A medium efficiency dust filter service life prediction system and method

Abstract

This application provides a system and method for predicting the service life of a medium-efficiency dust filter, relating to the field of medium-efficiency dust filter operation and maintenance technology. By calculating the cumulative slope and applying sliding window filtering to historical resistance sequences, a trend-enhancing feature sequence and a fluctuation-suppressing feature sequence are obtained. These two sequences are then subjected to parallel heterogeneous prediction to obtain trend life prediction values ​​and fluctuation life prediction values. The airflow stability index and dust concentration change rate are extracted from the operating condition data. Fuzzy inference is performed based on the airflow stability index, dust concentration change rate, trend life prediction value, and fluctuation life prediction value to obtain the weight decision factors for heterogeneous prediction. Finally, the trend life prediction value and fluctuation life prediction value are weighted and fused to obtain the life prediction value of the medium-efficiency dust filter. This application enables intelligent weighted fusion of heterogeneous prediction results driven by fuzzy inference, thereby improving the reliability of life prediction.

Description

Technical Field

[0001] This application relates to the field of operation and maintenance technology for medium-efficiency dust filters, and more specifically, to a system and method for predicting the service life of medium-efficiency dust filters. Background Technology

[0002] Medium-efficiency dust filters are one of the core components for ensuring air quality and normal equipment operation. The operation and maintenance management of medium-efficiency dust filters mainly relies on a preventive replacement strategy, which aims to replace medium-efficiency dust filters in time before their performance fails, so as to avoid dust penetrating and contaminating clean areas or causing insufficient system airflow and a surge in energy consumption.

[0003] Two modes are used to predict the lifespan of medium-efficiency dust filters: periodic replacement based on a fixed cycle or replacement based on a set final resistance (usually 2 to 4 times the initial resistance). However, both modes have revealed significant technical drawbacks in long-term engineering practice. First, periodic replacement easily overlooks the dynamic changes in external dust concentration and internal actual load of the air conditioning system under different seasons and operating frequencies. This leads to premature replacement during clean periods, increasing procurement costs, and potentially delayed replacement during periods of severe pollution, resulting in high system energy consumption. Second, the replacement mode based on set final resistance is applicable to variable frequency air conditioning systems. When the air conditioning system operates at low airflow for extended periods, the resistance of the medium-efficiency dust filter may not reach the preset final resistance value, easily causing the filter to exceed its service life, reduce filtration efficiency, and lead to the potential risk of losing early warning functionality due to resistance sensing failure. Therefore, how to achieve intelligent weighted fusion of heterogeneous prediction results driven by fuzzy reasoning to improve the reliability of lifespan prediction is a challenge facing the industry. Summary of the Invention

[0004] This application provides a system and method for predicting the service life of a medium-efficiency dust filter, which can realize intelligent weighted fusion of heterogeneous prediction results driven by fuzzy inference to improve the reliability of service life prediction.

[0005] In a first aspect, this application provides a method for predicting the service life of a medium-efficiency dust filter, the method comprising the following steps:

[0006] Collect historical resistance sequences and operating condition data of medium-efficiency dust filters;

[0007] The historical resistance sequences are subjected to cumulative slope calculation and sliding window filtering to obtain trend strengthening feature sequences and volatility suppression feature sequences.

[0008] The trend enhancement feature sequence and the fluctuation suppression feature sequence are heterogeneously predicted in parallel to obtain the trend lifetime prediction value and the fluctuation lifetime prediction value. The air volume stability index and dust concentration change rate are extracted from the operating condition data.

[0009] Based on the air volume stability index, the dust concentration change rate, the trend lifetime prediction value, and the fluctuation lifetime prediction value, fuzzy reasoning is performed to obtain the weight decision factors for heterogeneous prediction.

[0010] The predicted lifespan of the medium-efficiency dust filter is obtained by weighting and fusing the trend lifespan prediction value and the fluctuation lifespan prediction value according to the weight decision factor.

[0011] In this embodiment, the collection of historical resistance sequences and operating condition data of medium-efficiency dust filters specifically includes:

[0012] Historical resistance sequences of medium-efficiency dust filters were collected using a micro differential pressure sensor.

[0013] Operating condition data of medium-efficiency dust filters are collected through an IoT gateway.

[0014] In this embodiment, the cumulative slope calculation and sliding window filtering are performed on the historical resistance sequence to obtain the trend strengthening feature sequence and the volatility suppression feature sequence, specifically including:

[0015] Perform point-by-point cumulative slope calculation and linear regression analysis on the historical resistance sequence to obtain the trend strengthening characteristic sequence;

[0016] By applying median filtering to the historical resistance sequence using a sliding window, a fluctuation suppression feature sequence is obtained.

[0017] In this embodiment, performing heterogeneous prediction on the trend enhancement feature sequence and the volatility suppression feature sequence in parallel to obtain the trend lifetime prediction value and volatility lifetime prediction value specifically includes:

[0018] Construct a heterogeneous prediction model, which includes a support vector machine regression model and a temporal convolutional network model;

[0019] The trend reinforcement feature sequence is input into the support vector machine regression model for kernel function mapping and regression analysis to obtain the trend lifetime prediction value.

[0020] The fluctuation suppression feature sequence is input into the temporal convolutional network model for temporal feature convolution to obtain the fluctuation lifetime prediction value.

