Intelligent water vapor centralized sampling detection system for thermal power plant

Abstract

The invention provides an intelligent water vapor centralized sampling detection system for a thermal power plant, and relates to the technical field of computer information processing.The intelligent water vapor centralized sampling detection system is characterized in that a monitoring risk field intensity graph is constructed, double-light-path differential scanning is carried out on a detection air mass of the thermal power plant, a penetration spectrum is obtained, decoupling inversion is carried out on the penetration spectrum, and a water vapor concentration field is obtained; determining a dynamic risk state matrix according to the monitoring risk field intensity graph and the water vapor concentration field, and performing dual-network deviation correction control on the dynamic risk state matrix to obtain a sniffing course angle; performing upwind sampling according to the sniffing course angle to obtain sniffing water vapor parameters, and performing baseline drift compensation on the sniffing water vapor parameters to obtain a water vapor convergence set; and determining a reconstruction risk vector set through the water vapor convergence set and the dynamic risk state matrix, and determining the water vapor corrosion risk through the reconstruction risk vector set, so that high-precision dynamic sensing and intelligent centralized sampling of the water vapor parameters of the thermal power plant can be realized, and the accuracy of water vapor corrosion risk prediction is improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of industrial visual detection, and more particularly, to an intelligent water vapor centralized sampling detection system for thermal power plants. BACKGROUND

[0002] Industrial visual detection technology is an important means for key state perception and target recognition in complex working conditions. Industrial visual detection can integrate optical imaging, spectral analysis and artificial intelligence decision algorithms, and has advantages such as non-contact, high precision, fast dynamic response, etc., and is widely used in intelligent manufacturing, energy monitoring and environmental governance, etc.

[0003] In the operation system of a thermal power plant, the water vapor system is a core link connecting the boiler, the steam turbine and the condenser. The temperature, pressure, moisture content and other parameters of the water vapor system directly affect the thermal efficiency and operation safety of the equipment. Therefore, the intelligent water vapor centralized sampling detection method for thermal power plants has become an important means to realize fine operation control and safety protection. The existing water vapor detection technology mainly relies on fixed-point sampling combined with spectral analysis or electrochemical analysis. A fixed measurement point layout is usually used, and the water vapor risk is judged based on the single-path detection result. However, the existing technology mostly uses static sensing devices for point layout, which is difficult to cover dynamic change areas, resulting in that the water vapor abnormal aggregation area is easily missed. Moreover, the spectral sampling path is single, and there is a lack of deep modeling and real-time regulation of the risk field state. The sampling path is not optimized, which cannot reflect the actual water vapor concentration peak in time, resulting in low precision of the corrosion early warning mechanism. Therefore, how to realize high-precision dynamic perception and intelligent centralized sampling of water vapor parameters in thermal power plants, and further improve the accuracy of water vapor corrosion risk prediction is a difficult problem in the industry. SUMMARY

[0004] The present application provides an intelligent water vapor centralized sampling detection system for thermal power plants, which can realize high-precision dynamic perception and intelligent centralized sampling of water vapor parameters in thermal power plants, and further improve the accuracy of water vapor corrosion risk prediction.

[0005] The present application provides an intelligent water vapor centralized sampling detection system for thermal power plants, which can realize high-precision dynamic perception and intelligent centralized sampling of water vapor parameters in thermal power plants, and further improve the accuracy of water vapor corrosion risk prediction.

[0006] A risk field construction module is configured to construct a monitoring risk field intensity map based on the wind speed vector, the wind direction angle and the thermodynamic parameters of the thermal power plant.

[0007] A water vapor decoupling inversion module is configured to perform double-light-path differential scanning on the detection air mass of the thermal power plant to obtain a penetration spectrum, and then perform decoupling inversion on the penetration spectrum to obtain a water vapor concentration field.

[0008] The intelligent sniffing module is configured to determine a dynamic risk state matrix according to the monitoring risk field intensity map and the water vapor concentration field, perform double-network deviation correction control on the dynamic risk state matrix based on reinforcement learning decision-making, and obtain a sniffing heading angle during water vapor sampling.

[0009] The intelligent compensation module is configured to perform an upwind sampling operation on the detection air mass of the thermal power plant according to the sniffing heading angle, obtain sniffing water vapor parameters, and perform baseline drift compensation on the sniffing water vapor parameters to obtain a water vapor convergence set.

[0010] The intelligent evaluation module is configured to perform convergence reconstruction on the dynamic risk state matrix through the water vapor convergence set to obtain a reconstructed risk vector set, and further determine a water vapor corrosion risk of the thermal power plant from the reconstructed risk vector set.

[0011] In this embodiment, constructing a monitoring risk field intensity map based on a wind speed vector, a wind direction angle, and thermodynamic parameters of the thermal power plant specifically includes:

[0012] The wind speed vector, the wind direction angle, and the thermodynamic parameters of the thermal power plant are collected through an environmental perception device;

[0013] The wind speed vector, the wind direction angle, and the thermodynamic parameters of the thermal power plant are modeled through spatial coordinate information of the thermal power plant to obtain the monitoring risk field intensity map.

[0014] In this embodiment, performing double-optical-path differential scanning on the detection air mass of the thermal power plant to obtain a penetration spectrum specifically includes:

[0015] A waveband laser signal is emitted from both sides of the optical path by using an infrared laser emitting device to obtain a transmission light intensity corresponding to each side of the optical path;

[0016] The transmission light intensity corresponding to each side of the optical path is compared based on a differential algorithm to obtain the penetration spectrum.

