for the system under abnormal operating conditions is constructed, and an adaptive control strategy set is generated; The adaptive control strategy set is optimized by balancing desulfurization efficiency, energy consumption, and stability to generate system reconfiguration instructions.

Images