[0021] In this embodiment, extracting the airflow stability index and dust concentration change rate from the operating condition data specifically includes:

[0022] Obtain the air volume data sequence from the operating condition data;

[0023] The airflow stability index is determined by the coefficient of variation of the airflow data sequence;

[0024] Obtain dust concentration data from the operating condition data, and perform sliding window linear regression analysis on the dust concentration data to obtain a slope sequence;

[0025] The dust concentration change rate is determined based on the changing characteristics of the slope sequence.

[0026] In this embodiment, fuzzy inference is performed based on the airflow stability index, the dust concentration change rate, the trend lifetime prediction value, and the fluctuation lifetime prediction value to obtain the weighted decision factors for heterogeneous prediction, specifically including:

[0027] The air volume stability index, the dust concentration change rate, the trend lifetime prediction value, and the fluctuation lifetime prediction value are fuzzified using a triangular membership function to obtain an input fuzzy set.

[0028] The input fuzzy set and the preset fuzzy rule base are subjected to inference aggregation to obtain the output fuzzy set;

[0029] The centroid method is used to defuzzify the output fuzzy set to obtain the trend model weight coefficients and the fluctuation model weight coefficients.

[0030] The weighted decision factors for heterogeneous prediction are composed of the weighted coefficients of the trend model and the weighted coefficients of the volatility model.

[0031] In this embodiment, the weighted fusion of the trend lifetime prediction value and the fluctuation lifetime prediction value based on the weighted decision factor to obtain the lifetime prediction value of the medium-efficiency dust filter specifically includes:

[0032] The weight decision factors are vectorized to obtain a weight vector;

[0033] A prediction value vector is determined using the trend lifetime prediction value and the fluctuation lifetime prediction value;

[0034] The predicted lifespan of the medium-efficiency dust filter is determined based on the weight vector and the predicted value vector.

[0035] In this embodiment, the weight decision factor for heterogeneous prediction is a binary tuple consisting of trend model weight coefficients and volatility model weight coefficients. The trend model weight coefficients are used to quantify the contribution of the trend lifetime prediction value in the heterogeneous prediction process, and the volatility model weight coefficients are used to quantify the contribution of the volatility lifetime prediction value in the heterogeneous prediction process.

[0036] In this embodiment, the medium-efficiency dust filter includes a bag-type medium-efficiency dust filter and a box-type medium-efficiency dust filter.

[0037] Secondly, this application provides a medium-efficiency dust filter lifespan prediction system for performing a medium-efficiency dust filter lifespan prediction method, the lifespan prediction system comprising:

[0038] The data acquisition module is used to collect historical resistance sequences and operating condition data of medium-efficiency dust filters;

[0039] The feature construction module is used to perform cumulative slope calculation and sliding window filtering on the historical resistance sequence to obtain trend strengthening feature sequence and volatility suppression feature sequence;

[0040] The heterogeneous prediction module is used to perform heterogeneous prediction on the trend enhancement feature sequence and the fluctuation suppression feature sequence in parallel to obtain the trend lifetime prediction value and the fluctuation lifetime prediction value, and to extract the air volume stability index and dust concentration change rate from the operating condition data.

[0041] The weighted decision module is used to perform fuzzy reasoning based on the air volume stability index, the dust concentration change rate, the trend lifetime prediction value, and the fluctuation lifetime prediction value to obtain the weighted decision factors for heterogeneous prediction.

[0042] The fusion output module is used to perform weighted fusion of the trend lifetime prediction value and the fluctuation lifetime prediction value according to the weight decision factor to obtain the lifetime prediction value of the medium-efficiency dust filter.

[0043] The technical solutions provided by the embodiments disclosed in this application have the following beneficial effects:

[0044] Historical resistance sequences and operating condition data of medium-efficiency dust filters are collected. Cumulative slope calculation and sliding window filtering are performed on the historical resistance sequences to obtain trend enhancement feature sequences and fluctuation suppression feature sequences. The trend enhancement feature sequences and fluctuation suppression feature sequences are then subjected to heterogeneous prediction in parallel to obtain trend lifetime prediction values ​​and fluctuation lifetime prediction values. Airflow stability index and dust concentration change rate are extracted from the operating condition data. Fuzzy inference is performed based on the airflow stability index, dust concentration change rate, trend lifetime prediction value, and fluctuation lifetime prediction value to obtain the weighted decision factors for heterogeneous prediction. The trend lifetime prediction value and fluctuation lifetime prediction value are then weighted and fused according to the weighted decision factors to obtain the lifetime prediction value of the medium-efficiency dust filter.

