Coal ash component real-time monitoring method based on Internet of Things intelligent sensor

Abstract

The invention relates to the technical field of coal ash monitoring, and discloses a coal ash component real-time monitoring method based on an Internet of Things intelligent sensor. The method comprises the following steps: acquiring a plurality of coal ash component index data acquired by an intelligent sensor so as to realize real-time component monitoring. In the parameter optimization stage, obtaining a target feature vector of the current stage and a preset number of reference parameters, and in the optimization stage, dividing data into a data set corresponding to the feature vector according to the similarity between component index data and the feature vector in each stage; determining candidate feature vectors according to the correlation degree of the current data set of the target feature vector and other feature vector reference data sets; calculating a weight coefficient of each candidate feature vector, wherein the coefficient comprehensively considers the number of overlapped data points, a preset number, the total amount of a current data set, a similarity difference value between the data points and the feature vectors, a fluctuation amplitude of the feature vectors in a reference stage and a preset basic weight; and updating and optimizing the candidate feature vector and the target feature vector according to the weight coefficient.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of coal ash monitoring, in particular to a coal ash component real-time monitoring method based on an Internet of Things intelligent sensor. BACKGROUND

[0002] In industrial production, coal ash, as the main residue after coal combustion, its composition directly relates to the combustion efficiency, equipment wear and tear, and environmental emissions, etc. The traditional coal ash composition monitoring method relies on offline sampling analysis, that is, after manually collecting coal ash samples, the composition data is obtained by chemical analysis or instrument detection in the laboratory. This method not only needs to consume a lot of manpower and time cost, but also has obvious time lag, which is difficult to meet the needs of real-time production control.

[0003] With the improvement of industrial automation level, online monitoring technology is gradually applied to the field of coal ash composition analysis. Early online monitoring equipment mostly uses single sensor for data collection. Due to the complexity of coal ash composition and its susceptibility to temperature, humidity, air flow and other environmental factors, the measurement accuracy of single sensor is often difficult to guarantee, and there are often large fluctuations in data. In order to improve the monitoring accuracy, some systems try to use multi-sensor data fusion technology, but the existing fusion algorithms are mostly based on fixed parameter settings, which cannot be dynamically adjusted according to real-time data characteristics. When the composition of coal ash changes suddenly or the monitoring environment is disturbed, these algorithms are prone to fusion deviation, resulting in distorted monitoring results. In the data processing process, the traditional method usually regards all sensor data as equally important, lacking differentiated processing of different data points. For those data points with low correlation with the target characteristics, if they are not distinguished and included in the calculation, it will increase the redundancy of data processing and reduce the monitoring efficiency. At the same time, the existing technology often only relies on the data information of the current stage when updating the feature vector, ignoring the rules contained in the historical data, making the adaptability of the feature vector weak and difficult to cope with complex changes in the long-term monitoring process. In the industrial production site, the real-time monitoring data of coal ash composition needs to be fed back to the control system in time to realize the dynamic optimization of the combustion process. However, due to the shortcomings of the existing monitoring method in timeliness and accuracy, the production control often lags behind the actual working condition changes, which not only affects the energy utilization efficiency, but also may cause equipment corrosion, slagging and other problems due to excessive coal ash composition, increasing the production and maintenance cost. At the same time, inaccurate monitoring data will also interfere with environmental assessment, making the monitoring of environmental indicators lose reliable basis. Although the development of Internet of Things technology and intelligent sensors provides new possibilities for real-time monitoring of coal ash composition, how to effectively integrate these technologies and build a monitoring method that can adaptively adjust parameters and dynamically optimize feature vectors is still a problem to be solved. In the prior art, the parameter setting of the data fusion algorithm lacks flexibility, and the updating mechanism of the feature vector fails to fully utilize the correlation information between historical data and real-time data, resulting in that the monitoring system shows strong limitations when facing complex working conditions. The existence of these problems makes it difficult to achieve the precision and efficiency required by industrial production in real-time monitoring of coal ash composition. SUMMARY

[0004] The purpose of the present application is to provide a real-time monitoring method for coal ash composition based on Internet of Things intelligent sensors to solve the problems raised in the background art.

[0005] To achieve the above-mentioned purpose, the present application provides a real-time monitoring method for coal ash composition based on Internet of Things intelligent sensors, which comprises: obtaining a plurality of coal ash composition index data collected by intelligent sensors; using the composition index data to complete the parameter iterative optimization of the adaptive weighted fusion algorithm, and realizing real-time monitoring of the composition; in each parameter optimization stage, obtaining the target feature vector of the current parameter optimization stage, obtaining the previous preset number of parameter optimization stages, denoted as reference parameter optimization stages, and dividing the composition index data into the data set of the corresponding feature vector according to the similarity of the composition index data and the feature vector in each parameter optimization stage; obtaining the candidate feature vector according to the correlation degree between the data set of the current parameter optimization stage of the target feature vector and the data set of the reference optimization stage of the other feature vectors; calculating the weight coefficient of each candidate feature vector, which is determined by the number of overlapping data points of each candidate feature vector in the data set of the reference optimization stage and the current data set of the target feature vector, the preset number, the total amount of data points in the current data set, the difference in similarity of each data point and the target and candidate feature vectors, the fluctuation amplitude of the feature vector in the reference stage, and the preset basic weight; updating and optimizing the candidate feature vector and the target feature vector according to the weight coefficient.

