CN117150298A - Deep learning-based subway FAS fire alarm system debugging method - Google Patents
Deep learning-based subway FAS fire alarm system debugging method Download PDFInfo
- Publication number
- CN117150298A CN117150298A CN202311122909.XA CN202311122909A CN117150298A CN 117150298 A CN117150298 A CN 117150298A CN 202311122909 A CN202311122909 A CN 202311122909A CN 117150298 A CN117150298 A CN 117150298A
- Authority
- CN
- China
- Prior art keywords
- fire
- deep learning
- data
- learning model
- subway
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
- 238000000034 method Methods 0.000 title claims abstract description 56
- 238000013135 deep learning Methods 0.000 title claims abstract description 32
- 238000013136 deep learning model Methods 0.000 claims abstract description 117
- 230000006870 function Effects 0.000 claims abstract description 38
- 238000012544 monitoring process Methods 0.000 claims abstract description 17
- 230000004044 response Effects 0.000 claims abstract description 9
- 238000012360 testing method Methods 0.000 claims description 56
- 238000012549 training Methods 0.000 claims description 51
- 238000012795 verification Methods 0.000 claims description 23
- 239000000779 smoke Substances 0.000 claims description 18
- 238000005457 optimization Methods 0.000 claims description 15
- 230000008859 change Effects 0.000 claims description 13
- 238000004364 calculation method Methods 0.000 claims description 11
- 230000000694 effects Effects 0.000 claims description 10
- 238000011156 evaluation Methods 0.000 claims description 10
- 238000004422 calculation algorithm Methods 0.000 claims description 8
- 230000008569 process Effects 0.000 claims description 8
- ORILYTVJVMAKLC-UHFFFAOYSA-N Adamantane Natural products C1C(C2)CC3CC1CC2C3 ORILYTVJVMAKLC-UHFFFAOYSA-N 0.000 claims description 4
- 238000013528 artificial neural network Methods 0.000 claims description 3
- 238000012937 correction Methods 0.000 claims description 3
- 238000007781 pre-processing Methods 0.000 claims description 3
- 238000001514 detection method Methods 0.000 abstract description 22
- 238000013461 design Methods 0.000 abstract description 2
- 238000005070 sampling Methods 0.000 abstract description 2
- 230000009286 beneficial effect Effects 0.000 description 4
- 238000003909 pattern recognition Methods 0.000 description 3
- 238000010200 validation analysis Methods 0.000 description 3
- 206010000369 Accident Diseases 0.000 description 1
- 230000002159 abnormal effect Effects 0.000 description 1
- 230000006978 adaptation Effects 0.000 description 1
- 230000004075 alteration Effects 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 238000010586 diagram Methods 0.000 description 1
- 230000002349 favourable effect Effects 0.000 description 1
- 230000006872 improvement Effects 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 230000003449 preventive effect Effects 0.000 description 1
- 238000012545 processing Methods 0.000 description 1
- 230000035945 sensitivity Effects 0.000 description 1
- 238000000638 solvent extraction Methods 0.000 description 1
- 230000007480 spreading Effects 0.000 description 1
- 238000006467 substitution reaction Methods 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/213—Feature extraction, e.g. by transforming the feature space; Summarisation; Mappings, e.g. subspace methods
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
- G06N3/084—Backpropagation, e.g. using gradient descent
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
- G06N3/0985—Hyperparameter optimisation; Meta-learning; Learning-to-learn
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- Life Sciences & Earth Sciences (AREA)
- Artificial Intelligence (AREA)
- General Physics & Mathematics (AREA)
- Evolutionary Computation (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Bioinformatics & Computational Biology (AREA)
- Evolutionary Biology (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Health & Medical Sciences (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- General Health & Medical Sciences (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Fire Alarms (AREA)
Abstract
The invention discloses a debugging method of a subway FAS fire alarm system based on deep learning, relates to the technical field of subway fire alarm system debugging, improves subway fire early warning and safety monitoring capability, and solves some problems of a traditional subway fire alarm system, such as high false alarm rate and untimely response. By adopting the deep learning technology and combining the image data and the sensor data, the system can more accurately detect the fire disaster in the subway carriage and alarm in time, thereby effectively guaranteeing the life safety of passengers and staff. The deep learning model facilitates better sample distribution imbalance between the benefit classes when dealing with unbalanced data sets of minority classes (fire data) and majority classes (normal data). Through reasonable sampling strategy and loss function design of the data set, the problem caused by unbalanced data can be overcome, and the detection capability of fire data is improved, so that potential fire hazards can be found more effectively.
