Intelligent monitoring method for convection heating surface of biomass bulk combustion process

Abstract

The present application relates to the technical field of boiler, and especially relates to a kind of intelligent monitoring method of convection heating surface in biomass bulk combustion process, the temperature and pressure of each preset monitoring position of convection heating surface are obtained to form a real-time sample, all real-time samples are classified using target support vector machine model, and the real-time samples on interval boundary and in interval boundary are obtained, if target support vector machine model needs model parameter updating, the adjustment bias determination coefficient of penalty parameter and adjustment coefficient are obtained according to the real-time samples on interval boundary and in interval boundary, and new penalty parameter is obtained;According to new penalty parameter, model parameter updating is carried out to target support vector machine model, and the final classification result of final classification of all real-time samples is obtained, the state of ash deposition on convection heating surface is monitored according to final classification result, and model parameter is dynamically adjusted to better adapt to real-time sample classification.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of boilers, in particular to an intelligent monitoring method for a convection heating surface in a biomass bulk combustion process. BACKGROUND

[0002] In the biomass bulk combustion process, the convection heating surface is an important part of the boiler, and its performance directly affects the thermal efficiency and emission quality of the boiler. Biomass fuel (such as wood chips, straw, rice husk, etc.) has characteristics such as high volatility, low ash melting point, and high alkali metal content. During combustion, it is easy to produce deposits such as solid fly ash particles, and the ash deposits on the convection heating surface. The thickness of the ash deposits on the convection heating surface affects the performance of the boiler, the corrosion of the convection heating surface, shortens the service life of the equipment, and affects the safe operation of the coal-fired unit. Therefore, the monitoring of the ash deposits on the convection heating surface is a very important process to timely discover and handle the ash deposition problem and ensure the performance of the boiler.

[0003] In the prior art, the monitoring data (such as temperature and pressure) of the convection heating surface is analyzed for the ash deposition state characteristics of the convection heating surface by using a support vector machine model to determine whether the convection heating surface needs to be sooted blown, so as to perform the soot blowing operation. Since the support vector machine model has a lag, that is, the trained support vector machine model will always use fixed model parameters for model classification. However, when the biomass fuel is replaced, a sudden ash deposition event occurs, which causes the result of the model classification to deviate and the soot blowing operation to be misoperated.

[0004] Therefore, how to enhance the robustness and generalization of the support vector machine model to ensure the accuracy of the ash deposition characteristic analysis of the convection heating surface has become a problem to be solved. SUMMARY

[0005] Therefore, the embodiments of the present application provide an intelligent monitoring method for a convection heating surface in a biomass bulk combustion process to solve the problem of how to enhance the robustness and generalization of the support vector machine model to ensure the accuracy of the ash deposition characteristic analysis of the convection heating surface.

[0006] The embodiments of the present application provide an intelligent monitoring method for a convection heating surface in a biomass bulk combustion process, which comprises the following steps:

[0007] The temperature and pressure of each preset monitoring position of the convection heating surface are obtained to form a real-time sample, and a target support vector machine model is used to perform current classification on all real-time samples to obtain real-time samples on and inside the interval boundary, and the target support vector machine model refers to the support vector machine model after the last final classification of the current classification;

[0008] judging whether the target support vector machine model needs model parameter updating according to the support vectors, if the target support vector machine model needs model parameter updating, obtaining an adjustment bias determination coefficient of the penalty parameter according to the probability that each real-time sample belongs to noise and the real-time samples on and inside the margin boundary;

[0009] obtaining an adjustment coefficient of the penalty parameter according to the difference between the sample closest to the decision boundary inside the margin boundary of the current classification and the sample closest to the decision boundary inside the margin boundary of the last final classification, and obtaining a new penalty parameter according to the adjustment bias determination coefficient and the adjustment coefficient;

[0010] updating the model parameters of the target support vector machine model according to all real-time samples and the new penalty parameter to obtain an updated support vector machine model, performing final classification on all real-time samples by using the updated support vector machine model to obtain a final classification result, and monitoring the ash deposition state on the convective heating surface according to the final classification result.

[0011] Preferably, the support vectors include normal samples and ash deposition samples, and the judging whether the target support vector machine model needs model parameter updating according to the support vectors includes:

[0012] obtaining a proportion value and a quantity difference value of the normal samples and the ash deposition samples in the support vectors of the current classification and a proportion value and a quantity difference value of the normal samples and the ash deposition samples in the support vectors of a preset number of historical classifications before the current classification to form a proportion value sequence and a quantity difference value sequence;

[0013] respectively performing first-order difference processing on the proportion value sequence and the quantity difference value sequence to correspondingly obtain a proportion difference value sequence and a quantity difference value sequence, calculating the mean value of all elements except the last element in the proportion difference value sequence, denoted as a proportion difference value threshold, and calculating the mean value of all elements except the last element in the quantity difference value sequence, denoted as a quantity difference value threshold;

[0014] if the last element in the proportion difference value sequence is greater than the proportion difference value threshold and the last element in the quantity difference value sequence is greater than the quantity difference value threshold, it is determined that the target support vector machine model needs model parameter updating.