[0006] Preferably, the dividing of the composition index data into the data set of the corresponding feature vector according to the similarity of the composition index data and the feature vector in each parameter optimization stage comprises: in each parameter optimization stage, obtaining the similarity of each composition index data and each feature vector, dividing each composition index data into the data set of the feature vector with the highest similarity, and obtaining the data set of each feature vector in each parameter optimization stage.

[0007] Preferably, the degree of association of the data set of the current parameter optimization stage of the target feature vector with the data set of the reference optimization stage of the other feature vector obtains the candidate feature vector, comprising: if the data set of the current parameter optimization stage of the target feature vector and the data set of the reference optimization stage of the other feature vector exist overlapping data points, the other feature vector is recorded as a suspected candidate feature vector; the data points existing in the data set of the current parameter optimization stage of the target feature vector and the data set of the reference optimization stage of the suspected candidate feature vector are recorded as fluctuation data points; the deviation degree of the fluctuation data points is calculated; the fluctuation data points with the deviation degree less than a preset deviation threshold are recorded as effective fluctuation data points; the suspected candidate feature vector with the effective fluctuation data points existing in the data set of the reference optimization stage is recorded as the candidate feature vector.

[0008] Preferably, the deviation degree of the fluctuation data points is calculated, comprising: obtaining the neighborhood comparison data points of the fluctuation data points; recording the average of the similarity difference between the fluctuation data points and the neighborhood comparison data points as a metric difference, and judging the normalized value of the ratio of the metric difference of the fluctuation data points to the average of the metric differences of all component indicator data as the deviation degree of the fluctuation data points.

[0009] Preferably, the neighborhood comparison data points of the fluctuation data points are obtained, comprising: obtaining the area in the preset range of the fluctuation data points as a comparison neighborhood, and recording the component indicator data in the comparison neighborhood as neighborhood reference data points; evenly dividing the comparison neighborhood into a preset number of directional areas with the fluctuation data points as the center; recording the average of the similarity difference between the neighborhood reference data points and the fluctuation data points in each directional area as an area difference; and recording the neighborhood reference data points in the area with the area difference less than the average of the differences of all directional areas as the neighborhood comparison data points.

[0010] Preferably, the fluctuation amplitude obtaining method comprises: obtaining the feature values of each feature vector in each parameter reference stage, calculating the difference degree of the feature values of each feature vector in two reference parameter optimization stages, and recording the variance of the feature value difference degrees of each feature vector in all two reference parameter optimization stages as the fluctuation amplitude of the feature vector in all reference parameter optimization stages.

[0011] Preferably, the component real-time monitoring comprises: inputting the newly collected coal ash component index data into the self-adaptive weighted fusion algorithm with optimized parameters to obtain the component category to which the newly collected component index data belongs; obtaining the lower quartile of the similarity between each component index data and the feature center of the component category to which the component index data belongs; obtaining the similarity between the newly collected component index data and the feature center of the component category to which the component index data belongs, and determining that the newly collected component index data is an abnormal data point if the similarity between the newly collected component index data and the feature center of the component category to which the component index data belongs is less than the lower quartile; and determining that the coal ash component monitoring is abnormal if the number of continuously collected abnormal data points is greater than a preset number threshold.

[0012] Preferably, after obtaining the plurality of coal ash component index data collected by the intelligent sensor, the method further comprises a pre-processing step of the component index data, and the pre-processing comprises: removing outliers in the data, interpolating and filling missing data, and standardizing the data to a preset range.

[0013] Preferably, the stop condition of the parameter iterative optimization is that, in a continuous preset number of parameter optimization stages, the change rate of the data set of the target feature vector is less than a preset change threshold, or the total number of parameter optimizations reaches a preset maximum number.

[0014] Preferably, the component real-time monitoring process further comprises a dynamic sampling adjustment step, and the dynamic sampling adjustment step comprises: counting the frequency of abnormal data points in a unit time, increasing the sampling interval density of the intelligent sensor when the frequency is higher than a preset frequency threshold, decreasing the sampling interval density of the intelligent sensor when the frequency is lower than the preset frequency threshold, and the adjustment amplitude of the sampling interval density is positively correlated with the degree of deviation of the frequency from the threshold.