[0017] In this embodiment, decoupling and inversion are performed on the penetration spectrum to obtain a water vapor concentration field, specifically including:

[0018] The penetration spectrum is preprocessed to obtain a preprocessed penetration spectrum, wherein the preprocessing includes wavelength correction, baseline normalization, and signal denoising;

[0019] The preprocessed penetration spectrum is processed by a multi-peak nonlinear least squares method to obtain a water vapor absorption characteristic;

[0020] An absorption coefficient and an integral absorbance of water vapor are extracted from the water vapor absorption characteristic;

[0021] A water vapor concentration field is output based on a concentration inversion equation from the absorption coefficient and the integral absorbance.

[0022] In the embodiment, the determining a dynamic risk state matrix according to the monitored risk field intensity map and the water vapor concentration field specifically includes:

[0023] The risk field intensity map is grid partitioned, and then a wind speed vector and a wind direction angle of each grid partition are extracted;

[0024] A spatial gradient of the water vapor concentration field is extracted, and a concentration gradient vector is determined based on an eight-quadrant division method and the spatial gradient;

[0025] A multi-dimensional state feature is determined according to the wind speed vector, the wind direction angle and the concentration gradient vector of each grid partition;

[0026] The multi-dimensional state feature is matrix encoded to obtain a dynamic risk state matrix.

[0027] In the embodiment, the double-net deviation correction control of the dynamic risk state matrix based on a reinforcement learning decision-making is performed to obtain a sniffing heading angle during water vapor sampling specifically includes:

[0028] A heading action set is generated based on a reinforcement learning strategy network through the dynamic risk state matrix;

[0029] A value function estimation is performed on the heading action set through a main value evaluation network and an auxiliary value network respectively to obtain an expected heading angle and a deviation correction value;

[0030] The sniffing heading angle during water vapor sampling is determined according to the expected heading angle and the deviation correction value.

[0031] In the embodiment, the detection air mass of the thermal power plant is sampled against the wind according to the sniffing heading angle to obtain sniffing water vapor parameters specifically includes:

[0032] A wind speed variation and a wind direction variation around the detection air mass of the thermal power plant are obtained;

[0033] The sniffing heading angle is corrected based on the wind speed variation and the wind direction variation to obtain an upwind sampling path;

[0034] Sniffing water vapor parameters are collected based on the upwind sampling path through an infrared laser absorption device.

[0035] In the embodiment, the sniffing water vapor parameters are baseline drift compensated to obtain a water vapor convergence set specifically includes:

[0036] Original signal intensities of absorption peaks in the sniffing water vapor parameters are obtained;

[0037] A baseline offset is obtained by analyzing a change trend of the original signal intensities;

[0038] Compensate and correct the baseline offset based on the compensation function to obtain a steady-state water vapor signal sequence;

[0039] Smoothly filter and remove outliers from the steady-state water vapor signal sequence to obtain a water vapor convergence set.

[0040] In this embodiment, the dynamic risk state matrix is reconstructed by the water vapor convergence set to obtain a reconstructed risk vector set, which specifically includes:

[0041] The water vapor convergence set and the dynamic risk state matrix are matched in convergence value, and then a convergence matrix is obtained.

[0042] The convergence matrix and the dynamic risk state matrix are updated by weighting, and then a reconstructed risk vector set is obtained.

[0043] In this embodiment, the water vapor corrosion risk of the thermal power plant is determined by the reconstructed risk vector set, which specifically includes:

[0044] The peak value, duration and diffusion trend of the concentrated water vapor concentration of the thermal power plant are determined by the reconstructed risk vector set.

[0045] The corrosion risk index is determined according to the peak value, duration and diffusion trend of the concentrated water vapor concentration of the thermal power plant.

[0046] The risk warning information is judged according to the corrosion risk index, and then the water vapor corrosion risk of the thermal power plant is obtained.

[0047] The technical scheme provided by the embodiments disclosed in the application has the following beneficial effects:

[0048] Based on the wind speed vector, wind direction angle and thermodynamic parameters of the thermal power plant, a monitoring risk field intensity map is constructed. The detection air mass of the thermal power plant is scanned by double optical path difference to obtain a penetration spectrum, and then the penetration spectrum is decoupled and inverted to obtain a water vapor concentration field. The dynamic risk state matrix is determined according to the monitoring risk field intensity map and the water vapor concentration field, the dynamic risk state matrix is controlled by double network rectification based on reinforcement learning decision, and the sniffing heading angle at the water vapor sampling time is obtained. The detection air mass of the thermal power plant is sampled against the wind according to the sniffing heading angle to obtain sniffing water vapor parameters, and the baseline drift compensation is performed on the sniffing water vapor parameters to obtain a water vapor convergence set. The dynamic risk state matrix is reconstructed by the water vapor convergence set to obtain a reconstructed risk vector set, and then the water vapor corrosion risk of the thermal power plant is determined by the reconstructed risk vector set.

[0049] It can be seen that in the present application, high-precision dynamic perception and intelligent centralized sampling of water vapor parameters of thermal power plants can be realized. First, the risk field intensity map constructed based on the wind speed vector, wind direction angle and thermodynamic parameters can comprehensively depict the risk situation distribution of the local space of the thermal power plant, providing risk prior support for subsequent sniffing path planning and target area identification. Second, through the water vapor decoupling inversion module to perform double light path differential scanning and spectral decoupling analysis, not only the extraction ability of weak signals in complex air masses is enhanced, but also the water vapor concentration field at different heights and directions can be accurately obtained, realizing dynamic capture of multi-dimensional water vapor distribution. Then, through the intelligent sniffing module to fuse the concentration field and risk field information, combined with the reinforcement learning strategy network and the double value network for rectification control, the optimal sniffing heading angle can be dynamically generated, significantly improving the environmental adaptability and spatio-temporal matching degree of the water vapor sampling path. Then, through the intelligent compensation module to perform upwind sampling and baseline drift compensation and signal optimization on the sampling signal, effectively reducing the influence of environmental disturbance on the sampling result, ensuring that the sampling data has strong steady-state convergence. Finally, based on the collected water vapor convergence set, the dynamic risk state matrix is reconstructed, which can realize accurate prediction of the corrosion risk vector, thereby forming a high-reliability water vapor corrosion risk assessment model, and comprehensively improving the operation safety of the power plant and the accuracy of the corrosion early warning.