[0015] Preferably, the obtaining an adjustment bias determination coefficient of the penalty parameter according to the probability that each real-time sample belongs to noise and the real-time samples on and inside the margin boundary includes:

[0016] obtaining a probability that each of the real-time samples belongs to noise based on the historical samples at the preset number of historical classifications, marking a noise sample according to the probability that each of the real-time samples belongs to noise, and counting a number of noise samples in the real-time samples on and inside the interval boundary;

[0017] For any real-time sample on and inside the interval boundary, obtaining a Euclidean distance between the any real-time sample and each real-time sample on and inside the interval boundary, obtaining a mean value of the Euclidean distance corresponding to the any real-time sample, calculating a mean value of the mean values of the Euclidean distance, and recording a difference value between a constant 1 and a reciprocal of the mean value as a sample dispersion degree;

[0018] calculating a proportion of the number of noise samples in a total number of real-time samples on and inside the interval boundary, and obtaining an adjustment bias determination coefficient of a penalty parameter according to the proportion and the mean value of the sample dispersion degree.

[0019] Preferably, the obtaining of the probability that each of the real-time samples belongs to noise based on the historical samples at the preset number of historical classifications comprises:

[0020] For any historical sample in any historical classification, obtaining a straight-line distance from the any historical sample to a decision boundary of the any historical classification, taking a reciprocal of the straight-line distance as an influence coefficient of the any historical sample, obtaining an influence coefficient of each support vector in the any historical classification, taking a minimum influence coefficient as an influence coefficient threshold, and taking the any historical sample as a target historical sample if the influence coefficient of the any historical sample is greater than or equal to the influence coefficient threshold.

[0021] obtaining the target historical samples at the preset number of historical classifications, performing density clustering on all target historical samples and all real-time samples to obtain at least one cluster center, obtaining a minimum Euclidean distance between any real-time sample and each of the cluster centers, taking the any real-time sample as a center, taking a preset multiple of the minimum Euclidean distance as a radius, constructing a circular region, and counting a number of samples in the circular region.

[0022] obtaining a difference value between a reciprocal of the minimum Euclidean distance and a constant 1, obtaining a reciprocal of an added value of the number of samples and the constant 1, and performing weighted summation on the difference value and the reciprocal of the added value to obtain the probability that the any real-time sample belongs to noise.

[0023] Preferably, the difference between the position of the sample closest to the decision boundary within the interval boundary of the current classification and the sample closest to the decision boundary within the interval boundary of the last final classification is used to obtain an adjustment coefficient of the penalty parameter, including:

[0024] In the interval boundary of the current classification, the sample closest to the decision boundary within the upper interval and the sample closest to the decision boundary within the lower interval are calculated, which are denoted as the upper sample and the lower sample, and the minimum Euclidean distance between the upper sample and the lower sample is obtained;

[0025] The upper sample and the lower sample closest to the decision boundary within the interval boundary of the last final classification are obtained to obtain a minimum reference Euclidean distance, and the reciprocal of the absolute value of the difference between the minimum reference Euclidean distance and the minimum Euclidean distance is calculated, and the difference between the constant 1 and the reciprocal is used as an adjustment coefficient of the penalty parameter.

[0026] Preferably, the new penalty parameter is obtained according to the adjustment bias determination coefficient and the adjustment coefficient, including:

[0027] The penalty parameter of the target support vector machine model is obtained, the product of the penalty parameter and the adjustment coefficient is obtained, denoted as an adjustment value, the vertical distance between the interval boundary and the decision boundary of the current classification is obtained, the difference between the constant 1 and the reciprocal of the vertical distance is calculated, and the sum of the preset value and the difference is denoted as an adjustment bias determination coefficient threshold;

[0028] If the adjustment bias determination coefficient is greater than or equal to the adjustment bias determination coefficient threshold, the difference between the penalty parameter of the target support vector machine model and the adjustment value is used as the new penalty parameter; if the adjustment bias determination coefficient is less than the adjustment bias determination coefficient threshold, the sum of the penalty parameter of the target support vector machine model and the adjustment value is used as the new penalty parameter.

[0029] Preferably, the model parameter of the target support vector machine model is updated according to all real-time samples and the new penalty parameter to obtain an updated support vector machine model, including:

[0030] All real-time samples are clustered to obtain at least one cluster, for any cluster, the distance between the cluster center of the any cluster and each target historical sample is calculated, the target historical sample corresponding to the minimum distance is denoted as a reference sample, and the sample classification label of each real-time sample in the any cluster is obtained according to the classification result of the reference sample;

[0031] The sample classification label of each real-time sample is obtained, and after the penalty parameter in the target support vector machine model is replaced by the new penalty parameter, other model parameters in the target support vector machine model are updated based on each real-time sample and the corresponding sample classification label, so as to obtain an updated support vector machine model.

[0032] Preferably, the monitoring of the ash deposition state on the convective heating surface according to the final classification result comprises:

[0033] The final classification result comprises normal samples and ash deposition samples, the distance between each ash deposition sample and the final classification decision boundary is calculated to obtain a distance mean value, and the difference between the reciprocal of the distance mean value and a constant 1 is obtained to obtain a real-time ash deposition degree; the number of preset monitoring positions involved in the ash deposition sample is obtained to obtain a proportion of the number in the total number of preset monitoring positions, which is recorded as an ash deposition range.

[0034] If the ash deposition range is less than a preset ash deposition range threshold value, and the real-time ash deposition degree is greater than or equal to a preset ash deposition degree threshold value, only a local soot-blowing operation is performed on the area where the preset monitoring position involved in the ash deposition sample is located.

[0035] If the ash deposition range is greater than or equal to a preset ash deposition range threshold value, and the real-time ash deposition degree is greater than or equal to a preset ash deposition degree threshold value, a whole soot-blowing operation is performed on the convective heating surface.