[0015] Compared with the prior art, the method has the following beneficial effects: The coal ash component real-time monitoring method based on the Internet of Things intelligent sensor realizes dynamic processing and optimization of the coal ash component index data by introducing the self-adaptive weighted fusion algorithm. Compared with the traditional offline monitoring method, the method realizes real-time collection of data by means of the Internet of Things intelligent sensor, breaks the dependence on manual sampling and laboratory analysis, can obtain the change information of the coal ash component in the first time, makes the monitoring process keep synchronous with the industrial production rhythm, and avoids the production control delay caused by data lag. In the data processing link, the method realizes accurate classification of different data characteristics by dividing the component index data into the corresponding feature vector data set. This similarity-based division method can effectively distinguish data points with different degrees of association with the target feature, reduce the interference of redundant information on the calculation process, and make the data processing more targeted. At the same time, in determining the candidate feature vector, by analyzing the correlation degree of the current data set of the target feature vector and the reference data set of other feature vectors, the internal relationship between historical data and real-time data is fully tapped, and the selection of the candidate feature vector is more in line with the actual needs of the current monitoring working condition. The calculation method of the weight coefficient comprehensively considers various factors, including the number of coincident data points, the preset number, the total amount of data points, the similarity difference, the fluctuation amplitude, and the preset basic weight. This multi-dimensional consideration makes the weight distribution more reasonable. Compared with the traditional fixed weight fusion algorithm, the dynamically adjusted weight coefficient can better adapt to the changes in data characteristics. When the reliability of some data points decreases or environmental interference increases, the weight coefficient will be adjusted accordingly, thereby reducing the influence of abnormal data on the final monitoring result. The updating and optimization mechanism of the feature vector combines the information of the candidate feature vector and the target feature vector, and realizes the adaptive evolution of the feature vector through the dynamic adjustment of the weight coefficient. This mechanism not only responds to the mutation of coal ash composition in a timely manner, but also retains effective features in the long-term monitoring process through learning from historical reference stage data, making the adaptability of the monitoring model stronger. In the face of fluctuations in environmental factors such as temperature and humidity, the optimized feature vector can maintain stable monitoring accuracy, reducing errors caused by external interference. From the perspective of industrial application, this method can provide more accurate basis for real-time regulation of the combustion process. By timely grasping the changes in coal ash composition, the combustion parameters can be dynamically adjusted to reduce energy waste. At the same time, accurate composition monitoring can provide early warning of potential corrosion and slagging risks for equipment, reducing equipment maintenance frequency and cost. In addition, due to the improvement of real-time and accuracy of monitoring data, the monitoring of environmental emission indicators is also more timely, which helps to take measures to control pollutant emissions in a timely manner, meeting the environmental protection requirements of industrial production. BRIEF DESCRIPTION OF DRAWINGS

[0016] Figure 1 The working principle diagram of the coal ash composition real-time monitoring method based on the Internet of Things intelligent sensor described in the present application; Figure 2 The working principle diagram for generating a candidate feature vector; Figure 3 The flowchart for calculating the deviation degree of fluctuating data points; Figure 4 The flowchart for calculating the fluctuation amplitude. DETAILED DESCRIPTION

[0017] The technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, but not all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by a person of ordinary skill in the art without creative work fall within the protection scope of the present application.

[0018] Please refer to Figures 1-4 The present application provides a coal ash composition real-time monitoring method based on an Internet of Things intelligent sensor, which comprises the following steps: Obtaining a plurality of coal ash composition index data collected by an intelligent sensor: deploying an Internet of Things intelligent sensor in a coal ash monitoring area, and collecting a plurality of composition index data in coal ash in real time, such as ash content, sulfur content, moisture content, etc., to form an initial data set.

[0019] Using the composition index data to complete parameter iterative optimization of an adaptive weighted fusion algorithm, and realizing real-time composition monitoring: in the parameter iterative optimization process, the algorithm parameters are adjusted constantly through multiple parameter optimization stages, so that the monitoring result gradually becomes accurate. In each parameter optimization stage, the following operations are performed: Obtaining a target feature vector of the current parameter optimization stage, which is a data set reflecting the main features of the coal ash composition in the current stage.

[0020] Determining a preset number of previous parameter optimization stages as reference parameter optimization stages, and the number of reference stages can be set according to actual monitoring requirements, for example, 5 times.

[0021] Dividing the composition index data into the data set of the corresponding feature vector according to the similarity of the composition index data and the feature vector in each parameter optimization stage: calculating the similarity of each composition index data and each feature vector, and allocating the data to the data set of the most matched feature vector according to the similarity, so that each feature vector has a corresponding data set in different parameter optimization stages.

[0022] Obtaining a candidate feature vector according to the correlation degree between the data set of the current parameter optimization stage of the target feature vector and the data set of the reference optimization stage of other feature vectors: analyzing the coincidence degree and data fluctuation between the current data set and other data sets in the reference stage, and screening the feature vectors with correlation as candidates.

[0023] Calculate the weight coefficient of each candidate feature vector: The calculation of the weight coefficient needs to take into account multiple factors, including the number of overlapping data points between the dataset of each candidate feature vector and the current dataset of the target feature vector in the reference optimization stage, the preset number, the total number of data points in the current dataset, the similarity difference between each data point and the target and candidate feature vectors, the fluctuation range of the feature vector in the reference stage, and the preset basic weight. The final weight is determined by the combination of these parameters.