[0050] In summary, the technical scheme adopted by the present application can realize high-precision dynamic perception and intelligent centralized sampling of water vapor parameters of thermal power plants, thereby improving the accuracy of water vapor corrosion risk prediction. BRIEF DESCRIPTION OF DRAWINGS

[0051] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the drawings needed to be used in the embodiments or prior art description will be briefly introduced. Obviously, the drawings in the following description are only the embodiments of the present application, and other drawings can be obtained by those skilled in the art without creative labor.

[0052] Figure 1 is a module structure diagram of an intelligent water vapor centralized sampling detection system for thermal power plants provided by the present application;

[0053] Figure 2 is a flowchart for determining the penetration spectrum in some embodiments of the present application;

[0054] Figure 3 is an exemplary flowchart for determining the dynamic risk state matrix in some embodiments of the present application. DETAILED DESCRIPTION

[0055] With reference to the accompanying drawings, the technical solutions in the embodiments of the present application will be described clearly and completely. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all the embodiments of the present application. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative work fall within the scope of the present application.

[0056] The embodiment of the present application provides a kind of intelligent water vapor centralized sampling detection system for thermal power plant, its core is based on the wind speed vector of thermal power plant, wind direction angle and thermodynamic parameter constructs monitoring risk field intensity chart;Double optical path difference scanning is carried out to the detection air mass of thermal power plant, and the penetration spectrum is obtained, and then the water vapor concentration field is obtained by decoupling inversion to the penetration spectrum;Dynamic risk state matrix is determined according to the monitoring risk field intensity chart and the water vapor concentration field, and double network rectification control is carried out to the dynamic risk state matrix based on reinforcement learning decision, and the sniffing heading angle when water vapor sampling is obtained;According to the sniffing heading angle, the detection air mass of thermal power plant is operated against the wind sampling, and the sniffing water vapor parameter is obtained, the baseline drift compensation is carried out to the sniffing water vapor parameter, and the water vapor convergence set is obtained;The dynamic risk state matrix is reconstructed by convergence through the water vapor convergence set, and the reconstructed risk vector set is obtained, and then the water vapor corrosion risk of thermal power plant is determined from the reconstructed risk vector set.

[0057] In order to better understand the above technical solutions, the above technical solutions will be described in detail in combination with the drawings in the specification and specific embodiments. Referring to Figure 1 As shown in the figure, the figure is a module structure diagram of an intelligent water vapor centralized sampling detection system for thermal power plant according to the present application. The sampling detection system includes a risk field construction module 100, a water vapor decoupling inversion module 200, an intelligent sniffing module 300, an intelligent compensation module 400 and an intelligent evaluation module 500, which are described as follows:

[0058] The risk field construction module 100 is used to construct a monitoring risk field intensity chart based on the wind speed vector, wind direction angle and thermodynamic parameter of the thermal power plant.

[0059] In the embodiment, the monitoring risk field intensity chart can be constructed based on the wind speed vector, wind direction angle and thermodynamic parameter of the thermal power plant by the following method, i.e.:

[0060] The wind speed vector, wind direction angle and thermodynamic parameter of the thermal power plant are collected by an environmental perception device.

[0061] The wind speed vector, wind direction angle and thermodynamic parameter of the thermal power plant are modeled by the spatial coordinate information of the thermal power plant, and the monitoring risk field intensity chart is obtained.

[0062] In practice, firstly, wind speed vectors, wind direction angles, and thermodynamic parameters are acquired through environmental sensing devices deployed within the thermal power plant. These devices include 3D ultrasonic anemometers, infrared thermal imagers, integrated temperature and humidity sensors, and air pressure monitoring devices, which can be deployed within the thermal power plant according to actual needs. The wind speed vector represents the speed and direction of airflow, the wind direction angle describes the projection angle of the wind speed vector onto the horizontal plane, and the thermodynamic parameters include ambient temperature, relative humidity, absolute air pressure, and saturated vapor pressure. Then, the wind speed vectors, wind direction angles, and thermodynamic parameters of the thermal power plant are normalized using existing Z-Score technology. The spatial coordinates of the thermal power plant are obtained through its building information model, which can be derived from the plant's construction layout. A 3D interpolation algorithm is used to map the normalized wind speed vectors, wind direction angles, and thermodynamic parameters onto the spatial coordinates, generating a monitoring risk field strength map.

[0063] It should be noted that the monitoring risk field strength map in this application refers to a multi-dimensional rasterized map used to characterize the intensity and spatial distribution of environmental disturbances within the area of ​​a thermal power plant. The risk field strength refers to the superimposed response intensity of areas with drastic changes in wind and thermal parameters. The monitoring risk field strength map has the characteristics of high spatial resolution and strong environmental modeling capabilities, which can be used to guide the identification of high-risk areas and help to identify areas in thermal power plants where water vapor may accumulate in advance.

[0064] The water vapor decoupling inversion module 200 is used to perform dual-optical-path differential scanning on the detected gas mass of a thermal power plant to obtain a transmission spectrum, and then to decouple and invert the transmission spectrum to obtain a water vapor concentration field.

[0065] It should be noted that, in this application, the detection air mass of a thermal power plant refers to a local air mass with a certain water vapor concentration that can be identified by optical equipment and sampled and analyzed during the operation of a thermal power plant due to factors such as boiler emissions, pipeline leaks, equipment cooling, or environmental heat and humidity exchange. The detection air mass is dynamic and non-uniform, and its boundary may change with wind speed and direction, heat flow disturbance, and plant structure.