[0045] Therefore, this application can achieve intelligent weighted fusion of heterogeneous prediction results driven by fuzzy inference. First, by collecting historical resistance sequences and operating condition data of medium-efficiency dust filters, a comprehensive and multi-dimensional data foundation is provided for life assessment. Then, by performing cumulative slope calculation and sliding window filtering on the historical resistance sequences, trend-enhancing feature sequences and fluctuation-suppressing feature sequences are obtained. This effectively decouples the long-term irreversible aging trend contained in the original resistance sequence from short-term random fluctuation noise, providing an accurate feature foundation for subsequent heterogeneous prediction. Second, by inputting the trend-enhancing feature sequences and fluctuation-suppressing feature sequences in parallel into a support vector machine regression model and a temporal convolutional network model for heterogeneous prediction, trend life prediction values ​​and fluctuation life prediction values ​​are obtained respectively. This leverages the advantages of different model architectures to obtain macro-level trend predictions. The prediction results are generated from two complementary perspectives: grasping the trend and adapting to micro-fluctuations. This approach helps overcome the limitations of a single model perspective. Then, the airflow stability index and dust concentration change rate are extracted from the operating condition data. Combined with trend and fluctuation predictions, fuzzy inference is performed based on a triangular membership function and a fuzzy rule base to obtain the weighted decision factors for heterogeneous predictions. This incorporates the dynamic changes in real-time operating conditions and the inherent uncertainties of the prediction model into the decision-making process, facilitating intelligent and adaptive weight allocation. Finally, the two predictions are weighted and fused according to the weighted decision factors to generate a lifetime prediction value. This allows for the reliable fusion of multi-perspective prediction information at the decision-making level, thereby enabling intelligent weighted fusion of heterogeneous prediction results driven by fuzzy inference and improving the reliability of lifetime prediction results under complex and variable operating conditions.

[0046] In summary, the technical solution adopted in this application can realize intelligent weighted fusion of heterogeneous prediction results driven by fuzzy reasoning, so as to improve the credibility of lifespan prediction. Attached Figure Description

[0047] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only for this embodiment of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0048] Figure 1 This is a flowchart of a method for predicting the service life of a medium-efficiency dust filter according to the present application;

[0049] Figure 2 This is an exemplary flowchart for determining trend reinforcement feature sequences and volatility suppression feature sequences according to the present application;

[0050] Figure 3This is an exemplary flowchart for determining the airflow stability index and dust concentration change rate according to the present application;

[0051] Figure 4 This is a module structure diagram of a medium-efficiency dust filter lifespan prediction system provided in this application. Detailed Implementation

[0052] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0053] This application provides a system and method for predicting the service life of a medium-efficiency dust filter. The core of this system involves collecting historical resistance sequences and operating condition data of the medium-efficiency dust filter; calculating the cumulative slope and applying sliding window filtering to the historical resistance sequences to obtain trend enhancement feature sequences and fluctuation suppression feature sequences; performing heterogeneous prediction on the trend enhancement feature sequences and fluctuation suppression feature sequences in parallel to obtain trend life prediction values ​​and fluctuation life prediction values; extracting the airflow stability index and dust concentration change rate from the operating condition data; performing fuzzy inference based on the airflow stability index, dust concentration change rate, trend life prediction values, and fluctuation life prediction values ​​to obtain weighted decision factors for heterogeneous prediction; and weighting and fusing the trend life prediction values ​​and fluctuation life prediction values ​​according to the weighted decision factors to obtain the predicted service life value of the medium-efficiency dust filter.

[0054] Example 1: To better understand the above technical solution, the following will provide a detailed description of the technical solution in conjunction with the accompanying drawings and specific implementation methods. (Refer to...) Figure 1 As shown in the figure, this is a flowchart of a method for predicting the service life of a medium-efficiency dust filter according to this embodiment of the present application. The method for predicting the service life includes the following steps:

[0055] In step S1, the historical resistance sequence and operating condition data of the medium-efficiency dust filter are collected.

[0056] It should be noted that the medium-efficiency dust filter in this application includes bag-type medium-efficiency dust filters and box-type medium-efficiency dust filters. The bag-type medium-efficiency dust filter refers to a medium-efficiency dust filter that uses synthetic fiber non-woven filter media to form a bag-like structure, increasing dust holding capacity by increasing the filtration area. The box-type medium-efficiency dust filter refers to a medium-efficiency dust filter in which filter media are assembled within a metal or plastic frame to form a box-like structure. Furthermore, the medium-efficiency dust filter in this application includes: a filter material layer, a support frame, and a sealing component. The filter material layer is the core functional layer for dust filtration, made of glass fiber or synthetic fiber material, used to capture particulate pollutants in the air. The support frame is made of metal and high-strength plastic to maintain the structural stability of the medium-efficiency dust filter. The sealing component is made of rubber and foam material to ensure an airtight connection between the medium-efficiency dust filter and the mounting frame.

[0057] In this embodiment, the historical resistance sequence and operating condition data of the medium-efficiency dust filter can be collected in the following manner:

[0058] Historical resistance sequences of medium-efficiency dust filters were collected using a micro differential pressure sensor.

[0059] Operating condition data of medium-efficiency dust filters are collected through an IoT gateway.

[0060] In a specific implementation, a micro differential pressure sensor can be used to collect the historical resistance sequence of the medium-efficiency dust filter. This micro differential pressure sensor is a differential pressure detection device used to collect resistance data of the medium-efficiency dust filter in real time during operation. The measurement range of the micro differential pressure sensor can be set to 0-500 Pa, and the sampling frequency can be set to 1 Hz, facilitating long-term recording of the resistance changes of the medium-efficiency dust filter. Alternatively, in a specific implementation, an IoT gateway can be used to collect the operating condition data of the medium-efficiency dust filter. This IoT gateway refers to a communication hub device. As a preferred embodiment, the IoT gateway can be an industrial gateway supporting the Modbus / TCP protocol, which can connect to an environmental monitoring sensor via an RS485 interface. The environmental monitoring sensor is used to collect operating airflow and environmental dust concentration data, with a collection interval of once every 5 minutes, facilitating tracking of changes in the system's operating status.