[0024] The candidate feature vector and the target feature vector are updated and optimized based on the weight coefficients: the weight coefficients are applied to the candidate feature vector and the target feature vector to adjust the data distribution and feature ratio in the vector, thus completing this round of parameter optimization.

[0025] Through multiple rounds of parameter optimization until the stopping conditions are met, real-time monitoring of coal ash composition is finally achieved.

[0026] Example 1: In each parameter optimization stage, when classifying the component index data into the corresponding feature vector dataset based on the similarity between each component index data and the feature vector, a series of specific operations are required. First, in each parameter optimization stage, the system calls a preset similarity calculation mechanism to calculate the similarity between each component index data in that stage and all feature vectors existing in that stage. Here, the component index data covers various quantifiable component information in coal ash collected by intelligent sensors, such as the silica content, alumina content, and iron oxide content in the ash. The feature vector, on the other hand, is a dataset extracted from historical data and partial data from the current stage that reflects the characteristics of a certain type of component.

[0027] Similarity can be calculated in various ways. For example, it can be determined by calculating the distance between data points and feature vectors in multidimensional space; the closer the distance, the higher the similarity. Alternatively, it can be measured by the cosine of the angle between the vectors; the closer the cosine is to 1, the more consistent the directions are, and the higher the similarity. After obtaining the similarity values ​​between each component index data and each feature vector, the system compares these values ​​one by one, finds the maximum similarity value for each component index data, and assigns that component index data to the dataset corresponding to the feature vector that generated this maximum similarity.

[0028] In this way, each feature vector forms an independent dataset at different parameter optimization stages. All component index data within this dataset represent the features vector with the highest similarity at that stage. This partitioning method ensures strong internal consistency within each dataset, facilitating more precise analysis and optimization of the feature vectors later on.

[0029] After obtaining the several coal ash composition index data collected by the intelligent sensor, data preprocessing is an essential link. The first step of preprocessing is to remove outliers in the data. Due to the external environmental interference during the collection process of the intelligent sensor, such as temperature fluctuation, dust adhesion, temporary equipment failure, etc., part of the collected data may deviate from the normal range. At this time, specific identification methods are needed to identify these outliers, such as observing whether the data is outside the reasonable interval determined according to historical data, or by analyzing the distribution characteristics of the data, identifying data points far from the majority of data in the data set as outliers. Once the outliers are identified, they will be removed from the original data set to avoid interference of these unreasonable data on subsequent parameter optimization and monitoring results.

[0030] The second step of preprocessing is to interpolate and fill in the missing data. During data collection, due to temporary sensor offline, data transmission interruption, etc., some data may be missing in some time periods or for some indicators, resulting in gaps in the data set. For these missing data, appropriate methods need to be used for filling. For example, when the data before and after the missing data changes smoothly, linear interpolation can be used to calculate the estimated value of the missing position according to the change trend of the known data before and after the missing point; when the data shows a certain periodicity or correlation, the nearest known data near the missing point can be directly used as the filling value; if the data is large and has obvious distribution rules, statistical interpolation methods can also be used, such as filling according to the average value, median, etc. of the index data. Through these ways, the data set can maintain integrity, ensuring the continuity and usability of the data in the subsequent processing process.

[0031] The third step of preprocessing is to standardize the data to a pre-set range. Different composition index data often have different dimensions and value ranges, for example, some index values may be between 0-10, while some index values may be between 100-1000. Such differences in dimensions and ranges will affect the similarity calculation and the weight of each index in the subsequent parameter optimization process. Therefore, standardization is needed to eliminate such differences. Common standardization methods include converting data to the [0, 1] interval, which specifically operates by first finding the maximum and minimum values of the index data, then subtracting the minimum value from each data point value, and dividing by the difference between the maximum and minimum values to obtain the standardized data; or converting data to normal distribution data with mean 0 and standard deviation 1, and achieving it by subtracting the mean and dividing by the standard deviation. After standardization, all composition index data are in the same numerical interval, making different indexes comparable, so that in the subsequent similarity calculation and parameter optimization, each index can have a corresponding impact on the result according to its actual characteristics.

[0032] When obtaining the candidate feature vector according to the degree of association between the data set of the current parameter optimization stage of the target feature vector and the data set of the reference optimization stage of the other feature vector, the following procedure needs to be operated. The data set of the current parameter optimization stage of the target feature vector and the data set of the reference optimization stage of the other feature vector are compared to check whether there are same data points between them. The data set here contains all the coal ash composition index data associated with the feature vector in the corresponding stage, and each data point carries specific composition index information, such as carbon content and hydrogen content in coal ash. If it is found that there are data points that exist in both data sets during the comparison, the other feature vector corresponding to the data point is marked as a suspected candidate feature vector.

[0033] The data points that exist in both the data set of the current parameter optimization stage of the target feature vector and the data set of the reference optimization stage of the suspected candidate feature vector are defined as fluctuation data points. These fluctuation data points reflect the changes in data associated with different feature vectors in different parameter optimization stages. Then, the deviation degree of these fluctuation data points needs to be calculated to determine whether they have actual reference value.