[0066] Preferred, Reference Figure 2 As shown in the figure, this is a schematic flowchart of the process for determining the transmission spectrum in some embodiments of this application. In this embodiment, the process of performing dual-path differential scanning on the detection gas mass of a thermal power plant to obtain the transmission spectrum specifically includes the following steps:

[0067] In step S21, an infrared laser emitting device is used to emit band laser signals from both optical paths to obtain the transmitted light intensity corresponding to each optical path.

[0068] In step S22, the transmission spectrum is obtained by comparing the transmission light intensity corresponding to each side light path based on a difference algorithm.

[0069] It should be noted that the infrared laser emitting device in the present application refers to an active optical detection device for emitting specific waveband infrared laser to detect air mass in a thermal power plant and receiving transmission information thereof, for obtaining the absorption characteristics of gas components in the target area to the infrared waveband, so as to realize accurate construction of the transmission spectrum, wherein the infrared laser emitting device comprises an optoelectronic conversion unit for converting the received transmission laser signal into a corresponding electrical signal, and performing preliminary amplification and digitization processing on the electrical signal to obtain transmission light intensity information at different wavelengths.

[0070] In specific implementation, first, the infrared laser emitting device is symmetrically arranged on the upwind side and the downwind side of the detected air mass, wherein the infrared laser emitting device can generate adjustable waveband laser covering the range of 1.3-2.5 μm, the laser beam is emitted by the infrared laser emitting device, and the transmission light intensity is collected by the optoelectronic conversion unit in the infrared laser emitting device to obtain the transmission light intensity corresponding to each side light path; then, the transmission light intensity of the left and right light paths is compared based on the light path symmetry principle using a difference algorithm, and the transmission spectrum can be obtained by normalization operation and background subtraction using the difference algorithm.

[0071] It should be noted that the transmission spectrum in the present application refers to a relative absorption spectrum obtained by comparing the transmission light intensity of the air mass through the double-side path difference, for describing the energy attenuation characteristics of the laser under the absorption of water vapor molecules; in the present embodiment, the influence of environmental background light, path disturbance and device system drift on the spectrum result can be effectively suppressed through the double-path difference mechanism, and the spectrum line quality and absorption characteristic resolution capability are improved.

[0072] In the present embodiment, the transmission spectrum is decoupled and inverted to obtain the water vapor concentration field, which can be achieved in the following manner, that is:

[0073] The transmission spectrum is preprocessed to obtain a preprocessed transmission spectrum, wherein the preprocessing includes wavelength correction, baseline normalization and signal denoising;

[0074] The preprocessed transmission spectrum is processed by a multi-peak nonlinear least squares method to obtain water vapor absorption characteristics.

[0075] The absorption coefficient and the integral absorbance of water vapor are extracted from the water vapor absorption characteristics.

[0076] The absorption coefficient and the integral absorbance are output based on the concentration inversion equation to obtain the water vapor concentration field.

[0077] In a specific implementation, first, the preprocessing includes wavelength correction, baseline normalization and signal denoising, wherein the wavelength correction refers to aligning the spectrum axis with the standard wavelength and eliminating the spectral line shift caused by device drift, the baseline normalization can use the minimum-maximum normalization method to keep the spectral absorption intensity relatively consistent, and the signal denoising can use the wavelet transform method to remove random noise interference; second, the preprocessed transmission spectrum is processed by a multi-peak nonlinear least squares method through a Gaussian function, and the center wavelength, half-height width, peak height and the like of the preprocessed transmission spectrum are extracted, and then the center wavelength, half-height width, peak height and the like are taken as water vapor absorption characteristics, which refer to a set of characteristic parameters extracted from the spectrum and capable of reflecting the existence and concentration of water vapor; then, the water vapor absorption characteristics are substituted into the Beer-Lambert law to calculate the absorption coefficient of water vapor, wherein the absorption coefficient is an index for describing the light intensity attenuation of the wavelength under unit concentration, and the integral absorbance is the total area of the characteristic absorption peak in a specific waveband, which can be calculated by integrating the peak height in the water vapor absorption characteristics and is used to reflect the overall absorption capacity of water vapor; finally, the absorption coefficient and the integral absorbance are substituted into the calculation based on the Beer-Lambert law deformation equation, and the water vapor concentration value corresponding to the spatial positioning is output, and then a spatial interpolation algorithm is used to construct a water vapor concentration field.

[0078] It should be noted that the water vapor concentration field in the present application refers to a water vapor distribution intensity map in a three-dimensional space obtained by inversion of an absorption model based on laser transmission spectrum characteristics, and the unit is mg / m 3 , which can accurately express the dynamic change characteristics of the water vapor content at different positions in the plant area; in the embodiment, the nonlinear least squares fitting and integral absorbance inversion method can improve the parameter extraction accuracy under weak absorption conditions and water vapor disturbance background.

[0079] The intelligent sniffing module 300 is configured to determine a dynamic risk state matrix according to the monitored risk field intensity map and the water vapor concentration field, perform double-network deviation correction control on the dynamic risk state matrix based on reinforcement learning decision, and obtain a sniffing heading angle during water vapor sampling.

[0080] Preferably, as shown in Figure 3 , the figure is an exemplary flowchart for determining a dynamic risk state matrix in some embodiments of the present application, and the determination of the dynamic risk state matrix according to the monitored risk field intensity map and the water vapor concentration field in the embodiment specifically includes the following steps:

[0081] In step S31, the risk field intensity map is subjected to grid partition processing, and then the wind speed vector and the wind direction angle of each grid partition are extracted;

[0082] In step S32, a spatial gradient of the water vapor concentration field is extracted, and a concentration gradient vector is determined based on an eight-quadrant division method and the spatial gradient;

[0083] In step S33, a multi-dimensional state feature is determined according to the wind speed vector, the wind direction angle and the concentration gradient vector of each grid partition;

[0084] In step S34, the multi-dimensional state feature is matrix encoded to obtain a dynamic risk state matrix.

