Fire-fighting smoke exhaust fan state detection method

Abstract

The invention discloses a fire-fighting smoke exhaust fan state detection method, and relates to the technical field of air conditioning, and the method specifically comprises the steps: obtaining the operation data of a fire-fighting smoke exhaust fan sample, and adding a category label to the operation data to obtain a first labeled data set; performing abnormal value removal based on labels in the first labeling data set to obtain a second labeling data set, obtaining a first feature data set through feature extraction, performing calculation based on the first feature data set to obtain composite features, and fusing the first feature data set and the composite features to obtain a second feature data set; performing normalization processing to obtain a third feature data set; sampling the third feature data set through a self-service sampling method to obtain a sampling sub-data set, constructing a decision tree through the sampling sub-data set and a random forest algorithm, and establishing a fire-fighting smoke exhaust fan state detection model; and acquiring operation data of a to-be-detected fire-fighting smoke exhaust fan, performing prediction through the fire-fighting smoke exhaust fan state detection model, and outputting a state result.

Description

Technical Field

[0001] The invention relates to the technical field of air conditioning, and in particular to a method for detecting the state of a fire smoke exhaust fan. Background Art

[0002] In recent years, the construction rate of high-rise buildings in many cities has begun to slow down, gradually shifting from the pursuit of quantity to the improvement of quality, and the attention paid to building fire safety issues has also been increasing. In confined places such as comprehensive commercial buildings, high-rise buildings and underground shopping malls, the thick smoke produced after a fire is the most important factor threatening the safety of people's lives. It not only obscures the vision and makes people lose their way in the fire, but also the smoke produced by the combustion contains a variety of toxic and harmful components, which can easily cause trapped people to suffocate and poison. Fire smoke exhaust fans have the function of quickly removing smoke and toxic gases in buildings, effectively reducing the hazards of fire, and have now become an important part of the building's fire protection system. Therefore, it is very important to ensure that the fire smoke exhaust fan is in good operating condition.

[0003] At present, the Chinese invention patent with application number CN201710398838.4 discloses a building fire fan monitoring system based on the Internet of Things. The application scheme includes: a fire fan, a manual regulating valve and an air outlet are arranged on the fire duct in sequence, and information collection points of differential pressure sensors are respectively set on the fire ducts on both sides of the fire fan; the signal collected by the differential pressure sensor is transmitted to the fire fan information monitoring device; at the same time, the signal collected by the fire fan is transmitted to the fire fan information monitoring device; the information of the fire fan information monitoring device is transmitted to the wind system information device; the information communication of the wind system information device is uploaded to the fire protection facility Internet of Things application platform. The system has the function of real-time monitoring of the state of the fire fan, the function of collecting the pressure difference before and after the fan and analyzing and judging it, and the function of uploading information and displaying it on the spot. However, the detection of the fire smoke exhaust fan in this scheme is only evaluated based on the wind pressure data collected by the differential pressure sensor. The evaluation index is single and easily affected by the environment. It fails to comprehensively and comprehensively examine the state of the fire smoke exhaust fan, which is not conducive to ensuring the accuracy of the state detection result. Summary of the invention

[0004] The technical problem solved by the present invention is that there is a lack of mature solutions for detecting the status of fire smoke exhaust fans on the market. Traditional methods only judge the status based on experience through a small amount of data, and fail to comprehensively examine the status of fire smoke exhaust fans, which is not conducive to ensuring the accuracy of the status detection results.

[0005] In order to solve the above technical problems, the present invention provides the following technical solutions: Step S1, obtaining the operation data of the fire smoke exhaust fan sample, and adding a category label to the operation data based on the state of the fire smoke exhaust fan sample to obtain a first labeled data set; Step S2, removing outliers based on the labels in the first annotated data set to obtain a second annotated data set, extracting features from the second annotated data set to obtain a first feature data set, calculating composite features based on the first feature data set, fusing the first feature data set with the composite features to obtain a second feature data set, and normalizing the second feature data set to obtain a third feature data set; Step S3, sampling the third feature data set by a self-service sampling method to obtain a sampling sub-data set, constructing a decision tree by using the sampling sub-data set and a random forest algorithm to establish a fire smoke exhaust fan state detection model; Step S4, collecting the operating data of the fire smoke exhaust fan to be detected, making predictions through the fire smoke exhaust fan state detection model, and outputting the state results.