[0034] When calculating the deviation degree of the fluctuation data point, the neighborhood comparison data points of the fluctuation data point need to be obtained first. The neighborhood comparison data points refer to other composition index data within a certain range around the fluctuation data point, which can provide a reference benchmark for measuring the deviation degree of the fluctuation data point. After obtaining the neighborhood comparison data points, the similarity difference between the fluctuation data point and these neighborhood comparison data points is calculated, and then the average of these difference values is taken to obtain the measurement difference. This measurement difference reflects the overall difference between the fluctuation data point and its surrounding data.

[0035] The ratio of the measurement difference of the fluctuation data point to the average of the measurement differences of all composition index data is calculated. The average of the measurement differences of all composition index data is obtained by calculating the average of the similarity differences between all data points and their respective neighborhood comparison data points, which represents the average difference level of the entire data set. Comparing the measurement difference of the fluctuation data point with this average value, the ratio obtained can reflect the deviation degree of the fluctuation data point relative to the overall average difference.

[0036] In order to make the deviation degrees of different fluctuation data points comparable, the above ratio needs to be normalized. Normalization can convert the ratio to a unified numerical range, such as [0, 1], so as to eliminate the influence of different original numerical ranges between different data points. The value obtained after normalization is the deviation degree of the fluctuation data point.

[0037] A preset deviation threshold is set according to the actual monitoring requirements and data characteristics, which is used to determine whether the fluctuation data point is valid. The deviation degree of the fluctuation data point is compared with the preset deviation threshold. If the deviation degree is less than the preset deviation threshold, it means that the difference of the fluctuation data point is within an acceptable range, and it is recorded as a valid fluctuation data point. Otherwise, it is considered that the difference of the fluctuation data point is too large and does not have reference value.

[0038] The suspected candidate feature vector is screened to see if the data set of the reference optimization stage contains valid fluctuation data points. If there are valid fluctuation data points in the data set of the reference optimization stage of a suspected candidate feature vector, it means that the feature vector has a meaningful association with the target feature vector in the current parameter optimization stage, and it is determined as a candidate feature vector. If there is no valid fluctuation data point, the suspected candidate feature vector is excluded and not included in the range of candidate feature vectors.

[0039] Through such a process, candidate feature vectors with actual correlation with the target feature vector can be selected from a large number of feature vectors. These candidate feature vectors can provide important reference information for the parameter iterative optimization of the adaptive weighted fusion algorithm, which helps to improve the processing accuracy of the algorithm for coal ash composition index data, and further realizes more accurate real-time monitoring of coal ash composition.

[0040] In embodiment 3, when obtaining the neighborhood comparison data points of the fluctuation data point, the following steps are needed. First, a preset range of area is defined as a comparison neighborhood with the fluctuation data point as the center. The preset range can be set according to the distribution density of the coal ash composition index data and the actual monitoring scene. For example, when the data distribution is relatively dense, the range can be set smaller to ensure that the data points in the neighborhood have strong correlation. When the data distribution is sparse, the range can be appropriately expanded to ensure that there are enough data points in the neighborhood as reference. The shape of the comparison neighborhood can be circular, square or other regular shape, and the specific selection depends on the spatial distribution characteristics of the data.

[0041] All coal ash composition index data contained in the comparison neighborhood are marked as neighborhood reference data points. These neighborhood reference data points cover various related data around the fluctuation data point and can reflect the overall distribution characteristics of the data in this region. Next, the comparison neighborhood is evenly divided into a preset number of directional regions with the fluctuation data point as the center. The preset number can be determined according to actual needs, for example, 8 directional regions corresponding to east, south, west, north, northeast, southeast, northwest and southwest. In this way, the relationship between the neighborhood reference data points and the fluctuation data point can be analyzed from different directions. The division process needs to ensure that the angle or area of each directional region is equal to ensure the fairness of the analysis.

[0042] For each directional region, the similarity difference between all neighborhood reference data points and the fluctuation data point in the directional region is calculated, and the difference values are averaged to obtain the region difference of the directional region. The calculation of the similarity difference can be based on the numerical difference of the data points on each component indicator, for example, by calculating the distance between the two data points in the multi-dimensional feature space to reflect the similarity difference, the greater the distance, the greater the difference. The region difference reflects the average difference level of the neighborhood reference data points and the fluctuation data point in the direction.

[0043] The difference average of all directional regions is calculated, which is obtained by adding the region differences of each directional region and dividing by the total number of directional regions, representing the overall average difference between the neighborhood reference data points and the fluctuation data point in all directions. The region difference of each directional region is compared with the difference average, and if the region difference of a certain directional region is less than the difference average, it means that the neighborhood reference data points in that direction have smaller difference with the fluctuation data point, and can better reflect the normal characteristics of the fluctuation data point. Therefore, the neighborhood reference data points in the directional region are selected as the neighborhood comparison data points.