[0036] Compared with the prior art, the embodiment of the present application has the following beneficial effects:

[0037] The application obtains the temperature and pressure of each preset monitoring position of the convective heating surface to form a real-time sample, uses a target support vector machine model to perform current classification on all real-time samples to obtain real-time samples on and inside a support vector interval boundary, the target support vector machine model refers to a support vector machine model after the last final classification of the current classification, judges whether the target support vector machine model needs model parameter updating according to the support vector, if the target support vector machine model needs model parameter updating, obtains an adjustment bias judgment coefficient of a penalty parameter according to the probability that each real-time sample belongs to noise and the real-time samples on and inside the interval boundary, obtains an adjustment coefficient of the penalty parameter according to the position difference between the sample closest to the decision boundary inside the interval boundary of the current classification and the sample closest to the decision boundary inside the interval boundary of the last final classification, and obtains a new penalty parameter according to the adjustment bias judgment coefficient and the adjustment coefficient, performs model parameter updating on the target support vector machine model according to all real-time samples and the new penalty parameter to obtain an updated support vector machine model, performs final classification on all real-time samples using the updated support vector machine model to obtain a final classification result, and monitors the ash deposition state on the convective heating surface according to the final classification result. Wherein, the model parameters of the support vector machine model are dynamically adjusted and updated based on the last classification result of the support vector machine model and in combination with the result of the support vector machine model in the current classification to obtain the updated support vector machine model, which improves the robustness and generalization ability of the support vector machine model, better applies the classification of real-time samples, and ensures the accuracy of the ash deposition judgment based on the real-time classification result. BRIEF DESCRIPTION OF DRAWINGS

[0038] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the following will briefly introduce the drawings needed to be used in the embodiments or prior art description. Obviously, the drawings in the following description are only some embodiments of the present application, and other drawings can be obtained by those skilled in the art without creative labor.

[0039] Figure 1 It is a method flow chart of a biomass bulk combustion process convective heating surface intelligent monitoring method provided by the first embodiment of the present application. DETAILED DESCRIPTION

[0040] The embodiments of the present application will be described in detail below, and examples of the embodiments are shown in the drawings. The embodiments described below by referring to the drawings are exemplary and are intended to explain the present application, and cannot be understood as a limitation of the present application.

[0041] It should be noted that the terms "first", "second", etc. in the specification of the present disclosure and the above-mentioned drawings are used to distinguish similar objects, and do not necessarily describe a specific order or sequence. It should be understood that the data thus used can be interchanged under appropriate circumstances, so that the embodiments of the present disclosure described herein can be implemented in an order other than those illustrated or described herein. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with the present disclosure. Rather, they are merely examples of devices and methods consistent with some aspects of the present disclosure.

[0042] In order to illustrate the technical solutions of the present application, the following will be illustrated by specific embodiments.

[0043] Referring to Figure 1 , it is a method flow chart of a biomass bulk material combustion process convection heating surface intelligent monitoring method provided by the embodiment one of the present application, as Figure 1 indicated, the method can include:

[0044] Step S101, acquiring the temperature and pressure of each preset monitoring position of the convection heating surface to form a real-time sample, using a target support vector machine model to classify all real-time samples to obtain real-time samples on and in the interval boundary, and the target support vector machine model refers to the support vector machine model after the last final classification of the current classification.

[0045] The purpose of the present application is to better monitor the ash deposition of the convection heating surface of the boiler in real time, to monitor the real-time ash deposition degree by constructing a support vector machine model, and the fixed model parameters cannot adapt to all scenes, for example, the replacement of biomass fuel, therefore, in order to improve the robustness and generalization ability of the support vector machine model, the model parameters of the support vector machine model are updated in real time.

[0046] Since the ash deposition on the convection heating surface is produced slowly with the operation of the boiler, it does not exist at the beginning, and since the ash deposition will cause the convection heating surface to reduce the heat absorption, the flow area of the working medium (such as water, steam or air) is reduced, so that the flow velocity of the working medium in the convection heating surface is reduced and the flow resistance is increased, according to the principle of fluid dynamics, the flow resistance is increased, which will inevitably cause the pressure to rise, therefore, in the present application, after the boiler runs for a period of time, the temperature data of multiple sub-regions of the convection heating surface are acquired by the thermocouple and the pressure data acquired by the pressure sensor, and the temperature data and pressure data of each sub-region are combined to form a sample, one sub-region corresponds to one sample. The sampling frequency of the temperature data and the pressure data is consistent, and the present application sets the sampling frequency to 1 minute.

[0047] It should be noted that the number of sub-regions of the convective heating surface is related to the number of sensors arranged, assuming that 10 sensors are arranged, the convective heating surface is evenly divided into 10 sub-regions, and one sub-region represents one preset monitoring position, and correspondingly, 10 samples can be obtained at each sampling time, which are used to construct an initial support vector machine model.

[0048] Specifically, the embodiment of the present application sets the sample group within 10 minutes as a training set for constructing an initial support vector machine model. The construction process of the initial support vector machine model is as follows: the theoretical temperature threshold and the theoretical pressure threshold of the convective heating surface when the known combustion material is combusted are obtained. Since one sample has two attributes of temperature and pressure, and the accumulated ash causes the convective heating surface to absorb less heat and the pressure to rise, for any sample in the training set, if the temperature in the sample is greater than or equal to the theoretical temperature threshold and the pressure is less than or equal to the theoretical pressure threshold, the sample is regarded as a normal sample, that is, the normal sample has the characteristics of high temperature and low pressure, and the samples in other cases are regarded as ash samples, which have the characteristics of low temperature and high pressure. Thus, the sample label setting of each sample in the training set is completed.