[0061] It should be noted that the historical resistance sequence in this application refers to a continuous data set describing the pressure difference between the front and rear ends of a medium-efficiency dust filter. The historical resistance sequence includes: timestamps, instantaneous pressure difference values, and average pressure difference values, which are used to quantify the degree of clogging and aging rate of the medium-efficiency dust filter. The operating condition data is a set of parameters describing the operating environment of the medium-efficiency dust filter, including: system air volume, ambient dust concentration, and operating time, which are used to characterize the impact of external loads on the aging process of the medium-efficiency dust filter.

[0062] In step S2, the historical resistance sequence is subjected to cumulative slope calculation and sliding window filtering to obtain the trend strengthening feature sequence and the fluctuation suppression feature sequence.

[0063] Preferably, in this embodiment, reference Figure 2 As shown, this figure is an exemplary flowchart for determining the trend reinforcement feature sequence and the volatility suppression feature sequence according to the present application. In this embodiment, the cumulative slope calculation and sliding window filtering are performed on the historical resistance sequence to obtain the trend reinforcement feature sequence and the volatility suppression feature sequence, which can be achieved by the following steps:

[0064] First, in step S21, point-by-point cumulative slope calculation and linear regression analysis are performed on the historical resistance sequence to obtain the trend strengthening characteristic sequence;

[0065] Then, in step S22, the historical resistance sequence is subjected to median filtering through a sliding window to obtain a fluctuation suppression feature sequence.

[0066] In specific implementation, firstly, a point-by-point cumulative slope calculation is performed on the historical resistance sequence. That is, starting from the first valid data point in the historical resistance sequence, each data point is traversed sequentially in time. For the currently processed data point, all historical resistance values ​​from the start point of the sequence to that data point are extracted to form a subsequence. The least squares method is then used to linearly fit the subsequence to obtain its slope value. This process is repeated until the entire historical resistance sequence is traversed. The sequence of all obtained slope values ​​in time order is then used as the trend reinforcement feature sequence. Next, a median filter is applied to the historical resistance sequence using a sliding window. That is, a fixed time period can be preset. A window, for example, contains 3600 data points (corresponding to 1 hour). This time window starts from the beginning of the historical resistance sequence and slides by a fixed step size (e.g., 60 data points). At each position where the time window stops, the median of all historical resistance values ​​within the time window is extracted using statistical techniques. This sliding calculation process is repeated until the time window covers the entire historical resistance sequence. Then, the sequence of medians extracted from all time windows in chronological order is used as the fluctuation suppression feature sequence. The time window and step size can be preset according to the actual clogging degree and actual aging rate of the medium-efficiency dust filter, and are not limited here.

[0067] It should be noted that the trend enhancement feature sequence in this application is a feature sequence describing the long-term cumulative growth trend of the resistance value over time, used to capture the irreversible clogging process of the medium-efficiency dust filter; the fluctuation suppression feature sequence is a feature sequence representing the macroscopic aging state and resistance change profile of the medium-efficiency dust filter, which can filter out short-term spike pulses and random fluctuations caused by sudden changes in airflow or sensor noise in the resistance sequence, thus facilitating the provision of a stable signal basis.

[0068] In step S3, the trend enhancement feature sequence and the fluctuation suppression feature sequence are heterogeneously predicted in parallel to obtain the trend lifetime prediction value and the fluctuation lifetime prediction value. The air volume stability index and dust concentration change rate are extracted from the operating condition data.

[0069] In this embodiment, the trend enhancement feature sequence and the volatility suppression feature sequence are heterogeneously predicted in parallel to obtain the trend lifetime prediction value and volatility lifetime prediction value. Specifically, this can be achieved in the following manner:

[0070] Construct a heterogeneous prediction model, which includes a support vector machine regression model and a temporal convolutional network model;

[0071] The trend reinforcement feature sequence is input into the support vector machine regression model for kernel function mapping and regression analysis to obtain the trend lifetime prediction value.

[0072] The fluctuation suppression feature sequence is input into the temporal convolutional network model for temporal feature convolution to obtain the fluctuation lifetime prediction value.

[0073] In specific implementation, firstly, a heterogeneous prediction model is constructed: a support vector machine regression model is initialized using Python's Scikit-learn machine learning library, and a temporal convolutional network model is built using the PyTorch deep learning framework. The combination of the support vector machine regression model and the temporal convolutional network model serves as the heterogeneous prediction model. Then, the trend reinforcement feature sequence is input into the support vector machine regression model: the `fit` method of the support vector machine regression model is called to train the model using the trend reinforcement feature sequence as training data. After training, the `predict` method of the support vector machine regression model is used to infer the trend reinforcement feature sequence, and the predicted value output after inference is used as the trend lifetime prediction value. Finally, the fluctuation suppression feature sequence is input into the temporal convolutional network model: the fluctuation suppression feature sequence is converted into a tensor format supported by PyTorch and then input into the temporal convolutional network model. The scalar value output by the last fully connected layer of the temporal convolutional network model is automatically extracted through the convolutional layers inside the model, and this scalar value is used as the fluctuation lifetime prediction value.