[0044] The neighborhood comparison data points selected by the above method can more accurately reflect the surrounding data characteristics of the fluctuation data point under normal circumstances, providing a reliable reference benchmark for subsequent calculation of the deviation degree of the fluctuation data point. In actual operation, the range of the comparison neighborhood, the number of directional regions and other parameters can be adjusted according to the specific monitoring environment and data characteristics to adapt to the needs in different scenarios. For example, in areas with large data fluctuations, the comparison neighborhood range can be appropriately reduced to reduce the interference of irrelevant data; in areas with relatively uniform data distribution, the number of directional regions can be increased to analyze the data difference in different directions more carefully.

[0045] Through such steps, representative comparison data points can be selected from the neighborhood of the fluctuation data point, laying a foundation for subsequent accurate calculation of the deviation degree of the fluctuation data point, and further ensuring that the selection of candidate feature vectors is more reasonable and reliable.

[0046] When calculating the average of the similarity difference between the neighborhood reference data points and the fluctuation data point in each directional region, the formula can be used:

[0047] wherein, represents the region difference of the i-th directional region, represents the number of neighborhood reference data points in the i-th directional region, represents the similarity between the k-th neighborhood reference data point in the i-th directional region and the fluctuation data point, represents the reference similarity of the fluctuation data point itself (which can be set as the similarity between the fluctuation data point and itself, usually 1).

[0048] In the embodiment 4, the fluctuation range of the feature vector in the reference stage is obtained through a series of operations. First, the feature values of each feature vector in each reference parameter optimization stage are collected, which are extracted from the feature vector in the corresponding stage and can reflect the core characteristics of the feature vector in the stage. For example, the feature value of a certain feature vector in the reference stage 1 can be a representative value of the ash content, sulfur content and other component indicators in the data set of the stage, and the feature value in the reference stage 2 is another value obtained based on the data set of the stage.

[0049] Then, the difference degrees of the feature values of the same feature vector in any two reference parameter optimization stages are calculated. For example, for the feature vector A, the reference stages include stage 1, stage 2 and stage 3, and the difference degrees of the feature values of stage 1 and stage 2, stage 1 and stage 3, and stage 2 and stage 3 need to be calculated. When calculating, the absolute difference of the two feature values can be directly used. If the feature value of stage 1 is 50 and the feature value of stage 2 is 55, the difference degree of the two stages is 5. The relative difference, i.e. the ratio of the absolute difference to the average of the two feature values, can also be used, which is 5 divided by 52.5 in this case, obtaining the corresponding relative difference value.

[0050] Then, the feature value difference degrees of the feature vector in all two reference stages are counted, and the variance of the difference degrees is calculated, which is the fluctuation range of the feature vector in all reference parameter optimization stages. For example, the difference degrees of feature vector A in each two reference stages are 5, 3 and 7 respectively. The variance of these three values is calculated, and the result can reflect the fluctuation of the feature vector in the reference stage. The larger the variance, the more unstable the feature value of the feature vector in different reference stages.

[0051] When realizing real-time monitoring of components, the newly collected coal ash component index data is first input into the self-adaptive weighted fusion algorithm with optimized parameters. Assuming that the newly collected data includes ash content of 20%, sulfur content of 1.5%, moisture content of 8% and other indicators, the algorithm will analyze and process these data according to the optimized parameters to determine the category to which they belong, such as the category of “high ash content and low sulfur content”.

[0052] After collecting all the component index data of the category, the similarity of these data with the feature center of the category is calculated. The feature center is the core representation of all data of the category, such as the feature center of the category of "high ash content and low sulfur content" may be ash content 22%, sulfur content 1.2%, moisture content 7% and so on. When calculating the similarity of each data with the feature center, the closeness of each index can be compared to determine that the data of ash content 21%, sulfur content 1.3%, and moisture content 7.5% is relatively close to the index values of the feature center, and the similarity is higher; the data of ash content 18%, sulfur content 2.0%, and moisture content 9% is relatively far from the feature center, and the similarity is lower.

[0053] The lower quartile of these similarities is found, which is the value at the 25% position after sorting all similarity values from small to large. For example, a category has 100 data similarity values, and the 25th value after sorting is the lower quartile, which reflects the critical level of lower similarity in the category.

[0054] The similarity of the newly collected component index data with the feature center of the category to which it belongs is calculated and compared with the above-mentioned lower quartile. If the similarity of the new data is less than the lower quartile, it means that the data is relatively far from the feature center compared with most of the data in the category, and it is determined as an abnormal data point. For example, the lower quartile is 0.6, and the similarity of the new data is 0.5, which is less than the lower quartile, so the data is an abnormal data point.

[0055] A preset number threshold is set, which can be determined according to the accuracy requirements and production scenarios of actual monitoring, such as 3 times. When the number of continuously collected abnormal data points is greater than the threshold, such as 4 abnormal data points, it is determined that there is an abnormality in the coal ash component monitoring. This may mean that the data collected by the intelligent sensor has a systematic deviation, or the coal ash component itself has changed greatly, and further inspection of the sensor state or adjustment of the production process is needed.