[0049] After the sample label of each sample in the known training set is set, the support vector machine model is trained using the training set to obtain a trained support vector machine model, which is recorded as an initial support vector machine model. The training of the support vector machine model belongs to the prior art, and will not be described here. It should be noted that the setting of the theoretical temperature threshold and the theoretical pressure threshold is not limited here, and can be set according to the combustion material and the combustion scene. For example, when the combustion material is straw, the temperature and pressure of the convective heating surface will be affected by various factors, such as the type of boiler, design parameters, operating conditions, etc. The following are the approximate ranges in some common situations:

[0050] (1) Temperature: low temperature range, in some small or conservative design straw combustion boilers, the temperature of the convective heating surface may be about 200-400℃, for example, some small biomass hot water boilers, whose convective heating surface is mainly used for heating hot water, have a relatively low working temperature; medium temperature range, for medium-sized straw combustion boilers, the temperature of the convective heating surface is usually between 400-600℃, which may be used to supply steam of a certain scale for industrial use or heating; high temperature range, in large, high-efficiency straw combustion power generation boilers, the temperature of the convective heating surface may reach 600-800℃ or even higher, because power generation boilers need to generate high-temperature and high-pressure steam to drive steam turbines to generate electricity, and the convective heating surface needs to absorb more heat to increase the parameters of the steam.

[0051] (2) Pressure: low pressure range, some small straw burning equipment, such as small steam generator or low pressure hot water boiler, the pressure at the convection heating surface may be about 0.1-1 MPa; medium pressure range, medium scale straw burning boiler, for industrial steam supply and other purposes, the pressure at the convection heating surface is generally between 1-4 MPa; high pressure range, in the straw burning power generation boiler, the pressure at the convection heating surface is usually higher, which may reach 4-10 MPa or even higher, to meet the requirements of power generation on steam pressure and temperature.

[0052] After obtaining the initial support vector machine model, subsequent real-time samples are continuously input according to the initial support vector machine model for real-time classification, but because different combustion materials produce different heat, when the combustion material changes, etc., the result distribution of the model classification will change as follows: because the characteristics of the new sample may have similar characteristics to a certain original category, for example, the heat of wood is about 15-20 megajoules per kilogram, and the heat of straw is about 14-17 megajoules per kilogram, because the heat of straw is slightly lower than that of wood, the normal heat may be regarded as the heat of wood when the ash accumulates, that is, different combustion materials produce different heat, so the temperature obtained by normal combustion may cause the overall temperature data to deviate, the new sample may show greater or smaller differences in the ash sample, and it may also cause the support vectors originally near the decision boundary to no longer be support vectors after the addition of the new sample, at the same time, sample changes may also introduce noise or make some abnormal samples more significant under the new sample. In the embodiment of the present application, the model parameters of the initial support vector machine model can be adjusted and updated by collecting real-time samples, for the scene of replacing biomass fuels.

[0053] In the embodiment of the present application, real-time samples are collected every 1 minute, which are used for model parameter updating at different time based on the initial support vector machine model, that is, sample classification is performed every minute, for example: when the initial support vector machine model is obtained by training at the i th sampling time, after obtaining the real-time sample at the i+1 th sampling time, the classification result of the real-time sample by the initial support vector machine model is used to update the model parameters of the initial support vector machine model, and the updated support vector machine model is used as the target support vector machine model at the i+1 th sampling time, then after obtaining the real-time sample at the i+2 th sampling time, the classification result of the real-time sample at the i+2 th sampling time by the target support vector machine model at the i+1 th sampling time is used to update the model parameters of the target support vector machine model at the i+1 th sampling time, and the updated support vector machine model is used as the target support vector machine model at the i+2 th sampling time, and the like, the support vector machine model at each sampling time is updated and adjusted for adaptive model parameters.

[0054] The specific analysis of the updating and adjusting of the model parameters at any sampling time is as follows: first, real-time samples at any sampling time of each preset monitoring position of the convective heating surface are obtained, then, the current classification of all real-time samples is performed by using the target support vector machine model to obtain real-time samples on and inside the interval boundary, wherein the support vectors include normal samples and soot samples, and the target support vector machine model refers to the support vector machine model of the last final classification (the last sampling time of any sampling time) of the current classification (any sampling time).

[0055] In step S102, it is judged whether the target support vector machine model needs model parameter updating according to the support vectors, if the target support vector machine model needs model parameter updating, then the adjustment bias judgment coefficient of the penalty parameter is obtained according to the probability that each real-time sample belongs to noise and the real-time samples on and inside the interval boundary.

[0056] Since the sample points of soot samples will be farther and farther away from the decision boundary in the classification process as the soot degree deepens, the support vectors of soot samples will probably remain unchanged, and the support vectors of normal samples may change slightly, therefore, a model parameter updating judgment mechanism is set to judge whether the target support vector machine model needs model parameter updating, and the model parameter updating judgment mechanism is as follows:

[0057] The proportion value and the number difference value of the normal samples and the soot samples in the support vectors of the current classification and the proportion value and the number difference value of the normal samples and the soot samples in the support vectors of at least three historical classifications before the current classification are obtained to form a proportion value sequence and a number difference value sequence;

[0058] First-order difference processing is performed on the proportion value sequence and the number difference value sequence respectively to obtain a proportion difference value sequence and a number difference value sequence, the mean value of all elements except the last element in the proportion difference value sequence is calculated and recorded as a proportion difference value threshold, and the mean value of all elements except the last element in the number difference value sequence is calculated and recorded as a number difference value threshold;

[0059] If the last element in the proportion difference value sequence is greater than the proportion difference value threshold and the last element in the number difference value sequence is greater than the number difference value threshold, it is determined that the target support vector machine model needs model parameter updating, otherwise, it is considered that the target support vector machine model does not need model parameter updating, and the result of the current classification is the final classification result of the real-time samples.