[0006] As a preferred solution of the fire smoke exhaust fan state detection method described in the present invention, wherein: adding category labels to N groups of operation data of N fire smoke exhaust fan samples to obtain a first labeled data set, the category labels include good state, electrical fault, mechanical fault, electrical hidden danger and mechanical hidden danger; The operation data of the fire smoke exhaust fan sample includes air volume data, wind pressure data, sound data, current data, speed data, vibration data, temperature data, humidity data, dust data and historical data.

[0007] As a preferred solution of the fire smoke exhaust fan state detection method described in the present invention, wherein: data with a category label of good state in the first labeled data set is selected to form a first labeled data subset, and data with category labels of electrical fault, mechanical fault, electrical hidden danger and mechanical hidden danger in the first labeled data set is selected to form a second labeled data subset; The first anomaly value set is obtained by filtering outliers on the first annotated data subset through the interquartile range method, the second anomaly value set is obtained by filtering outliers on the second annotated data subset through a preset threshold interval, and the second anomaly value set is obtained by removing outliers from the first annotated data set based on the first anomaly value set and the second anomaly value set to obtain the second annotated data set.

[0008] As a preferred solution of a fire smoke exhaust fan state detection method described in the present invention, wherein: feature extraction is performed on the second annotation set to obtain a first feature data set, and the calculation logic of the feature extraction specifically includes: The median, maximum value and standard deviation of air volume are calculated through air volume data; The peak value, valley value, average value and standard deviation of wind pressure are calculated through wind pressure data; The average sound pressure level and standard deviation are calculated from the sound data, the sound data is subjected to high-pass filtering to remove environmental noise, the sound spectrum is obtained through fast Fourier transform processing, and the spectrum information entropy and spectrum flux are calculated based on the sound spectrum; The peak-to-peak value and kurtosis are calculated by vibration data, and the vibration spectrum is obtained by low-pass filtering and fast Fourier transform processing on the vibration data. The center frequency and bandwidth are calculated based on the vibration spectrum; The average value, standard deviation and zero-crossing rate of the current are calculated through the current data; The average, maximum and standard deviation of the speed are calculated through the speed data; The average, maximum and change rate of temperature are calculated through temperature data; The average value of ambient humidity is calculated through humidity data; The average value of air dust concentration is calculated through dust data; The historical data is one-hot encoded according to the maintenance within one month, maintenance within three months, and no maintenance for more than half a year to obtain a one-hot vector.

[0009] Among them, the calculation expressions of vibration kurtosis, average sound pressure level of sound and spectrum information entropy are:

[0010]

[0011]

[0012] in, represents the kurtosis of vibration, represents the i-th vibration acceleration, Mean vibration acceleration, represents the standard deviation of vibration acceleration, Indicates the total number of vibration acceleration data, i indicates vibration acceleration data, represents the average sound pressure level, T represents the total detection time, t represents the time point, represents the sound pressure intensity at time t, lg represents the logarithm with base 10, H represents the spectrum information entropy, Indicates The probability of a frequency component, Indicates the frequency component number, represents the total number of frequency components, Represents the base 2 logarithm.

[0013] As a preferred solution of a fire smoke exhaust fan state detection method described in the present invention, wherein: a composite feature is calculated based on the data of a first feature data set, and the composite feature is fused with the first feature data set to obtain a second feature data set, and the composite feature specifically includes: the product of sound spectrum flux and vibration kurtosis, the ratio of average sound pressure level to average speed, the ratio of median air volume to average wind pressure, and the product of average temperature and average current.

[0014] As a preferred solution of the fire smoke exhaust fan state detection method described in the present invention, the second feature data set is linearly mapped by Min-Max normalization method to obtain the third feature data set.

[0015] As a preferred solution of the fire smoke exhaust fan status detection method described in the present invention, wherein: the third feature data set is randomly sampled by the self-service sampling method to obtain K sampling data subsets and corresponding K out-of-bag data subsets, where K is the number of sampling data subsets, that is, the total number of decision trees generated.