[0074] It should be noted that the support vector machine regression model in this application is a machine learning model for small sample, high-dimensional data. It can map low-dimensional nonlinearly separable trend sequences to a high-dimensional feature space through kernel functions, and find an optimal linear regression hyperplane in the high-dimensional feature space, which is beneficial for making point predictions on the long-term lifespan trend of medium-efficiency dust filters. The temporal convolutional network model is a deep learning architecture for processing time-series data. It can capture multiple time-scale dependent patterns in fluctuation sequences from short-term to long-term by using dilated convolutional layers, and thus predict the lifespan affected by complex fluctuations. Heterogeneous prediction can process input features with different physical meanings separately, and obtain differentiated results from multiple perspectives. The prediction results are complementary; in addition, in this application, the trend life prediction value is a numerical estimate of the remaining service life that a medium-efficiency dust filter can achieve under ideal and stable operating conditions according to its long-term aging trend, which can provide a macro-trend judgment on the degradation of the basic performance of the medium-efficiency dust filter; the fluctuation life prediction value represents a numerical estimate of the impact of operating condition fluctuations on the aging rate of the medium-efficiency dust filter, which can capture the deviation of the expected life caused by the dynamic operating environment, and provide a sensitive response and micro-correction judgment to the current actual load state; in this embodiment, the trend life prediction value and the fluctuation life prediction value can be used for synergistic diagnosis of the health status of the medium-efficiency dust filter from different time scales and physical perspectives.

[0075] Preferably, in this embodiment, reference Figure 3 As shown, this figure is an exemplary flowchart for determining the airflow stability index and dust concentration change rate according to the present application. In this embodiment, the extraction of the airflow stability index and dust concentration change rate from the operating condition data can be achieved by the following steps:

[0076] First, in step S31, the air volume data sequence in the operating condition data is obtained;

[0077] Secondly, in step S32, the air volume stability index is determined by the coefficient of variation of the air volume data sequence;

[0078] Then, in step S33, dust concentration data in the operating condition data is obtained, and sliding window linear regression analysis is performed on the dust concentration data to obtain a slope sequence;

[0079] Finally, in step S34, the dust concentration change rate is determined based on the change characteristics of the slope sequence.

[0080] In specific implementation, firstly, the air volume data sequence in the operating condition data is obtained. That is, the system air volume in the operating condition data can be extracted in time-stamp order through a database query statement, and the sequence of system air volume in time-stamp order is taken as the air volume data sequence. Secondly, the air volume stability index is determined by the coefficient of variation of the air volume data sequence. That is, the standard deviation and arithmetic mean of the air volume data sequence can be calculated by calling the std and mean functions of the NumPy library, respectively, and the ratio of the standard deviation to the arithmetic mean is taken as the coefficient of variation of the air volume data sequence. Then, the reciprocal of the coefficient of variation is taken as the air volume stability index. Then, the dust concentration data in the operating condition data is obtained and the dust concentration data is analyzed. Dust concentration data is analyzed using sliding window linear regression. This involves extracting environmental dust concentrations from a database query and using a sequence of dust concentrations ordered by query timestamps. A sliding time window is then used to move the dust concentration data across the window. For each position the window remains in, the `linregress` function from the SciPy library is called to perform linear regression on the concentration data within the window. This calculates the slope at each position, and the sequence of all slopes ordered by timestamps is then used as the slope sequence. Finally, the rate of change of dust concentration is determined based on the characteristics of the slope sequence; specifically, the absolute value of the arithmetic mean of all slopes in the slope sequence is taken as the rate of change of dust concentration.

[0081] It should be noted that the airflow stability index in this application is a dimensionless index that quantifies the stability of the air supply of a medium-efficiency dust filter. The larger the value of the airflow stability index, the smaller the airflow fluctuation and the more stable the operation of the air supply system of the medium-efficiency dust filter. The dust concentration change rate is a scalar index that characterizes the degree of dynamic change of dust load in the external environment of the medium-efficiency dust filter. The larger the value of the dust concentration change rate, the stronger the trend of rising or falling dust concentration in the external environment of the medium-efficiency dust filter. In addition, in this embodiment, the sliding window linear regression analysis is a data processing method that captures the local change trend of a small segment of continuous time series data by fitting a straight line on the data. It can convert continuous and fluctuating dust concentration data into slope values ​​that represent the direction and intensity of change.

[0082] In step S4, fuzzy inference is performed based on the air volume stability index, the dust concentration change rate, the trend lifetime prediction value, and the fluctuation lifetime prediction value to obtain the weight decision factors for heterogeneous prediction.

[0083] In this embodiment, the weighted decision factors for heterogeneous prediction are obtained through fuzzy inference based on the airflow stability index, the dust concentration change rate, the trend lifetime prediction value, and the fluctuation lifetime prediction value. Specifically, this can be achieved in the following manner:

[0084] The air volume stability index, the dust concentration change rate, the trend lifetime prediction value, and the fluctuation lifetime prediction value are fuzzified using a triangular membership function to obtain an input fuzzy set.