[0060] After determining that the target support vector machine model needs model parameter updating, considering that the penalty parameter of the support vector machine model is artificially set and the remaining model parameters are obtained according to sample training, in the embodiment of the application, the penalty parameter in the target support vector machine model is preferentially updated, and the remaining model parameters are updated and obtained according to the normal training method.

[0061] A smaller penalty parameter leads to higher fault tolerance of the model, that is, a wider buffer zone (interval boundary), and a larger penalty parameter leads to lower fault tolerance of the model, that is, a narrower buffer zone. When the sample characteristics change, the sample distribution needs to be analyzed, especially the historical samples and real-time samples in the interval boundary before the penalty parameter is updated, and an adjustment bias judgment coefficient of the current classification corresponding penalty parameter is obtained.

[0062] Because too many samples are added each time (that is, real-time samples) for model classification, the calculation load is easily increased, and therefore, after each classification, the influence coefficient of each sample and the decision boundary needs to be calculated, and only the samples with a large influence coefficient are retained in the model. Therefore, in the embodiment of the application, for any historical sample in any historical classification, a straight-line distance of the any historical sample to the decision boundary of the any historical classification is obtained, and the reciprocal of the straight-line distance is taken as the influence coefficient of the any historical sample; the influence coefficient of each support vector in the any historical classification is obtained, the minimum influence coefficient is taken as an influence coefficient threshold, and if the influence coefficient of any historical sample is greater than or equal to the influence coefficient threshold, the any historical sample is taken as a target historical sample, that is, a sample with a large influence coefficient.

[0063] Because there may be noise in the real-time samples, the noise is caused by electromagnetic equipment in the industrial environment, which easily causes electromagnetic interference to the sensor, and at the same time, the stability of the sensor is affected by the mechanical vibration or impact on the convection heating surface during the operation of the boiler, leading to changes or interference of the samples and further forming noise. For the noise samples, a smaller penalty parameter can avoid overfitting, and therefore, based on the characteristics of the noise, such as position abnormality, small quantity, and strong randomness, whether there is a noise sample in the real-time samples is analyzed, and the method for obtaining the probability that each real-time sample belongs to noise is as follows:

[0064] The target historical samples in the three historical classifications before the current classification are taken as references, density clustering is performed on all target historical samples and all real-time samples, at least one cluster center is obtained, for any real-time sample, the minimum Euclidean distance between the any real-time sample and each cluster center is obtained according to the Euclidean distance between the any real-time sample and each cluster center A circular region is constructed with the any real-time sample as the center and the minimum Euclidean distance multiplied by a preset multiple as the radius, and the number of samples in the circular region is counted.

[0065] obtaining a difference between the reciprocal of the minimum Euclidean distance and a constant 1, obtaining a reciprocal of an addition value between the sample quantity and a constant 1, and performing a weighted summation on the difference and the reciprocal of the addition value to obtain the probability that any real-time sample belongs to noise.

[0066] In an embodiment, since the cluster centers near the samples are prevented from causing interference, the radius of the circle cannot be the complete minimum Euclidean distance, and the distance cannot be too close considering that the density of the cluster is not necessarily too dense, so a preset multiple is set as , that is, the radius of the circle is , and the calculation formula of the probability that any real-time sample belongs to noise is

[0067]

[0068] wherein, represents the probability that any real-time sample belongs to noise, represents a first weight, 1 represents a constant, represents the minimum Euclidean distance, represents a second weight, represents the sample quantity in the circular region.

[0069] It should be noted that the greater the distance between any real-time sample and the nearest cluster center, that is, the greater the minimum Euclidean distance, the greater the probability that any real-time sample is an isolated point, and the greater the possibility of position anomaly; if the sample quantity in the circular region of any real-time sample is smaller, it means that there are fewer similar samples around the real-time sample, and the greater the probability that any real-time sample is noise.

[0070] Preferably, since the cluster size is variable, that is, it is possible to select some cluster members with high density in the circular region, so , which is not limited here.

[0071] Further, according to experimental statistics, the probability threshold is set to 0.7, if the probability that any real-time sample belongs to noise is greater than or equal to 0.7, the real-time sample is marked as a noise sample, and then all noise samples in the real-time samples can be obtained. After obtaining the marked noise samples, the real-time samples on and in the interval boundary of the current classification are analyzed to obtain the adjustment bias judgment coefficient of the penalty parameter, and the specific analysis is as follows:

[0072] For any real-time sample on and within the interval boundary, the Euclidean distance between the any real-time sample and each real-time sample on and within the interval boundary is obtained, to obtain the Euclidean distance mean corresponding to the any real-time sample; according to the Euclidean distance mean corresponding to each real-time sample on and within the interval boundary, the average value of the Euclidean distance mean is calculated, and the difference between the constant 1 and the reciprocal of the average value is recorded as the sample dispersion degree;

[0073] The number of noise samples in the real-time samples on and within the interval boundary is counted, the number ratio of the number of noise samples in the total number of real-time samples on and within the interval boundary is calculated, and the adjustment bias determination coefficient of the penalty parameter is obtained according to the number ratio and the mean of the sample dispersion degree.