[0016] As a preferred solution of a fire smoke exhaust fan state detection method described in the present invention, wherein: a first sampling data subset is selected from K sampling data subsets, an optimal splitting value and an optimal splitting feature group of a node are calculated based on the first sampling data subset, a preset number of feature types M and the Gini coefficient minimum principle, node splitting is performed based on the optimal splitting value and the optimal splitting feature group, and a first decision tree is obtained by recursive calculation of node splitting, wherein the optimal splitting value is the sampling data subset segmentation point corresponding to the minimum Gini coefficient, and the optimal splitting feature group is the M feature combination corresponding to the minimum Gini coefficient; A second sampling data subset, a third sampling data subset... and a Kth sampling data subset are selected from the K sampling data subsets in turn, and node splitting recursive calculations are performed respectively to obtain a second decision tree, a third decision tree... and a Kth decision tree, and a first random forest model is established through the obtained K decision trees.

[0017] As a preferred scheme of the fire smoke exhaust fan status detection method described in the present invention, wherein: the first random forest model is verified by an out-of-bag data subset, the out-of-bag error of the first random forest model is calculated, the optimal number of sampling data subsets is obtained by the incremental method, and a fire smoke exhaust fan status detection model is established based on the optimal number of sampling data subsets.

[0018] As a preferred scheme of the fire smoke exhaust fan status detection method described in the present invention, the operating data of the fire smoke exhaust fan to be detected is collected, including: air volume data, wind pressure data, sound data, current data, speed data, vibration data, temperature data, humidity data, dust data and historical data, and prediction is made through the voting results of each decision tree in the fire smoke exhaust fan status detection model to output the final detection result.

[0019] Beneficial effects of the present invention: The present invention collects information from multiple dimensions to establish an evaluation model for fire smoke exhaust fans, which is conducive to fully evaluating fire smoke exhaust fans. By further calculating the extracted features to obtain composite features, the correlation between different categories of influencing factors is magnified, which is conducive to reducing the impact of the collected error data and increasing the robustness of the evaluation system. The fire smoke exhaust fan state detection model is established by a random decision tree, which makes full use of a large amount of sample data and improves the accuracy of the detection results. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] Figure 1 A basic flow chart of a fire smoke exhaust fan status detection method provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0021] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the specific implementation methods of the present invention are described in detail below in conjunction with the drawings. It is obvious that the described embodiments are only part of the embodiments of the present invention, but not all of the embodiments.

[0022] Example 1, reference Figure 1 , as an embodiment of the present invention, provides a fire smoke exhaust fan state detection method, comprising: Step S1, obtaining the operation data of the fire smoke exhaust fan sample, and adding a category label to the operation data based on the state of the fire smoke exhaust fan sample to obtain a first labeled data set; Step S2, removing outliers based on the labels in the first annotated data set to obtain a second annotated data set, extracting features from the second annotated data set to obtain a first feature data set, calculating composite features based on the first feature data set, fusing the first feature data set with the composite features to obtain a second feature data set, and normalizing the second feature data set to obtain a third feature data set; Step S3, sampling the third feature data set by a self-service sampling method to obtain a sampling sub-data set, constructing a decision tree by using the sampling sub-data set and a random forest algorithm to establish a fire smoke exhaust fan state detection model; Step S4, collecting the operating data of the fire smoke exhaust fan to be detected, making predictions through the fire smoke exhaust fan state detection model, and outputting the state results.

[0023] In this embodiment, category labels are added to N groups of operation data of N fire smoke exhaust fan samples to obtain a first labeled data set, wherein the category labels include good status, electrical fault, mechanical fault, electrical hidden danger, and mechanical hidden danger; The operation data of the fire smoke exhaust fan sample includes air volume data, wind pressure data, sound data, current data, speed data, vibration data, temperature data, humidity data, dust data and historical data.

[0024] In this embodiment, data with the category label of good status in the first labeled data set are selected to form a first labeled data subset, and data with the category labels of electrical fault, mechanical fault, electrical hidden danger and mechanical hidden danger in the first labeled data set are selected to form a second labeled data subset; The first anomaly value set is obtained by filtering outliers on the first annotated data subset through the interquartile range method, the second anomaly value set is obtained by filtering outliers on the second annotated data subset through a preset threshold interval, and the second anomaly value set is obtained by removing outliers from the first annotated data set based on the first anomaly value set and the second anomaly value set to obtain the second annotated data set.