[0085] In a specific implementation, first, the risk field intensity map is subjected to grid partition processing, that is, the risk field intensity map is divided into grid units by using a two-dimensional coordinate system, and the grid units are divided by unit length until the last grid unit is smaller than the unit length, where the unit length is set to 1, and the unit risk field intensity is 1. Figure 1 For each grid region, the wind speed vector is obtained through instantaneous wind speed data measured by the sensor array, and the wind direction angle is obtained through the wind direction sensor. Second, the spatial gradient of the water vapor concentration field is extracted, specifically, the Sobel operator is used to calculate the gradient of the water vapor concentration data of each grid point to obtain the change rate of the concentration in the horizontal and vertical directions, and the change rate of the concentration in the horizontal and vertical directions is divided based on the eight-quadrant division method to obtain the concentration gradient vector of the water vapor concentration field, which can reflect the concentration gradient size and the concentration gradient direction angle of the water vapor concentration. Then, for each grid partition, the wind speed size, the wind direction angle, the concentration gradient size and the concentration gradient direction angle are combined to form a four-dimensional feature vector, that is, the multi-dimensional state feature = [wind speed size, wind direction angle, gradient intensity, gradient direction angle], which can represent the comprehensive dynamic state of the current wind field and the water vapor concentration field of the grid partition. Finally, the multi-dimensional state features of each grid partition are encoded one by one according to the grid number to obtain a dynamic risk state matrix, and each row of the dynamic risk state matrix represents the risk state feature of a grid, which is convenient for subsequent model analysis, recognition or prediction processing.

[0086] It should be noted that the dynamic risk state matrix in the present application refers to a structured expression matrix for describing the current risk state of each spatial region of a thermal power plant, and each row of the matrix represents the risk state of a grid region, and each column represents a different feature dimension, including wind speed size, wind direction angle, concentration gradient direction, etc., which is beneficial to realize intelligent sampling control of multiple targets and the whole region. In the present embodiment, the eight-quadrant method can improve the discrete modeling capability of the concentration change direction and enhance the expression discrimination degree of the state vector.

[0087] In the present embodiment, the dynamic risk state matrix is subjected to double-net deviation correction control based on reinforcement learning decision to obtain the sniffing heading angle during water vapor sampling, which can be implemented in the following manner, that is:

[0088] The reinforcement learning policy network generates a heading action set based on the dynamic risk state matrix;

[0089] The main value evaluation network and the auxiliary value network respectively perform value function estimation on the heading action set to obtain an expected heading angle and a correction value;

[0090] The sniffing heading angle during water vapor sampling is determined according to the expected heading angle and the correction value.

[0091] In a specific implementation, first, a reinforcement learning policy network is constructed, and the dynamic risk state matrix is taken as input data of the network. The reinforcement learning policy network can adopt a deep neural network architecture, such as a three-layer fully connected network. The reinforcement learning policy network outputs a heading action set, for example: [θ1, θ2, …, θn], where θ represents a candidate sniffing heading angle of a sensing device, the unit is degree, and the value range is generally between 0-360°, and 1-n represents the serial number of the candidate sniffing heading angle. n Then, the main value evaluation network and the auxiliary value network are respectively used to process the heading action set. The main value evaluation network is used for heading decision output under the current policy, receives a heading action and corresponding state features as input each time, outputs the sum of the immediate reward and the long-term value of the heading, and represents the overall advantages and disadvantages of the direction at the current time. The auxiliary value network corrects and constrains the output of the main network, selects the heading angle corresponding to the highest value as the expected heading angle according to the value function evaluation result, and calculates the difference between the main network and the auxiliary network as a correction value. Finally, the sum of the expected heading angle and the correction value is taken as the sniffing heading angle, for example: the expected heading angle is 60°, and the correction value is -2°, and the final heading angle is 58°.

[0092] It should be noted that the sniffing heading angle in the present application refers to an intelligent navigation parameter for guiding the water vapor sampling device to autonomously adjust the direction in a complex disturbance environment, which can be dynamically calculated based on the environment state by a reinforcement learning model, and has the characteristics of strong real-time and high adaptability. In the present embodiment, the double-network correction control refers to a double-channel value evaluation mechanism in which a main value network and an auxiliary value network are introduced into the reinforcement learning architecture. The main network focuses on global strategy optimality evaluation, and the auxiliary network is used to enhance the robustness judgment of local disturbance and error fluctuation, which can avoid strategy deviation or local oscillation caused by single value evaluation, and is beneficial to improving the heading control accuracy in unstable wind field and complex water vapor disturbance working conditions.

[0093] The intelligent compensation module 400 is configured to perform an upwind sampling operation on the detected air mass of the thermal power plant according to the sniffing heading angle, obtain sniffing water vapor parameters, and perform baseline drift compensation on the sniffing water vapor parameters to obtain a water vapor convergence set.

[0094] In the embodiment, the detection air mass of the thermal power plant is sampled against the wind according to the sniffing heading angle, and the sniffing water vapor parameters are obtained by using the following method, that is,

[0095] The wind speed variation and the wind direction variation around the detection air mass of the thermal power plant are obtained.

[0096] The sniffing heading angle is corrected based on the wind speed variation and the wind direction variation, and the upwind sampling path is obtained.

[0097] The infrared laser absorption device is used to collect the sniffing water vapor parameters based on the upwind sampling path.