Description
Technical Field
The invention relates to the technical field of subway fire alarm system debugging, in particular to a method for debugging a subway FAS fire alarm system based on deep learning.
Background
The subway FAS refers to an abbreviation of a subway fire alarm system (SubwayFireAlarmSystem). The fire disaster detection and alarm system is applied to subway carriages and stations, and aims to improve the timely identification and response capability of the subway to fire disaster events during operation of the subway so as to ensure the safety of passengers and staff. Conventional subway fire alarm systems generally use a rule-based method, and rely on preset thresholds and rules for fire detection. However, this method often performs poorly for complex subway environments and fire scenes, and is prone to false positives or false negatives. The deep learning technology has excellent feature learning capability and pattern recognition capability, and can automatically learn and extract complex fire features, so that the accuracy and reliability of fire detection can be promoted to be improved.
In the existing deep learning model, in subway fire data, the subway fire data are often in a few categories, and the normal operation data are in a plurality of categories, so that unbalanced distribution of data sets is caused. This may make the model more prone to predicting normal conditions, and less effective for predicting fire conditions.
Disclosure of Invention
(one) solving the technical problems
Aiming at the defects of the prior art, the invention provides a subway FAS fire alarm system debugging method based on deep learning, which aims to improve the subway fire early warning and safety monitoring capability and solve some problems of the traditional subway fire alarm system, such as high false alarm rate, untimely response and the like. By adopting the deep learning technology and combining the image data and the sensor data, the system can more accurately detect the fire disaster in the subway carriage and alarm in time, thereby effectively guaranteeing the life safety of passengers and staff.
(II) technical scheme
In order to achieve the above purpose, the invention is realized by the following technical scheme: the debugging method of the subway FAS fire alarm system based on deep learning comprises the following steps,
collecting subway line, carriage, different time periods and weather condition sample data of different subway lines in a time axis mode, and collecting video and sensor sample data when the subway normally operates and when fire records occur; recorded as dataset a;
marking the data set A, marking the data of normal conditions and historical fire conditions, and marking the data including the time stamp, the position, the fire scale and related influence information of the occurrence of the fire;
after preprocessing the data set A, dividing the data set A into a training set B, a verification set C and a test set D;
extracting fire features from a fire alarm system of a training set B, calculating and obtaining a safety important task coefficient rwx, a data set complex index fz and a resource depth coefficient sd, fitting the safety important task coefficient rwx, the data set complex index fz and the resource depth coefficient sd to obtain a complex network structure coefficient fzwx, selecting a corresponding network structure and layer number model according to the complex network structure coefficient fzwx, and setting a definition loss function to construct a deep learning model;
the complex network structure coefficient fzwx is obtained by the following formula:
wherein Xn is expressed as recall, zq is expressed as accuracy, F1 is expressed as F1 score evaluation index, ypzl is expressed as total number of sample data in the set;
training the deep learning model by using a training set B, and adopting an optimization algorithm comprising one of SGD and Adam to continuously update parameters of the deep learning model to gradually converge, wherein in the training process, a verification set C is used for verifying the deep learning model; calculating recall rate Xn and accuracy zq of the deep learning model on the verification set C; calculating a monitoring performance coefficient Jkx according to the recall rate Xn and the accuracy rate zq, and adjusting the super-parameter configuration according to the monitoring performance coefficient Jkx;
evaluating the performance of the deep learning model using the test set D; the specific method comprises the steps of calculating recall rate Xn, accuracy zq and F1 fraction evaluation indexes of the deep learning model on a test set D, obtaining a test performance coefficient Csx, comparing the test performance coefficient Csx with a monitoring performance coefficient Jkx, obtaining a difference value Diff, and debugging a corresponding scheme aiming at the difference value Diff.
Preferably, the data set A comprises an image data set A1, a sensor data set A2 and historical fire data A3, and a camera and an infrared sensor are installed to collect image data in a subway carriage and marked as image data A1;
a temperature sensor, a smoke sensor, a gas sensor and a sensor data A2 are adopted to collect temperature data, smoke data and combustible gas leakage data in a carriage;
and collecting historical fire record report data which is published by related departments of the subway system through big data, wherein the historical fire record report data comprises fire reasons, fire spread conditions, alarm times and emergency treatment data, and is marked as historical fire data A3.