[0025] In this embodiment, feature extraction is performed on the second annotation set to obtain a first feature data set, and the calculation logic of the feature extraction specifically includes: The median, maximum value and standard deviation of air volume are calculated through air volume data; The peak value, valley value, average value and standard deviation of wind pressure are calculated through wind pressure data; The average sound pressure level and standard deviation are calculated from the sound data, the sound data is subjected to high-pass filtering to remove environmental noise, the sound spectrum is obtained through fast Fourier transform processing, and the spectrum information entropy and spectrum flux are calculated based on the sound spectrum; The peak-to-peak value and kurtosis are calculated by vibration data, and the vibration spectrum is obtained by low-pass filtering and fast Fourier transform processing on the vibration data. The center frequency and bandwidth are calculated based on the vibration spectrum; The average value, standard deviation and zero-crossing rate of the current are calculated through the current data; The average, maximum and standard deviation of the speed are calculated through the speed data; The average, maximum and change rate of temperature are calculated through temperature data; The average value of ambient humidity is calculated through humidity data; The average value of air dust concentration is calculated through dust data; The historical data is one-hot encoded according to the maintenance within one month, maintenance within three months, and no maintenance for more than half a year to obtain a one-hot vector.

[0026] Among them, the calculation expressions of vibration kurtosis, average sound pressure level of sound and spectrum information entropy are:

[0027]

[0028]

[0029] in, represents the kurtosis of vibration, represents the i-th vibration acceleration, Mean vibration acceleration, represents the standard deviation of vibration acceleration, Indicates the total number of vibration acceleration data, i indicates vibration acceleration data, represents the average sound pressure level, T represents the total detection time, t represents the time point, represents the sound pressure intensity at time t, lg represents the logarithm with base 10, H represents the spectrum information entropy, Indicates The probability of a frequency component, Indicates the frequency component number, represents the total number of frequency components, Represents the base 2 logarithm.

[0030] In this embodiment, a composite feature is calculated based on the data of the first feature data set, and the composite feature is fused with the first feature data set to obtain a second feature data set. The composite feature specifically includes: the product of sound spectrum flux and vibration kurtosis, the ratio of average sound pressure level to average rotation speed, the ratio of median wind volume to average wind pressure, and the product of average temperature and average current.

[0031] As a preferred solution of the fire smoke exhaust fan state detection method described in the present invention, the second feature data set is linearly mapped by Min-Max normalization method to obtain the third feature data set.

[0032] In this embodiment, the third feature data set is randomly sampled by a self-service sampling method to obtain K sampled data subsets and corresponding K out-of-bag data subsets, where K is the number of sampled data subsets, that is, the total number of decision trees generated.

[0033] In this embodiment, a first sampled data subset is selected from K sampled data subsets, and the optimal splitting value and the optimal splitting feature group of the node are calculated based on the first sampled data subset, the preset number of feature types M and the Gini coefficient minimum principle, and the node is split based on the optimal splitting value and the optimal splitting feature group. The first decision tree is obtained by recursive calculation of node splitting, wherein the optimal splitting value is the sampled data subset segmentation point corresponding to the minimum Gini coefficient, and the optimal splitting feature group is the M feature combination corresponding to the minimum Gini coefficient; Among them, the total number of feature types is 27, including the median, maximum value and standard deviation of the wind volume; the peak value, valley value, average value and standard deviation of the wind pressure; the average sound pressure level, standard deviation, spectral information entropy and spectral flux of the sound; the average value, standard deviation and zero-crossing rate of the peak-to-peak value, kurtosis, center frequency and bandwidth current of the vibration; the average value, maximum value and standard deviation of the rotation speed; the average value, maximum value and change rate of the temperature; the average value of the ambient humidity; the average value of the air dust concentration; the unique hot vectors representing maintenance within one month, maintenance within three months, and no maintenance for more than half a year; The preset number of categories M is 5, that is, 5 feature categories are randomly selected from the 27 feature categories, and the data corresponding to the 5 feature categories are selected from the first sampling data subset to calculate the Gini coefficient value, so that the split value and split variable when the Gini coefficient takes the minimum value are the optimal split value and optimal split feature group of the node; The splitting variables are the five randomly selected feature types, and the splitting values ​​are the splitting feature groups for the five selected feature types, and the specific splitting points are determined to divide the data into two new sub-nodes. Perform node splitting recursive calculation, split each child node to obtain two corresponding new nodes, stop node splitting when the data of the first sample data subset is exhausted, and obtain a decision tree generated by the first sample data subset.