[0074] In an embodiment, the calculation formula of the adjustment bias determination coefficient of the penalty parameter is:

[0075]

[0076] wherein, represents the adjustment bias determination coefficient of the penalty parameter, represents the total number of real-time samples on and within the interval boundary, represents the number of noise samples in the real-time samples on and within the interval boundary, represents the Euclidean distance between the jth real-time sample on and within the interval boundary and the ith real-time sample, and 1 represents a constant.

[0077] It should be noted that, represents the number ratio, the larger the number ratio, the larger the noise sample, and the larger the adjustment bias of the penalty parameter, the larger the adjustment bias determination coefficient; is used to represent the dispersion degree of the real-time sample, the larger the dispersion degree, the larger the adjustment bias of the penalty parameter, the larger the adjustment bias determination coefficient, and the smaller the penalty parameter.

[0078] In step S103, according to the position difference between the sample closest to the decision boundary within the interval boundary of the current classification and the sample closest to the decision boundary within the interval boundary of the last final classification, the adjustment coefficient of the penalty parameter is obtained, and the new penalty parameter is obtained according to the adjustment bias determination coefficient and the adjustment coefficient.

[0079] Since the different categories of samples should be separated as much as possible during classification, the adjustment of the penalty parameter can be performed according to the distribution difference between real-time samples. In the current classification interval boundary, the upper sample and the lower sample closest to the decision boundary in the upper interval and the lower interval are calculated, and the minimum Euclidean distance between the upper sample and the lower sample is obtained.

[0080] According to the above method for obtaining the minimum Euclidean distance corresponding to the interval boundary of the current classification, the minimum Euclidean distance between the upper sample and the lower sample closest to the decision boundary in the interval boundary of the last final classification is obtained, which is denoted as the minimum reference Euclidean distance. The reciprocal of the absolute value of the difference between the minimum reference Euclidean distance and the minimum Euclidean distance is calculated, and the difference between the constant 1 and the reciprocal is used as the adjustment coefficient of the penalty parameter.

[0081] In an embodiment, the calculation formula of the adjustment coefficient of the penalty parameter is:

[0082]

[0083] wherein, denotes the adjustment coefficient of the penalty parameter, denotes the minimum Euclidean distance corresponding to the interval boundary of the current classification, denotes the minimum reference Euclidean distance corresponding to the interval boundary of the last final classification, || denotes the absolute value symbol, and 1 denotes a constant.

[0084] It should be noted that, denotes the minimum Euclidean distance corresponding to the interval boundary of the current classification, denotes the minimum Euclidean distance corresponding to the interval boundary before the real-time sample is added. If the difference between the minimum Euclidean distances before and after the real-time sample is added is large, it means that the value of the penalty parameter adjustment is large, and the adjustment coefficient is larger.

[0085] Further, after obtaining the adjustment bias determination coefficient and the adjustment coefficient of the penalty parameter, a new penalty parameter can be obtained according to the adjustment bias determination coefficient and the adjustment coefficient, which is used to replace the penalty parameter in the target support vector machine model. The method for obtaining the new penalty parameter is:

[0086] First, the penalty parameter of the target support vector machine model is obtained, and the product of the penalty parameter and the adjustment coefficient is obtained, which is denoted as the adjustment value. Then, according to the calculation formula of the adjustment bias determination coefficient, the adjustment bias determination coefficient threshold is set according to the rounding principle. The specific setting method is as follows: the vertical distance between the interval boundary of the current classification and the decision boundary is obtained, the difference between the constant 1 and the reciprocal of the vertical distance is calculated, and the sum of the preset value and the difference is denoted as the adjustment bias determination coefficient threshold. Preferably, the adjustment bias determination coefficient threshold is wherein, represents a vertical distance.

[0087] Finally, if the adjustment bias determination coefficient is greater than or equal to an adjustment bias determination coefficient threshold, a difference between the penalty parameter of the target support vector machine model and the adjustment value is taken as a new penalty parameter; if the adjustment bias determination coefficient is less than the adjustment bias determination coefficient threshold, a sum of the penalty parameter of the target support vector machine model and the adjustment value is taken as the new penalty parameter.

[0088] In an embodiment, a calculation formula of the new penalty parameter is:

[0089]

[0090] wherein, represents a new penalty parameter, represents a penalty parameter of a target support vector machine model, represents an adjustment coefficient of the penalty parameter, represents an adjustment bias determination coefficient of the penalty parameter.

[0091] In step S104, model parameter updating is performed on the target support vector machine model according to all real-time samples and the new penalty parameter, to obtain an updated support vector machine model. The updated support vector machine model is used to perform final classification on all real-time samples, to obtain a final classification result. The final classification result is used to monitor the ash deposition state on the convective heating surface.

[0092] After the new penalty parameter is obtained, model parameter updating is performed on the target support vector machine model according to all real-time samples and the new penalty parameter. Since the features of real-time samples can change, the classification standard of the target support vector machine model can not be applicable to real-time samples. However, there is a certain inherent difference between real-time samples. Therefore, clustering is performed on all real-time samples, to obtain at least one cluster. For any cluster, a distance between a cluster center of the any cluster and each target historical sample is calculated. A target historical sample corresponding to a minimum distance is recorded as a reference sample. According to a classification result of the reference sample, a sample classification label of each real-time sample in the any cluster is obtained.

[0093] After obtaining the sample classification label of each real-time sample and replacing the penalty parameter in the target support vector machine model with the new penalty parameter, other model parameters in the target support vector machine model are updated based on each real-time sample and the corresponding sample classification label, to obtain an updated support vector machine model, that is, each real-time sample and the corresponding sample classification label are taken as a training set to retrain the target support vector machine model, to obtain a new support vector machine model, at this time, the model parameters in the new support vector machine model are the updated model parameters, and the new support vector machine model is the updated support vector machine model.