[0056] In the whole process, the calculation of the fluctuation amplitude of the feature vector can help understand the stability of different feature vectors in the historical stage and provide a basis for the determination of the subsequent weight coefficient; while the determination of abnormal data points and the monitoring of continuous abnormalities can timely find the abnormal situation in the coal ash component monitoring process and ensure the effectiveness of real-time monitoring.

[0057] Embodiment 5: The stopping condition of parameter iterative optimization contains two cases. The first case is that the change rate of the data set of the target feature vector is calculated in a continuous preset number of parameter optimization stages. The change rate is calculated by taking the difference between the current stage data set and the last stage data set, and then comparing the difference with the total amount of the last stage data set to obtain the ratio, which is the change rate. For example, if the last stage data set contains 100 data points, and the current stage data set has 5 more data points and 3 less data points compared with the last stage data set, then the difference is 8, and the change rate is the ratio of 8 to 100. The preset change threshold is a value set according to the actual application scenario. When the change rate of the data set of the target feature vector in a continuous number of preset parameter optimization stages is less than the preset change threshold, it means that the change of the data set has reached a relatively stable state, and further parameter optimization has little effect on the improvement of the monitoring result, so the parameter iterative optimization is stopped.

[0058] The second case is that when the total number of parameter optimization reaches a preset maximum number, the parameter iterative optimization is stopped regardless of whether the change rate of the current data set is less than the preset change threshold. The preset maximum number is determined by considering factors such as monitoring efficiency and computing resources, for example, it is set to 50, and when the parameter optimization reaches the 50th time, the iteration process is automatically terminated.

[0059] In the implementation of real-time component monitoring, the dynamic sampling adjustment step is performed in the following manner. First, the frequency of abnormal data points in a unit time is counted, and the calculation method is the ratio of the number of abnormal data points in a unit time to the total number of data points collected in that unit time. For example, in 1 hour, 200 data points are collected, of which 30 are abnormal data points, so the frequency is the ratio of 30 to 200.

[0060] The preset frequency threshold is set according to the accuracy requirements and historical data characteristics of coal ash component monitoring, and is used to determine whether the current sampling density is appropriate. When the frequency of abnormal data points obtained by counting is higher than the preset frequency threshold, it means that there are more abnormal situations in the current monitoring data, and more intensive data collection is needed to capture more details, so the sampling interval density of the intelligent sensor is increased, that is, the time interval between adjacent two data collections is shortened, for example, from every 10 minutes to every 5 minutes. When the frequency of abnormal data points is lower than the preset frequency threshold, it means that the current data is relatively stable, and there is no need for too high sampling frequency, so the sampling interval density of the intelligent sensor can be reduced, that is, the sampling time interval is extended, for example, from every 5 minutes to every 10 minutes.

[0061] The adjustment range of the sampling interval density is positively correlated with the degree of the frequency deviation threshold, that is, the more the frequency of abnormal data points exceeds the preset frequency threshold, the greater the adjustment range of the sampling interval. For example, the preset frequency threshold is 10%, when the actual frequency is 20%, which exceeds the threshold by 10 percentage points, the sampling interval can be shortened from 10 minutes to 5 minutes; when the actual frequency is 30%, which exceeds the threshold by 20 percentage points, the sampling interval can be shortened from 10 minutes to 3 minutes, so as to obtain data more quickly and grasp the change of the coal ash composition in time. Conversely, when the actual frequency is much lower than the threshold, the extension range of the sampling interval also increases accordingly. Through this dynamic adjustment mode, the collection resources can be reasonably allocated under the premise of ensuring the monitoring accuracy, and unnecessary waste of energy and computing resources can be avoided.

[0062] It should be noted that, in this document, the terms such as first and second are used only to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply that there is any such actual relationship or order between these entities or operations. Moreover, the terms "include", "contain" or any other variants thereof are intended to cover non-exclusive inclusion, so that the process, method, article or device including a series of elements not only includes those elements, but also includes other elements not explicitly listed or inherent to such process, method, article or device.

[0063] Although the embodiments of the present application have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and variations can be made to the embodiments without departing from the principles and spirit of the present application, and the scope of the present application is defined by the appended claims and their equivalents.

Claims

1. A method for real-time monitoring of coal ash composition based on an Internet of Things intelligent sensor, characterized in that, The method comprises the steps of: obtaining a plurality of coal ash component index data collected by an intelligent sensor; performing parameter iteration optimization of an adaptive weighted fusion algorithm using the component index data to realize real-time component monitoring; in each parameter optimization stage, obtaining a target feature vector of the current parameter optimization stage, and obtaining a preset number of previous parameter optimization stages, referred to as reference parameter optimization stages; dividing the component index data into data sets of corresponding feature vectors according to the similarity of the component index data and the feature vectors in each parameter optimization stage; obtaining a candidate feature vector according to the correlation degree of the data set of the target feature vector in the current parameter optimization stage and the data set of the reference optimization stage of the other feature vectors; calculating a weight coefficient of each candidate feature vector, which is determined by the number of overlapping data points of the data set of the reference optimization stage and the current data set of the target feature vector, the preset number, the total amount of data points in the current data set, the difference in similarity of each data point and the target and candidate feature vectors, the fluctuation amplitude of the feature vector in the reference stage, and a preset basic weight; updating and optimizing the candidate feature vector and the target feature vector according to the weight coefficient.