[0098] In the specific implementation, first, the wind speed variation and the wind direction variation around the detection air mass are obtained in real time by using a distributed wind speed and direction sensor array, wherein the wind speed variation represents the instantaneous change value of the local area wind speed per unit time, and the wind direction variation represents the offset amplitude of the wind direction angle. Preferably, the sampling frequency of the distributed wind speed and direction sensor array can be set to 1-5 Hz to ensure the environmental response sensitivity. Then, the wind speed variation and the wind direction variation are calculated by using a shortest disturbance path planning algorithm, and the sniffing heading angle is dynamically corrected based on the calculation result by using the shortest disturbance path planning algorithm to obtain the upwind sampling path. The upwind sampling path is the best motion route that enters the main wind flow direction in the opposite direction and avoids the high disturbance area. Finally, the infrared laser absorption device is controlled to perform air mass penetration sampling along the upwind path, to collect the water vapor absorption signals on the path, and to take the collected water vapor absorption signals as the sniffing water vapor parameters. It should be noted that the infrared laser absorption device in the embodiment can use an existing tunable laser absorption spectroscopy technology device, including a laser emitter, a collimating lens, a gas absorption cell, and a detector, which can measure the water vapor concentration based on the transmittance change of infrared light at a specific water vapor absorption wavelength.

[0099] It should be noted that the upwind sampling path in the present application refers to the heading trajectory corrected in the opposite direction of the wind direction, which is used to enable the sampling device to penetrate the center of the detection air mass against the air flow, to improve the pertinence and effectiveness of sampling. The sniffing water vapor parameters refer to the key water vapor indicators collected by the infrared laser absorption method, including the concentration, temperature, and infrared absorption characteristics, which can reflect the spatial distribution state of the water vapor in the thermal power plant environment. In addition, the infrared laser absorption device has the advantages of fast response speed and strong anti-interference ability, and is suitable for real-time water vapor sampling under wind field disturbance, which is conducive to improving the accuracy of subsequent risk modeling and corrosion trend analysis.

[0100] In the embodiment, the sniffing water vapor parameters are baseline drift compensated to obtain a water vapor convergence set by using the following method, that is,

[0101] acquire original signal intensity of each absorption peak in the sniffing water vapor parameter;

[0102] analyze the trend of the original signal intensity to obtain baseline offset;

[0103] compensate and correct the baseline offset based on a compensation function to obtain a steady-state water vapor signal sequence;

[0104] smooth filter and remove outliers from the steady-state water vapor signal sequence to obtain a water vapor convergence set.

[0105] In a specific implementation, first, a plurality of absorption peak positions are identified by a peak detection algorithm, and the original light intensity value at the corresponding time is recorded, and then the original light intensity value sequence in time order is taken as the original signal intensity sequence; second, the original signal intensity sequence is modeled by a sliding average algorithm, and then the trend change of the original signal intensity in the time dimension is determined through visual modeling analysis, and the trend change is taken as the baseline offset, that is, the trend curve obtained by modeling analysis is taken as the baseline offset, which is used to represent the drift degree of the system response; then, the value of the baseline offset at each time point is subtracted from the original signal intensity, thereby offsetting the errors caused by factors such as laser drift, temperature and humidity changes, and mirror contamination, to obtain a steady-state water vapor signal sequence, and preferably, the compensation and correction process can be completed by the scipy.optimize.curvefit library in Python; finally, a moving average filter is used to smooth the signal sequence, which can eliminate the instantaneous jitter caused by environmental fluctuations, and then the Z-score method is used to identify outliers, for example, the Z-score threshold is set to ±2.5, and the abnormal points exceeding the threshold are removed, and the remaining data is the signal set after baseline compensation and cleaning, and then the signal set is taken as the water vapor convergence set.

[0106] It should be noted that the water vapor convergence set in the present application refers to a water vapor parameter set with high time sequence stability and noise suppression capability after drift compensation, signal smoothing and outlier removal processing, which is used to reflect the steady-state water vapor concentration state under real working conditions, and the water vapor convergence set can more effectively avoid measurement errors caused by equipment jitter, wind disturbance or instrument thermal drift, and can represent the stable characteristics of the air mass body; in addition, the baseline drift in the present embodiment refers to the overall curve offset phenomenon of the laser spectrum absorption signal caused by environmental changes or equipment aging.

[0107] The intelligent evaluation module 500 is configured to perform convergence reconstruction on the dynamic risk state matrix based on the water vapor convergence set to obtain a reconstructed risk vector set, and then determine the water vapor corrosion risk of the thermal power plant based on the reconstructed risk vector set.

[0108] In the embodiment, the water vapor convergence set is used to converge and reconstruct the dynamic risk state matrix, and the following method can be used:

[0109] The convergence value of the water vapor convergence set and the dynamic risk state matrix is matched, and a convergence matrix is obtained.

[0110] The convergence matrix and the dynamic risk state matrix are updated by weighting, and a reconstructed risk vector set is obtained.

[0111] In a specific implementation, first, a representative value (such as a smoothed water vapor concentration value or a gradient change rate) at a corresponding position in the water vapor convergence set in the region is extracted from the dynamic risk state matrix, and the representative value and the water vapor convergence value are quantitatively compared by using the minimum distance matching principle. For example, the Euclidean distance between the representative value and the water vapor convergence value can be calculated, and all the Euclidean distances are arranged into a matrix as a convergence matching matrix according to the corresponding positions of the dynamic risk state matrix, where each element in the convergence matching matrix represents the convergence degree between the current grid state and the water vapor concentration response. Then, a weighting coefficient can be set, and each state unit in the original dynamic risk state matrix is fused by using the weighting coefficient. All the updated state units are recombined and arranged to form an updated multi-dimensional reconstruction matrix. The vector extraction and dimension compression of the multi-dimensional reconstruction matrix are performed by using an existing matrix conversion technology, and a reconstructed risk vector set is generated. Preferably, the weighting coefficient can be adaptively set according to the data density, self-confidence or transmission reliability of the water vapor convergence set under the current working condition.