[0094] After obtaining the updated support vector machine model, all real-time samples are finally classified by using the updated support vector machine model, to obtain a final classification result, and the final classification result includes normal samples and ash deposition samples. Then, the ash deposition state on the convective heating surface is monitored according to the final classification result, and the specific method is as follows:

[0095] The distance between each ash deposition sample and the decision boundary of the final classification is calculated, to obtain a distance mean, and the reciprocal of the distance mean and the difference between the constant 1 are used to obtain a real-time ash deposition degree; the number of preset monitoring positions involved in the ash deposition sample is obtained, to obtain a proportion of the number in the total number of preset monitoring positions, which is recorded as an ash deposition range.

[0096] If the ash deposition range is less than a preset ash deposition range threshold, and the real-time ash deposition degree is greater than or equal to a preset ash deposition degree threshold, only a local soot blowing operation is performed on the area where the preset monitoring position involved in the ash deposition sample is located;

[0097] If the ash deposition range is greater than or equal to a preset ash deposition range threshold, and the real-time ash deposition degree is greater than or equal to a preset ash deposition degree threshold, a whole soot blowing operation is performed on the convective heating surface.

[0098] In an embodiment, the calculation formula of the real-time ash deposition degree is as follows:

[0099]

[0100] wherein, the real-time ash deposition degree is represented by d, the distance mean is represented by d.

[0101] It should be noted that the greater the distance mean is, the farther the ash deposition sample is from the interval boundary, which indicates that the ash deposition is more, that is, the ash deposition degree is greater.

[0102] The calculation formula of the ash deposition range is as follows:

[0103]

[0104] wherein, represents the dusting range, that is, the proportion of the number of preset monitoring positions involved by the dusting sample in the total number of preset monitoring positions, represents the total number of preset monitoring positions, represents the number of preset monitoring positions involved by the dusting sample.

[0105] It should be noted that the more the number of preset monitoring positions involved by the dusting sample, the greater the dusting range.

[0106] Since the value range of the real-time dusting degree and the dusting range is [0, 1], the median is taken as the demarcation line of the local blowing operation and the overall blowing operation, and the dusting range threshold and the dusting degree threshold are set as , which is not limited here.

[0107] If , it indicates that the dusting may be local dusting serious, and the sub-regions at preset monitoring positions need to be blown (that is, blowing operation in one region of the boiler) to realize local blowing operation, if , double blowing operation (that is, blowing operation on the whole boiler) is performed to realize overall blowing operation, and the remaining cases select observation without blowing operation.

[0108] The above embodiments are only used to illustrate the technical solutions of the present application, but not to limit it; although the present application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that the technical solutions recorded in the foregoing embodiments can still be modified, or some technical features can be replaced by equivalents; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present application, and should be included in the protection scope of the present application.

Claims

1. A method for intelligent monitoring of the convection heating surfaces in a biomass spread combustion process, characterized in that, The method comprises: acquiring temperature and pressure of each preset monitoring position of the convective heating surface to form a real-time sample, using a target support vector machine model to perform current classification on all real-time samples to obtain support vectors, real-time samples on and within a margin boundary, and the target support vector machine model is a support vector machine model after last final classification of the current classification; judging whether the target support vector machine model needs model parameter updating according to the support vectors, if the target support vector machine model needs model parameter updating, acquiring an adjustment bias judgment coefficient of a penalty parameter according to a probability of each real-time sample belonging to noise and the real-time samples on and within the margin boundary; acquiring an adjustment coefficient of the penalty parameter according to a position difference between a sample closest to a decision boundary within the margin boundary of the current classification and a sample closest to the decision boundary within the margin boundary of the last final classification, and acquiring a new penalty parameter according to the adjustment bias judgment coefficient and the adjustment coefficient; performing model parameter updating on the target support vector machine model according to all real-time samples and the new penalty parameter to obtain an updated support vector machine model, performing final classification on all real-time samples using the updated support vector machine model to obtain a final classification result, and monitoring the ash deposition state on the convective heating surface according to the final classification result; if the support vectors comprise normal samples and ash deposition samples, the judging whether the target support vector machine model needs model parameter updating according to the support vectors comprises: acquiring a proportion value and a quantity difference value of normal samples and ash deposition samples in the support vectors of the current classification and a proportion value and a quantity difference value of normal samples and ash deposition samples in the support vectors of a preset number of historical classifications before the current classification to form a proportion value sequence and a quantity difference value sequence; performing first-order difference processing on the proportion value sequence and the quantity difference value sequence respectively to obtain a proportion difference value sequence and a quantity difference value sequence, calculating a mean value of all elements except the last element in the proportion difference value sequence as a proportion difference value threshold, and calculating a mean value of all elements except the last element in the quantity difference value sequence as a quantity difference value threshold; if the last element in the proportion difference value sequence is greater than the proportion difference value threshold and the last element in the quantity difference value sequence is greater than the quantity difference value threshold, it is determined that the target support vector machine model needs model parameter updating; the acquiring an adjustment bias judgment coefficient of a penalty parameter according to a probability of each real-time sample belonging to noise and the real-time samples on and within the margin boundary comprises: acquiring the probability of each real-time sample belonging to noise based on historical samples at the time of the preset number of historical classifications, marking noise samples according to the probability of each real-time sample belonging to noise, and counting the number of noise samples in the real-time samples on and within the margin boundary. For any real-time sample on and within the interval boundary, the Euclidean distance between the any real-time sample and each real-time sample on and within the interval boundary is obtained, to obtain the mean of the Euclidean distance corresponding to the any real-time sample; according to the mean of the Euclidean distance corresponding to each real-time sample on and within the interval boundary, the average of the mean of the Euclidean distance is calculated, and the difference between the constant 1 and the reciprocal of the average is recorded as the sample dispersion degree; The number proportion of the number of noise samples in the total number of real-time samples on and within the interval boundary is calculated, and the adjustment bias determination coefficient of the penalty parameter is obtained according to the number proportion and the mean of the sample dispersion degree.