Preferably, the data set A is divided into a training set B, a verification set C and a test set D in proportions of 70%, 15% and 15%;
and extracting fire features from the training set B fire alarm system, wherein the fire features comprise fire wire and dense smoke area intensity features, fire light intensity change features, target carriage internal object deformation features, carriage internal temperature change features, smoke concentration change features, combustible gas leakage features and fire spread change features.
Preferably, the specific method for defining the loss function is as follows:
s1, for a fire alarm system, setting the probability of fire occurrence or the classification result of fire occurrence as the output of a deep learning model; the real tag is marked according to the information in the historical fire data A3;
s2, selecting a set loss function to measure the difference between the predicted value and the true value according to the output of the deep learning model and the label type; for the second classification, including fire and non-fire classification, outputting by adopting a binary cross entropy loss function, and if the output is a probability value of fire occurrence, setting and calculating by using a mean square error loss function;
s3, inputting the output of the deep learning model and the corresponding real label into a loss function, and calculating to obtain a loss value of the deep learning model on a current sample; for all samples in the training set B, calculating average loss or total loss as the loss of the deep learning model on the whole training set B;
and S4, taking the loss function as an optimization target, and adjusting parameters of the deep learning model by using a gradient descent optimization algorithm in the training process, so that the loss function value is gradually reduced, the deep learning model is better fitted with training data, and a better fire disaster prediction effect is achieved.
Preferably, the safety-critical task coefficient rwx is obtained by calculating the following formula:
where rws represents the number of important tasks that the fire alarm system needs to perform, including the number of tasks that detect flame, smoke fire characteristics; aqx is expressed as the degree of safety of the fire, including the coverage of the fire precautions and the number of emergency responses; mygl is expressed as a spread probability value; α, β, γ are weight coefficients of rws, aqx, and mygl, and are set by the user, and α+β+γ=1.0.
Preferably, the dataset complexity index fz is calculated by the following formula:
fz＝(N*Ce)/(C1*C2)
where N represents the total number of samples in the dataset; ce represents a class balance index of the data set, and is used for measuring the balance degree of different class samples; metrics are measured using statistical indicators, including Gini coefficients and information entropy; c1 and C2 are adjustment coefficients representing the number of samples of the data set and the data distribution, respectively.
Preferably, the resource depth coefficient sd is obtained by calculating the following formula:
wherein ccL is expressed as a parameter quantity value of a deep learning model, and in the neural network, parameters refer to weight and bias, and the greater the parameter quantity, the higher the complexity of the deep learning model; xql represents the computational resource memory requirements required for training and deep learning models; jjxl is expressed as computational efficiency, which includes computational time and computational power ratio.
Preferably, the recall Xn, the accuracy zq, and the monitoring coefficient of performance Jkx are calculated by the following formula:
Xn＝TP/(TP+FN)
zq＝TP/(TP+FP)
wherein TP represents the number of samples for which the deep learning model correctly predicts as fire, and FN represents the number of samples for which the deep learning model incorrectly predicts as non-fire; FP represents the number of samples that the deep learning model mispredicts as a fire; f1 and f2 are coefficients for balancing weights between recall Xn and accuracy zq; the meaning of the parameters is that the recall rate Xn and the accuracy rate zq are balanced, if the higher recall rate Xn is more important, namely the fire disaster situation is not leaked as much as possible, the increase f1 is set, and if the higher accuracy rate zq is more important, namely the false alarm is avoided as much as possible, the increase f2 is set.
Preferably, the test coefficient of performance Csx is calculated by the following formula:
wherein F1 is expressed as F1 fraction on the test set D, and is used for comprehensively considering the balance of recall rate and accuracy rate; the F1 score is a harmonic mean value of the recall rate Xn and the accuracy rate zq and is used for measuring the comprehensive performance of the deep learning model; w1, w2, w3 are represented as weight coefficient values of F1 score, recall Xn, and accuracy zq, and w1+w2+w3=1.0; r represents a correction constant.