[0112] It should be noted that the convergence matrix in the present application refers to a matrix structure used to describe the convergence degree of the water vapor stable state to the current dynamic risk characteristics, and is an intermediate bridge connecting the perception data and the state modeling. The reconstructed risk vector set refers to a new risk expression set obtained by fusing the convergence matrix and the original state matrix information, and can be used for accurate modeling and hierarchical evaluation of the corrosion risk trend. In the embodiment, the convergence reconstruction can compensate for the deviation of the predicted state under the wind field disturbance, which is beneficial to enhancing the real evolution ability of the model to the water vapor behavior, and further improving the accuracy of the system in the high-risk area identification and dynamic regulation.

[0113] In the embodiment, the water vapor corrosion risk of the thermal power plant is determined by using the reconstructed risk vector set in the following manner:

[0114] The concentration peak value, duration and diffusion trend of the water vapor in the thermal power plant are determined by using the reconstructed risk vector set.

[0115] The corrosion risk index is determined according to the concentration peak value, duration and diffusion trend of the water vapor in the thermal power plant.

[0116] According to the corrosion risk index, risk early warning information is determined, and the water vapor corrosion risk of the thermal power plant is obtained.

[0117] In a specific implementation, first, statistical analysis is performed on the water vapor concentration in different time periods and spatial regions in the reconstruction risk vector set. The maximum concentration value extracted through the sliding time window is taken as the concentrated water vapor concentration peak value, and the time period during which the maximum concentration value exists continuously is recorded as the duration. Based on the water vapor coordinate changes at each time in the reconstruction risk vector set, the air mass movement trajectory is obtained by linear regression fitting, so as to determine the main direction of water vapor diffusion and obtain the diffusion trend. Then, the concentrated water vapor concentration peak value, the duration, and the diffusion trend are input into a preset corrosion risk calculation model for evaluation. Preferably, the corrosion risk index can be calculated in the following manner. The corrosion risk calculation model is: corrosion risk index = a x peak concentration + b x duration + g x diffusion intensity, where a, b, and g are empirical weight coefficients, which can be set according to historical equipment corrosion records and working conditions. Finally, the calculated corrosion risk index is compared with the system preset risk classification threshold to obtain the risk level, for example, low risk when the corrosion risk index is less than 50, medium risk when the corrosion risk index is between 50 and 75, and high risk when the corrosion risk index is greater than 75. The risk level is taken as the risk early warning information, and the water vapor corrosion risk result of the thermal power plant is output.

[0118] It should be noted that the corrosion risk index in the present application is a risk quantitative index formed based on the comprehensive evaluation of the space-time evolution characteristics of water vapor concentration, which is used to reflect the comprehensive strength of the potential for moisture corrosion in the region of the thermal power plant. The corrosion risk index comprehensively considers the concentration level, action duration, and spatial expansion ability, and is more targeted and predictive, which is beneficial to identifying high-risk areas and key nodes of corrosion risk in advance. In addition, the processing method of combining the sliding window with the spatial coordinate difference is adopted in the present application, which can take into account local precision and overall trend analysis, and is convenient for realizing accurate early warning and judgment under the condition of complex wind field interference.

[0119] In summary, the technical solution adopted in the present application can realize high-precision dynamic perception and intelligent centralized sampling of water vapor parameters in a thermal power plant, thereby improving the accuracy of water vapor corrosion risk prediction.

[0120] The computer program instructions can also be loaded onto a computer, other programmable data processing apparatus, or other processing devices to cause a series of operational steps to be performed on the computer, other programmable apparatus or other devices to produce a computer implemented process such that the instructions which execute on the computer or other programmable apparatus provide processes for implementing the functions specified in the flowchart block or blocks. Figure 1 The flowchart and / or block diagram in the Figures illustrate the architecture, functionality, and operation of possible implementations of systems, methods and computer program products according to various embodiments. In this regard, each flowchart block and / or combination of flowchart blocks can be implemented by various means, such as hardware, software, firmware, processor, circuitry, and / or other Figure 1 The flowchart and / or block diagram in the Figures illustrate the architecture, functionality, and operation of possible implementations of systems, methods and computer program products according to various embodiments. In this regard, each flowchart block and / or combination of flowchart blocks can be implemented by various means, such as hardware, software, firmware, processor, circuitry, and / or other

[0121] Those skilled in the art can understand that all or part of the steps in the above-mentioned various methods of the embodiments can be completed by a program instructing relevant hardware, and the program can be stored in a computer readable storage medium, including Read-Only Memory (ROM), Random Access Memory (RAM), Programmable Read-only Memory (PROM), Erasable Programmable Read Only Memory (EPROM), One-time Programmable Read-Only Memory (OTPROM), Electrically-Erasable Programmable Read-Only Memory (EEPROM), Compact Disc Read-Only Memory (CD-ROM) or other optical disk storage, magnetic disk storage, magnetic tape storage, or any other medium that can be used to carry or store data.

[0122] It should also be noted that the terms "comprising", "containing", or any other variant thereof, are intended to cover a non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements does not include only those elements, but can also include other elements not expressly listed or inherent to such process, method, article, or apparatus. Without further limitation, an element preceded by "comprises a" does not, without more constraints, foreclose the existence of additional identical elements in the process, method, article, or apparatus that includes the element.