2. The method of claim 1, wherein the biomass spread combustion process is a fluidized bed combustion process. The probability that each real-time sample belongs to noise is obtained based on the historical samples in the preset number of historical classifications, including: For any historical sample in any historical classification, the straight line distance from the any historical sample to the decision boundary of the any historical classification is obtained, and the reciprocal of the straight line distance is taken as the influence coefficient of the any historical sample; the influence coefficient of each support vector in the any historical classification is obtained, and the minimum influence coefficient is taken as the influence coefficient threshold; if the influence coefficient of any historical sample is greater than or equal to the influence coefficient threshold, the any historical sample is taken as a target historical sample; The target historical samples at the preset number of historical classifications are obtained, and density clustering is performed on all target historical samples and all real-time samples to obtain at least one cluster center; for any real-time sample, the minimum Euclidean distance between the any real-time sample and each cluster center is obtained, and a circular region is constructed with the any real-time sample as the center and the minimum Euclidean distance multiplied by a preset multiple as the radius; the number of samples in the circular region is counted. The difference between the reciprocal of the minimum Euclidean distance and the constant 1 is obtained, the reciprocal of the sum of the number of samples and the constant 1 is obtained, and the weighted sum of the difference and the reciprocal of the sum is obtained to obtain the probability that the any real-time sample belongs to noise.

3. The method of claim 1, wherein the biomass spread combustion process is a fluidized bed combustion process. The adjustment coefficient of the penalty parameter is obtained according to the position difference between the sample closest to the decision boundary within the interval boundary of the current classification and the sample closest to the decision boundary within the interval boundary of the last final classification, including: In the interval boundary of the current classification, the sample closest to the decision boundary within the upper interval and the sample closest to the decision boundary within the lower interval are calculated, which are recorded as the upper sample and the lower sample, respectively; the minimum Euclidean distance between the upper sample and the lower sample is obtained; The upper sample and the lower sample closest to the decision boundary within the interval boundary of the last final classification are obtained to obtain the minimum reference Euclidean distance, and the reciprocal of the absolute value of the difference between the minimum reference Euclidean distance and the minimum Euclidean distance is calculated; the difference between the constant 1 and the reciprocal is taken as the adjustment coefficient of the penalty parameter.

4. The method of claim 1, wherein the biomass spread combustion process is a fluidized bed combustion process. The new penalty parameter is obtained according to the adjustment bias determination coefficient and the adjustment coefficient, including: Obtaining the penalty parameter of the target support vector machine model, obtaining the product of the penalty parameter and the adjustment coefficient, denoted as an adjustment value; obtaining the vertical distance between the interval boundary and the decision boundary of the current classification, calculating the difference between the constant 1 and the reciprocal of the vertical distance, and denoting the sum of the preset value and the difference as an adjustment bias determination coefficient threshold; If the adjustment bias determination coefficient is greater than or equal to the adjustment bias determination coefficient threshold, the difference between the penalty parameter of the target support vector machine model and the adjustment value is taken as a new penalty parameter; if the adjustment bias determination coefficient is less than the adjustment bias determination coefficient threshold, the sum of the penalty parameter of the target support vector machine model and the adjustment value is taken as a new penalty parameter.

5. The method as claimed in claim 2, wherein the method is characterized by: The model parameter updating of the target support vector machine model according to all real-time samples and the new penalty parameter to obtain an updated support vector machine model, comprising: Clustering all real-time samples to obtain at least one cluster, for any cluster, calculating the distance between the cluster center of the any cluster and each target historical sample, obtaining the target historical sample corresponding to the minimum distance as a reference sample, and obtaining the sample classification label of each real-time sample in the any cluster according to the classification result of the reference sample; Obtaining the sample classification label of each real-time sample, replacing the penalty parameter in the target support vector machine model with the new penalty parameter, and updating other model parameters in the target support vector machine model based on each real-time sample and the corresponding sample classification label to obtain an updated support vector machine model.

6. The method of claim 1, wherein the biomass spread combustion process is a fluidized bed combustion process. The monitoring of the ash deposition state on the convective heating surface according to the final classification result, comprising: The final classification result includes normal samples and ash deposition samples, calculating the distance between each ash deposition sample and the decision boundary of the final classification to obtain a distance mean, obtaining the real-time ash deposition degree according to the difference between the reciprocal of the distance mean and the constant 1; obtaining the number of preset monitoring positions involved in the ash deposition sample to obtain the proportion of the number in the total number of preset monitoring positions, denoted as an ash deposition range; If the ash deposition range is less than a preset ash deposition range threshold, and the real-time ash deposition degree is greater than or equal to a preset ash deposition degree threshold, only a local soot blowing operation is performed on the area where the preset monitoring position involved in the ash deposition sample is located; If the ash deposition range is greater than or equal to a preset ash deposition range threshold, and the real-time ash deposition degree is greater than or equal to a preset ash deposition degree threshold, a whole soot blowing operation is performed on the convective heating surface.

Images