2. The real time monitoring of fly ash composition based on IoT enabled smart sensor as claimed in claim 1 wherein, The method comprises the following steps: in each parameter optimization stage, obtaining the similarity of each component index data and each feature vector, dividing each component index data into the data set of the feature vector with the highest similarity, and obtaining the data set of each feature vector in each parameter optimization stage.

3. The real time monitoring of fly ash composition based on IoT enabled smart sensor as claimed in claim 1 wherein, The method comprises the following steps: if the data set of the target feature vector in the current parameter optimization stage and the data set of the reference optimization stage of the other feature vector have overlapping data points, the other feature vector is recorded as a suspected candidate feature vector; the data points that exist in both the data set of the target feature vector in the current parameter optimization stage and the data set of the suspected candidate feature vector in the reference optimization stage are recorded as fluctuation data points; the deviation degree of the fluctuation data points is calculated; the fluctuation data points with a deviation degree less than a preset deviation threshold are recorded as effective fluctuation data points; the suspected candidate feature vector in the data set of the reference optimization stage that has effective fluctuation data points is recorded as a candidate feature vector.

4. The real-time monitoring method of fly ash composition based on IoT- based smart sensor as claimed in claim 3, wherein, The method comprises the following steps: obtaining a neighborhood comparison data point of the fluctuation data point; recording the average difference in similarity between the fluctuation data point and the neighborhood comparison data point as a measure difference; and judging the normalized value of the ratio of the measure difference of the fluctuation data point to the average measure difference of all component index data as the deviation degree of the fluctuation data point.

5. The real time monitoring of fly ash composition based on IoT enabled smart sensor as claimed in claim 4 wherein, The obtaining of the neighborhood contrast data point of the fluctuation data point comprises: obtaining a region in a preset range of the fluctuation data point, denoted as a contrast neighborhood, and recording the component index data in the contrast neighborhood as a neighborhood reference data point; dividing the contrast neighborhood into a preset number of directional regions uniformly with the fluctuation data point as the center; calculating the mean value of the similarity difference between the neighborhood reference data point and the fluctuation data point in each directional region, denoted as a regional difference; and recording the neighborhood reference data point in the region with a regional difference less than the mean value of the differences of all directional regions as the neighborhood contrast data point.

6. The real time monitoring of fly ash composition based on IoT enabled smart sensor as claimed in claim 1 wherein, The obtaining method of the fluctuation amplitude comprises: obtaining the feature values of each feature vector at each parameter reference stage, calculating the difference degrees of the feature values of each feature vector at two reference parameter optimization stages, and recording the variance of the feature value difference degrees of each feature vector at all two reference parameter optimization stages as the fluctuation amplitude of the feature vector at all reference parameter optimization stages.

7. The real time monitoring of fly ash composition based on IoT enabled smart sensor as claimed in claim 1 wherein, The implementation of the component real-time monitoring comprises: inputting the newly collected component index data of the coal ash into the self-adaptive weighted fusion algorithm with completed parameter optimization to obtain the belonging component category of the newly collected component index data; obtaining the lower quartile of the similarity between each component index data and the feature center of the belonging component category; obtaining the similarity between the newly collected component index data and the feature center of the belonging component category, and determining that the newly collected component index data is an abnormal data point if the similarity between the newly collected component index data and the feature center of the belonging component category is less than the lower quartile; and determining that the coal ash component monitoring is abnormal if the number of continuously collected abnormal data points is greater than a preset number threshold.

8. The real time monitoring of fly ash composition based on IoT enabled smart sensor as claimed in claim 1 wherein, After obtaining the plurality of coal ash component index data collected by the intelligent sensor, the method further comprises a pre-processing step of the component index data, wherein the pre-processing comprises: removing outliers in the data, interpolating and filling missing data, and normalizing the data to a preset range.

9. The real time monitoring of fly ash composition based on IoT enabled smart sensor as claimed in claim 1 wherein, The stop condition of the parameter iterative optimization is that the change rate of the data set of the target feature vector in a continuous preset number of parameter optimization stages is less than a preset change threshold, or the total number of parameter optimizations reaches a preset maximum number.

10. The real time monitoring of fly ash composition based on IoT enabled smart sensor as claimed in claim 1 wherein, In the implementation of the component real-time monitoring, a dynamic sampling adjustment step is further included, wherein the dynamic sampling adjustment step comprises: counting the frequency of abnormal data points in a unit time, increasing the sampling interval density of the intelligent sensor when the frequency is higher than a preset frequency threshold, and decreasing the sampling interval density of the intelligent sensor when the frequency is lower than the preset frequency threshold, and the adjustment amplitude of the sampling interval density is positively correlated with the degree of deviation of the frequency from the threshold.

Images