Claims

1. A smart water vapor centralized sampling and detection system for a thermal power plant, characterized in that, The sampling detection system comprises: a risk field construction module, configured to construct a monitoring risk field intensity map based on a wind speed vector, a wind direction angle and thermodynamic parameters of the thermal power plant; a water vapor decoupling inversion module, configured to perform double-light-path differential scanning on a detection air mass of the thermal power plant to obtain a transmission spectrum, and further perform decoupling inversion on the transmission spectrum to obtain a water vapor concentration field; an intelligent sniffing module, configured to determine a dynamic risk state matrix according to the monitoring risk field intensity map and the water vapor concentration field, perform double-network deviation correction control on the dynamic risk state matrix based on reinforcement learning decision-making, and obtain a sniffing heading angle during water vapor sampling; an intelligent compensation module, configured to perform headwind sampling on the detection air mass of the thermal power plant according to the sniffing heading angle to obtain sniffing water vapor parameters, and perform baseline drift compensation on the sniffing water vapor parameters to obtain a water vapor convergence set; an intelligent evaluation module, configured to perform convergence reconstruction on the dynamic risk state matrix through the water vapor convergence set to obtain a reconstructed risk vector set, and further determine a water vapor corrosion risk of the thermal power plant from the reconstructed risk vector set.

2. The intelligent water vapor centralized sampling and detection system for a thermal power plant according to claim 1, characterized in that, The method for constructing a monitoring risk field intensity map based on a wind speed vector, a wind direction angle and thermodynamic parameters of the thermal power plant comprises: collecting the wind speed vector, the wind direction angle and the thermodynamic parameters of the thermal power plant through an environmental perception device; modeling the wind speed vector, the wind direction angle and the thermodynamic parameters of the thermal power plant through spatial coordinate information of the thermal power plant to obtain the monitoring risk field intensity map.

3. The intelligent water vapor centralized sampling and detection system for a thermal power plant of claim 1, wherein, The method for performing double-light-path differential scanning on a detection air mass of the thermal power plant to obtain a transmission spectrum comprises: emitting a waveband laser signal from a double-sided light path by using an infrared laser emitting device to obtain a transmission light intensity corresponding to each side light path; performing transmission comparison on the transmission light intensity corresponding to each side light path based on a difference algorithm to obtain a transmission spectrum.

4. The intelligent water vapor centralized sampling and detection system for a thermal power plant of claim 1, wherein, The method for performing decoupling inversion on the transmission spectrum to obtain a water vapor concentration field comprises: preprocessing the transmission spectrum to obtain a preprocessed transmission spectrum, wherein the preprocessing includes wavelength correction, baseline normalization and signal denoising; performing multi-peak nonlinear least squares processing on the preprocessed transmission spectrum to obtain a water vapor absorption characteristic; extracting an absorption coefficient and an integral absorbance of water vapor from the water vapor absorption characteristic; outputting the absorption coefficient and the integral absorbance based on a concentration inversion equation to obtain a water vapor concentration field.

5. A smart water vapor centralized sampling and detection system for a thermal power plant as claimed in claim 1, wherein, The method for determining a dynamic risk state matrix according to the monitoring risk field intensity map and the water vapor concentration field comprises: performing grid partitioning processing on the risk field intensity map, and further extracting a wind speed vector and a wind direction angle of each grid partitioning; extracting a spatial gradient of the water vapor concentration field, and determining a concentration gradient vector based on an eight-quadrant division method and the spatial gradient; determining a multi-dimensional state feature according to the wind speed vector, the wind direction angle and the concentration gradient vector of each grid partitioning; matrix encoding the multi-dimensional state feature to obtain a dynamic risk state matrix.

6. A smart water vapor centralized sampling and detection system for a thermal power plant as claimed in claim 1, wherein, The method for performing double-network deviation correction control on the dynamic risk state matrix based on reinforcement learning decision-making to obtain a sniffing heading angle during water vapor sampling comprises: generating a heading action set through the dynamic risk state matrix based on a reinforcement learning strategy network; The main value evaluation network and the auxiliary value network respectively estimate the value function of the set of heading actions to obtain an expected heading angle and a deviation value; The sniffing heading angle of the water vapor sampling is determined according to the expected heading angle and the deviation value.

7. A smart water vapor centralized sampling and detection system for a thermal power plant as claimed in claim 1, wherein, The detection air mass of the thermal power plant is sampled against the wind according to the sniffing heading angle to obtain sniffing water vapor parameters, which specifically include: The wind speed variation and the wind direction variation around the detection air mass of the thermal power plant are obtained; The deviation of the sniffing heading angle is corrected based on the wind speed variation and the wind direction variation to obtain a windward sampling path; The sniffing water vapor parameters are collected based on the windward sampling path by an infrared laser absorption device.

8. A smart water vapor centralized sampling and detection system for a thermal power plant as claimed in claim 1, wherein, Baseline drift compensation is performed on the sniffing water vapor parameters to obtain a water vapor convergence set, which specifically includes: The original signal intensity of each absorption peak in the sniffing water vapor parameters is obtained; The baseline offset is obtained by analyzing the change trend of the original signal intensity; The baseline offset is compensated and corrected based on a compensation function to obtain a steady-state water vapor signal sequence; The steady-state water vapor signal sequence is smoothed and outlier removed to obtain a water vapor convergence set.

9. A smart water vapor centralized sampling and detection system for a thermal power plant as claimed in claim 1, wherein, The dynamic risk state matrix is reconstructed by the water vapor convergence set to obtain a reconstructed risk vector set, which specifically includes: The convergence value of the water vapor convergence set and the dynamic risk state matrix is matched to obtain a convergence matrix; The convergence matrix and the dynamic risk state matrix are weighted and updated to obtain a reconstructed risk vector set.

10. The intelligent water vapor centralized sampling and detection system for a thermal power plant of claim 1, wherein, The water vapor corrosion risk of the thermal power plant is determined by the reconstructed risk vector set, which specifically includes: The concentrated water vapor concentration peak value, duration and diffusion trend of the thermal power plant are determined by the reconstructed risk vector set; The corrosion risk index is determined according to the concentrated water vapor concentration peak value, duration and diffusion trend of the thermal power plant; The risk warning information is judged according to the corrosion risk index, and the water vapor corrosion risk of the thermal power plant is obtained.

Images