[0085] The input fuzzy set and the preset fuzzy rule base are subjected to inference aggregation to obtain the output fuzzy set;

[0086] The centroid method is used to defuzzify the output fuzzy set to obtain the trend model weight coefficients and the fluctuation model weight coefficients.

[0087] The weighted decision factors for heterogeneous prediction are composed of the weighted coefficients of the trend model and the weighted coefficients of the volatility model.

[0088] It should be noted that the triangular membership function in this embodiment refers to a mathematical function with a triangular shape, used to precisely calculate the extent to which a specific input value (such as an airflow stability index of 0.8) belongs to a certain vague linguistic concept (such as "high stability"). This triangular membership function is defined by three key parameters: the starting point a, the peak point m, and the ending point b. The mathematical expression of this triangular membership function is... as follows:

[0089] ;

[0090] Where x is the input value, the mathematical expression is as follows: when the input value x is less than or equal to a or greater than or equal to b, the degree of membership of the input value to the concept is 0; when the input value x is greater than a and less than or equal to m, the degree of membership of the input value is in the increasing interval (0, 1); when x is greater than m and less than or equal to b, the degree of membership of the input value is in the decreasing interval (1, 0). As a preferred embodiment, "high stability" can be set to a=0.6, m=0.8, b=1.0. For example, when the air volume stability index is 0.7, the degree of membership of the air volume stability index to "high stability" can be calculated to be 0.5 through the triangular membership function, which can be used to determine that the air supply system of the medium-efficiency dust filter operates at a moderate level.

[0091] In addition, in this embodiment, the input fuzzy set refers to the membership set obtained by transforming the airflow stability index, dust concentration change rate, trend lifetime prediction value, and fluctuation lifetime prediction value through triangular membership functions; the output fuzzy set refers to the result set composed of fuzzy language and membership degrees obtained by substituting the input fuzzy set into a preset fuzzy rule base for reasoning. The fuzzy language is a non-precise vocabulary set used to describe the state of variables, including fuzzy language for the airflow stability index (such as "low stability", "medium stability", "high stability") and fuzzy language for the dust concentration change rate (such as "change rate"). The system uses fuzzy language for predicting trend lifetime (e.g., "predicted value low", "predicted value medium", "predicted value high") and fuzzy language for predicting fluctuation lifetime (e.g., "predicted value low", "predicted value medium", "predicted value high") to convert precise numerical inputs into natural language, which can lay the foundation for fuzzy reasoning based on natural language rules. The preset fuzzy rule base is a knowledge base that stores expert experience knowledge. This knowledge base defines the nonlinear mapping relationship from multi-dimensional working conditions and prediction results to the final weight preference in the form of "IF (input condition) THEN (output conclusion)".

[0092] In practical implementation, firstly, for the four input variables—airflow stability index, dust concentration change rate, trend lifetime prediction value, and fluctuation lifetime prediction value—four fuzzy linguistic values ​​are defined respectively. These include fuzzy linguistic values ​​for the airflow stability index (e.g., "low stability", "medium stability", "high stability"), dust concentration change rate (e.g., "low change rate", "medium change rate", "high change rate"), trend lifetime prediction value (e.g., "low prediction value", "medium prediction value", "high prediction value"), and fluctuation lifetime prediction value (e.g., "low prediction value", "medium prediction value", "high prediction value"). For the input values ​​of the airflow stability index, dust concentration change rate, trend lifetime prediction value, and fluctuation lifetime prediction value, the membership degree of the corresponding four linguistic values ​​is calculated using a triangular membership function. The set of all linguistic values ​​and their membership degrees is then... The input fuzzy set is used as the input fuzzy set. Next, inference aggregation is performed on the input fuzzy set and a preset fuzzy rule base. Specifically, in the preset fuzzy rule base, a database query is used to find all "IF-THEN" rules whose preconditions (i.e., the "IF" part) match the current fuzzy state. For each rule, the minimum value among all the membership degrees of the preconditions is taken as the overall credibility of the rule. The triangular membership function corresponding to the conclusion part (i.e., the "THEN" part) of the rule is then "weakened" based on the credibility. Specifically, the peak value m in the triangular membership function is changed to the credibility value. Repeating the above steps yields the four fuzzy linguistic values ​​corresponding to the four input variables and the maximum membership degree of the four fuzzy linguistic values. Finally, the dataset consisting of the four fuzzy linguistic values ​​corresponding to the four input variables and the maximum membership degree of the four fuzzy linguistic values ​​is used as the output fuzzy set.

[0093] In addition, in specific implementation, the centroid method is used to defuzzify the output fuzzy set. That is, the variables in the output fuzzy set are discretized into multiple discrete points by the centroid method, and the weighted average of the coordinates of all discrete points and their corresponding membership degrees is calculated. The coordinate value of each discrete point is the object being averaged, and the membership degree corresponding to the discrete point is used as the weight when calculating the weighted average. This weighted average can be used as the trend model weight coefficient. Then, according to the constraint that the sum of all weight coefficients is 1, the fluctuation model weight coefficient can be calculated, that is: fluctuation model weight coefficient = 1 - trend model weight coefficient. Finally, the trend model weight coefficient and the fluctuation model weight coefficient constitute the weight decision factor of heterogeneous prediction. That is, the pair formed by the trend model weight coefficient and the fluctuation model weight coefficient is used as the weight decision factor of heterogeneous prediction, that is, the weight decision factor = [trend model weight coefficient, fluctuation model weight coefficient].