[0034] Select the second sampling data subset, the third sampling data subset... and the Kth sampling data subset from the K sampling data subsets in turn, repeat the above calculations respectively, obtain the second decision tree, the third decision tree... and the Kth decision tree through node splitting recursive calculation, and establish the first random forest model through the obtained K decision trees.

[0035] In this embodiment, the first random forest model is verified by using the out-of-bag data subset, the out-of-bag error of the first random forest model is calculated, the optimal number of sampling data subsets is obtained by the incremental method, and a fire smoke exhaust fan status detection model is established based on the optimal number of sampling data subsets.

[0036] The specific steps of the incremental method are as follows: assigning the number of sampled data subsets K to 10 according to the preset initial sampling value, calculating the first state detection model, and calculating the first out-of-bag error corresponding to the first random forest model, then incrementing the number of sampled data subsets K, assigning the number of sampled data subsets K to 20, and calculating the second out-of-bag error corresponding to the second state detection model; The number of sampling data subsets K assigned is increased by 10 successively, and the calculation is repeated until the number of sampling data subsets K is assigned a value of 200, the third out-of-bag error, the fourth out-of-bag error, ... and the twentieth out-of-bag error, all out-of-bag errors are compared, and the sampling data subset number K corresponding to the minimum out-of-bag error is selected as the optimal sampling data subset number.

[0037] In this embodiment, the operating data of the fire smoke exhaust fan to be tested is collected, including: air volume data, wind pressure data, sound data, current data, speed data, vibration data, temperature data, humidity data, dust data and historical data, and prediction is made through the voting results of each decision tree in the fire smoke exhaust fan state detection model to output the final detection result.

[0038] It should be understood by those skilled in the art that the embodiments of the present invention may be provided as methods, systems or computer program products. Therefore, the present invention may take the form of a complete hardware embodiment, a complete software embodiment or an embodiment combining software and hardware aspects. Moreover, the present invention may take the form of a computer program product implemented on one or more computer-usable storage media containing computer-usable program codes. Among them, the storage medium may be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (Static Random Access Memory, referred to as SRAM), electrically erasable programmable read-only memory (Electrically Erasable Programmable Read-Only Memory, referred to as EEPROM), erasable programmable read-only memory (Erasable Programmable Read Only Memory, referred to as EPROM), programmable read-only memory (Programmable Red-Only Memory, referred to as PROM), read-only memory (Read-Only Memory, referred to as ROM), magnetic memory, flash memory, magnetic disk or optical disk. These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1A function specified in one or more boxes.

[0039] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical solutions of the present invention, which should all be included in the scope of the claims of the present invention.

Claims

1. A fire smoke exhaust fan state detection method, characterized in that: include: Step S1, obtaining the operation data of the fire smoke exhaust fan sample, and adding a category label to the operation data based on the state of the fire smoke exhaust fan sample to obtain a first labeled data set; Step S2, removing outliers based on the labels in the first annotated data set to obtain a second annotated data set, extracting features from the second annotated data set to obtain a first feature data set, calculating composite features based on the first feature data set, fusing the first feature data set with the composite features to obtain a second feature data set, and normalizing the second feature data set to obtain a third feature data set; Step S3, sampling the third feature data set by a self-service sampling method to obtain a sampling sub-data set, constructing a decision tree by using the sampling sub-data set and a random forest algorithm to establish a fire smoke exhaust fan state detection model; Step S4, collecting the operating data of the fire smoke exhaust fan to be detected, making predictions through the fire smoke exhaust fan state detection model, and outputting the state results.

2. A fire smoke exhaust fan state detection method according to claim 1, characterized in that: Adding category labels to N groups of operation data of N fire smoke exhaust fan samples to obtain a first labeled data set, wherein the category labels include good status, electrical fault, mechanical fault, electrical hidden danger, and mechanical hidden danger; The operation data of the fire smoke exhaust fan sample includes air volume data, wind pressure data, sound data, current data, speed data, vibration data, temperature data, humidity data, dust data and historical data.