[0094] It should be noted that, in this embodiment, the weight decision factor for heterogeneous prediction is a binary tuple consisting of trend model weight coefficients and fluctuation model weight coefficients. That is, the heterogeneous prediction in this application includes support vector machine regression prediction and temporal convolutional network prediction. The trend model weight coefficients in the weight decision factor are applied to support vector machine regression prediction, and the fluctuation model weight coefficients in the weight decision factor are applied to temporal convolutional network prediction. The trend model weight coefficients are used to quantify the contribution of the trend lifetime prediction value to the heterogeneous prediction process, and the fluctuation model weight coefficients are used to quantify the contribution of the fluctuation lifetime prediction value to the heterogeneous prediction process. In addition, the centroid method is a conventional defuzzification method that obtains the most accurate value representing the fuzzy set by calculating the geometric center of the fuzzy set shape.

[0095] In step S5, the trend lifetime prediction value and the fluctuation lifetime prediction value are weighted and fused according to the weight decision factor to obtain the lifetime prediction value of the medium-efficiency dust filter.

[0096] In this embodiment, the predicted lifespan of the medium-efficiency dust filter is obtained by weighting and fusing the trend lifetime prediction value and the fluctuation lifetime prediction value according to the weight decision factor. Specifically, this can be done in the following way:

[0097] The weight decision factors are vectorized to obtain a weight vector;

[0098] A prediction value vector is determined using the trend lifetime prediction value and the fluctuation lifetime prediction value;

[0099] The predicted lifespan of the medium-efficiency dust filter is determined based on the weight vector and the predicted value vector.

[0100] In practice, firstly, the trend model weight coefficients and fluctuation model weight coefficients in the weight decision factors can be converted into vector form by one-hot encoding, and then this vector form can be used as the weight vector. Secondly, the trend lifetime prediction value and fluctuation lifetime prediction value are arranged in the same order as the weight vector to obtain the vector as the prediction value vector. Finally, the numpy.dot() function in the NumPy library is used to calculate the dot product of the prediction value vector and the weight vector, and then this dot product is used as the lifetime prediction value of the medium-efficiency dust filter.

[0101] It should be noted that the weighted fusion in this application refers to the linear combination of the weight vector and the predicted value vector through vector dot product operation, which can combine two complementary but different predicted values ​​into a reliable prediction result based on the weights determined by intelligent decision.

[0102] In summary, the technical solution adopted in this application can realize intelligent weighted fusion of heterogeneous prediction results driven by fuzzy reasoning, so as to improve the credibility of lifespan prediction.

[0103] Example 2: This application provides a system for predicting the service life of a medium-efficiency dust filter, referring to... Figure 4 As shown, this figure is a modular structure diagram of a medium-efficiency dust filter lifespan prediction system provided in this application. The lifespan prediction system includes:

[0104] Data acquisition module 100 is used to collect historical resistance sequences and operating condition data of medium-efficiency dust filters;

[0105] The feature construction module 200 is used to perform cumulative slope calculation and sliding window filtering on the historical resistance sequence to obtain trend strengthening feature sequence and volatility suppression feature sequence;

[0106] The heterogeneous prediction module 300 is used to perform heterogeneous prediction on the trend enhancement feature sequence and the fluctuation suppression feature sequence in parallel to obtain the trend lifetime prediction value and the fluctuation lifetime prediction value, and to extract the air volume stability index and dust concentration change rate from the operating condition data.

[0107] The weighted decision module 400 is used to perform fuzzy reasoning based on the air volume stability index, the dust concentration change rate, the trend lifetime prediction value and the fluctuation lifetime prediction value to obtain the weighted decision factors for heterogeneous prediction.

[0108] The fusion output module 500 is used to perform weighted fusion of the trend life prediction value and the fluctuation life prediction value according to the weight decision factor to obtain the life prediction value of the medium-efficiency dust filter.

[0109] This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this application. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0110] Those skilled in the art will understand that all or part of the steps in the various methods of the above embodiments can be implemented by a program instructing related hardware. The program can be stored in a computer-readable storage medium, including read-only memory (ROM), random access memory (RAM), programmable read-only memory (PROM), erasable programmable read-only memory (EPROM), one-time programmable read-only memory (OTPROM), electrically-Erasable Programmable Read-Only Memory (EEPROM), compactdisc read-only memory (CD-ROM) or other optical disc storage, disk storage, magnetic tape storage, or any other computer-readable medium capable of carrying or storing data.

[0111] It should also be noted that the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus. Unless otherwise specified, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes that element.