3. A fire smoke exhaust fan state detection method according to claim 1, characterized in that: The data with the category label of "good status" in the first labeled data set are selected to form a first labeled data subset, and the data with the category labels of "electrical failure", "mechanical failure", "electrical hidden danger" and "mechanical hidden danger" in the first labeled data set are selected to form a second labeled data subset; The first anomaly value set is obtained by filtering outliers on the first annotated data subset through the interquartile range method, the second anomaly value set is obtained by filtering outliers on the second annotated data subset through a preset threshold interval, and the second anomaly value set is obtained by removing outliers from the first annotated data set based on the first anomaly value set and the second anomaly value set to obtain the second annotated data set.

4. A fire smoke exhaust fan state detection method according to claim 1, characterized in that: Performing feature extraction on the second annotation set to obtain a first feature data set, wherein the calculation logic of the feature extraction specifically includes: The median, maximum value and standard deviation of air volume are calculated through air volume data; The peak value, valley value, average value and standard deviation of wind pressure are calculated through wind pressure data; The average sound pressure level and standard deviation are calculated from the sound data, the sound data is subjected to high-pass filtering to remove environmental noise, the sound spectrum is obtained through fast Fourier transform processing, and the spectrum information entropy and spectrum flux are calculated based on the sound spectrum; The peak-to-peak value and kurtosis are calculated by vibration data, and the vibration spectrum is obtained by low-pass filtering and fast Fourier transform processing on the vibration data. The center frequency and bandwidth are calculated based on the vibration spectrum; The average value, standard deviation and zero-crossing rate of the current are calculated through the current data; The average, maximum and standard deviation of the speed are calculated through the speed data; The average, maximum and change rate of temperature are calculated through temperature data; The average value of ambient humidity is calculated through humidity data; The average value of air dust concentration is calculated through dust data; The historical data is one-hot encoded according to the maintenance within one month, maintenance within three months, and no maintenance for more than half a year to obtain a one-hot vector.

5. A fire smoke exhaust fan state detection method according to claim 1, characterized in that: A composite feature is calculated based on the data of the first feature data set, and the composite feature is fused with the first feature data set to obtain a second feature data set. The composite feature specifically includes: the product of sound spectrum flux and vibration kurtosis, the ratio of average sound pressure level to average rotation speed, the ratio of median wind volume to average wind pressure, and the product of average temperature and average current.

6. A fire smoke exhaust fan status detection method according to claim 1, characterized in that: The second feature data set is linearly mapped using the Min-Max normalization method to obtain the third feature data set.

7. A fire smoke exhaust fan state detection method according to claim 1, characterized in that: The third feature data set is randomly sampled by the self-service sampling method to obtain K sampling data subsets and corresponding K out-of-bag data subsets, where K is the number of sampling data subsets.

8. A fire smoke exhaust fan state detection method according to claim 7, characterized in that: Select a first sampled data subset from the K sampled data subsets, calculate the optimal splitting value and the optimal splitting feature group of the node based on the first sampled data subset, the preset number of feature types M and the Gini coefficient minimum principle, perform node splitting based on the optimal splitting value and the optimal splitting feature group, and obtain a first decision tree through node splitting recursively, wherein the optimal splitting value is the sampled data subset splitting point corresponding to the minimum Gini coefficient, and the optimal splitting feature group is the M feature combination corresponding to the minimum Gini coefficient; A second sampling data subset, a third sampling data subset... and a Kth sampling data subset are selected from the K sampling data subsets in turn, and node splitting recursive calculations are performed respectively to obtain a second decision tree, a third decision tree... and a Kth decision tree, and a first random forest model is established through the obtained K decision trees.

9. A fire smoke exhaust fan state detection method according to claim 8, characterized in that: The first random forest model was verified by using the out-of-bag data subset, and the out-of-bag error of the first random forest model was calculated. The optimal number of sampling data subsets was obtained by the incremental method, and a fire smoke exhaust fan status detection model was established based on the optimal number of sampling data subsets.

10. A fire smoke exhaust fan status detection method according to claim 1, characterized in that: Collect the operating data of the fire smoke exhaust fan to be tested, including: air volume data, wind pressure data, sound data, current data, speed data, vibration data, temperature data, humidity data, dust data and historical data, and make predictions through the voting results of each decision tree in the fire smoke exhaust fan status detection model to output the final detection results.

Images