Claims

1. A method for predicting the service life of a medium efficiency dust filter, characterized by, The lifetime prediction method includes the following steps: Collect historical resistance sequences and operating condition data of medium-efficiency dust filters; The historical resistance sequences are subjected to cumulative slope calculation and sliding window filtering to obtain trend strengthening feature sequences and volatility suppression feature sequences. A heterogeneous prediction model is constructed, which includes a support vector machine regression model and a temporal convolutional network model. The trend enhancement feature sequence is input into the support vector machine regression model for kernel function mapping and regression analysis to obtain the trend lifetime prediction value. The fluctuation suppression feature sequence is input into the temporal convolutional network model for temporal feature convolution to obtain the fluctuation lifetime prediction value. The air volume stability index and dust concentration change rate are extracted from the operating condition data. Based on the air volume stability index, the dust concentration change rate, the trend lifetime prediction value, and the fluctuation lifetime prediction value, fuzzy reasoning is performed to obtain the weight decision factors for heterogeneous prediction. The predicted lifespan of the medium-efficiency dust filter is obtained by weighting and fusing the trend lifespan prediction value and the fluctuation lifespan prediction value according to the weight decision factor.

2. The method of claim 1, wherein the filter is a medium efficiency dust filter. The collection of historical resistance sequences and operating condition data for medium-efficiency dust filters specifically includes: Historical resistance sequences of medium-efficiency dust filters were collected using a micro differential pressure sensor. Operating condition data of medium-efficiency dust filters are collected through an IoT gateway.

3. The method for predicting the service life of a medium-efficiency dust filter as described in claim 1, characterized in that, The historical resistance sequences are subjected to cumulative slope calculation and sliding window filtering to obtain trend strengthening feature sequences and volatility suppression feature sequences, specifically including: Perform point-by-point cumulative slope calculation and linear regression analysis on the historical resistance sequence to obtain the trend strengthening characteristic sequence; By applying median filtering to the historical resistance sequence using a sliding window, a fluctuation suppression feature sequence is obtained.

4. The method for predicting the service life of a medium-efficiency dust filter as described in claim 1, characterized in that, Extracting the airflow stability index and dust concentration change rate from the aforementioned operating condition data specifically includes: Obtain the air volume data sequence from the operating condition data; The airflow stability index is determined by the coefficient of variation of the airflow data sequence; Obtain dust concentration data from the operating condition data, and perform sliding window linear regression analysis on the dust concentration data to obtain a slope sequence; The dust concentration change rate is determined based on the changing characteristics of the slope sequence.

5. The method for predicting the service life of a medium-efficiency dust filter as described in claim 1, characterized in that, Based on the airflow stability index, the dust concentration change rate, the trend lifetime prediction value, and the fluctuation lifetime prediction value, fuzzy inference is performed to obtain the weighted decision factors for heterogeneous prediction, which specifically include: The air volume stability index, the dust concentration change rate, the trend lifetime prediction value, and the fluctuation lifetime prediction value are fuzzified using a triangular membership function to obtain an input fuzzy set. The input fuzzy set and the preset fuzzy rule base are subjected to inference aggregation to obtain the output fuzzy set; The centroid method is used to defuzzify the output fuzzy set to obtain the trend model weight coefficients and the fluctuation model weight coefficients. The weighted decision factors for heterogeneous prediction are composed of the weighted coefficients of the trend model and the weighted coefficients of the volatility model.

6. The method for predicting the service life of a medium-efficiency dust filter as described in claim 1, characterized in that, The predicted lifespan of the medium-efficiency dust filter is obtained by weighting and fusing the trend lifespan prediction value and the fluctuation lifespan prediction value according to the weighted decision factors. Specifically, the predicted lifespan includes: The weight decision factors are vectorized to obtain a weight vector; A prediction value vector is determined using the trend lifetime prediction value and the fluctuation lifetime prediction value; The predicted lifespan of the medium-efficiency dust filter is determined based on the weight vector and the predicted value vector.

7. The method for predicting the service life of a medium-efficiency dust filter as described in claim 1, characterized in that, The weight decision factor for heterogeneous prediction is a binary tuple consisting of trend model weight coefficients and volatility model weight coefficients. The trend model weight coefficients are used to quantify the contribution of the trend lifetime prediction value to the heterogeneous prediction process, and the volatility model weight coefficients are used to quantify the contribution of the volatility lifetime prediction value to the heterogeneous prediction process.

8. The method for predicting the service life of a medium-efficiency dust filter as described in claim 1, characterized in that, The medium-efficiency dust filters include bag-type medium-efficiency dust filters and box-type medium-efficiency dust filters.

9. A system for predicting the service life of a medium-efficiency dust filter, used to execute a method for predicting the service life of a medium-efficiency dust filter as described in any one of claims 1 to 8, characterized in that, The lifetime prediction system includes: The data acquisition module is used to collect historical resistance sequences and operating condition data of medium-efficiency dust filters; The feature construction module is used to perform cumulative slope calculation and sliding window filtering on the historical resistance sequence to obtain trend strengthening feature sequence and volatility suppression feature sequence; The heterogeneous prediction module is used to perform heterogeneous prediction on the trend enhancement feature sequence and the fluctuation suppression feature sequence in parallel to obtain the trend lifetime prediction value and the fluctuation lifetime prediction value, and to extract the air volume stability index and dust concentration change rate from the operating condition data. The weighted decision module is used to perform fuzzy reasoning based on the air volume stability index, the dust concentration change rate, the trend lifetime prediction value, and the fluctuation lifetime prediction value to obtain the weighted decision factors for heterogeneous prediction. The fusion output module is used to perform weighted fusion of the trend lifetime prediction value and the fluctuation lifetime prediction value according to the weight decision factor to obtain the lifetime prediction value of the medium-efficiency dust filter.

Images