Preferably, the difference Diff between the test coefficient of performance Csx and the monitor coefficient of performance Jkx is calculated, i.e., diff=csx-Jkx; according to the magnitude and direction of the difference value Diff, judging whether the performance of the deep learning model on the test set D is better than the performance of the deep learning model on the verification set C or not; a large positive difference indicates that the deep learning model performs better on test set D, and a large negative difference may indicate that the deep learning model performs worse on test set D;
according to the result of the difference value Diff, corresponding scheme debugging is carried out; if the test performance coefficient Csx is better, namely Csx > Jkx, the performance of the deep learning model on the test set D is better than that on the verification set C, and the deep learning model is considered to be continuously used; if the test performance coefficient Csx is poor, i.e., csx < Jkx, then it is necessary to re-optimize the super-parameters, adjust the deep learning model structure, or add more data to improve the deep learning model.
(III) beneficial effects
The invention provides a debugging method of a subway FAS fire alarm system based on deep learning. The beneficial effects are as follows:
(1) According to the debugging method of the subway FAS fire alarm system based on deep learning, the deep learning technology has excellent feature learning and pattern recognition capability, and complex fire features can be automatically learned and extracted, so that the influence of manually set thresholds and rules on a fire detection result is reduced. Compared with the traditional rule-based method, the subway FAS based on deep learning can more accurately identify the fire situation, and reduce the risks of false alarm and missing report.
(2) According to the debugging method of the subway FAS fire alarm system based on deep learning, subway environments are complex and changeable, and the deep learning model can adaptively learn and fit fire characteristics in different scenes through a large amount of sample data, so that the method has stronger adaptability and robustness. Thus, even in the face of new or unusual fire situations, subway FAS based on deep learning can better cope with and improve the effect of fire detection.
(3) According to the debugging method of the subway FAS fire alarm system based on deep learning, when the deep learning model processes unbalanced data sets of minority classes (fire data) and majority classes (normal data), sample distribution imbalance among the classes can be better processed. Through reasonable sampling strategy and loss function design to the data set, subway FAS based on deep learning can overcome the problem caused by data unbalance, and improve the fire data detection capability, so that potential fire hazards can be found more effectively.
(4) According to the debugging method of the subway FAS fire alarm system based on the deep learning, the difference between the test performance coefficient Csx and the monitoring performance coefficient Jkx is compared and analyzed, the subway FAS based on the deep learning can find out the advantages and disadvantages of the model in time, and corresponding scheme debugging is performed according to the result. This allows for continuous optimization and improvement of the system, continuously increasing the effectiveness and reliability of fire detection.
Drawings
FIG. 1 is a schematic diagram of a structural calculation flow of a deep learning model of the method of the invention;
Detailed Description
The following description of the embodiments of the present invention will be made clearly and completely with reference to the accompanying drawings, in which it is apparent that the embodiments described are only some embodiments of the present invention, but not all embodiments. All other embodiments, which can be made by those skilled in the art based on the embodiments of the invention without making any inventive effort, are intended to be within the scope of the invention.
The subway FAS refers to an abbreviation of a subway fire alarm system (SubwayFireAlarmSystem). The fire disaster detection and alarm system is applied to subway carriages and stations, and aims to improve the timely identification and response capability of the subway to fire disaster events during operation of the subway so as to ensure the safety of passengers and staff. Conventional subway fire alarm systems generally use a rule-based method, and rely on preset thresholds and rules for fire detection. However, this method often performs poorly for complex subway environments and fire scenes, and is prone to false positives or false negatives. The deep learning technology has excellent feature learning capability and pattern recognition capability, and can automatically learn and extract complex fire features, so that the accuracy and reliability of fire detection can be promoted to be improved.
In the existing deep learning model, in subway fire data, the subway fire data are often in a few categories, and the normal operation data are in a plurality of categories, so that unbalanced distribution of data sets is caused. This may make the model more prone to predicting normal conditions, and less effective for predicting fire conditions.
Example 1
The invention provides a debugging method of a subway FAS fire alarm system based on deep learning, referring to FIG. 1, comprising the following steps,
collecting subway line, carriage, different time periods and weather condition sample data of different subway lines in a time axis mode, and collecting video and sensor sample data when the subway normally operates and when fire records occur; recorded as dataset a;
marking the data set A, marking the data of normal conditions and historical fire conditions, and marking the data including the time stamp, the position, the fire scale and related influence information of the occurrence of the fire;
after preprocessing the data set A, dividing the data set A into a training set B, a verification set C and a test set D; the data set A comprises an image data set A1, a sensor data set A2 and historical fire data A3, and a camera and an infrared sensor are installed to collect image data in a subway carriage and marked as image data A1;
a temperature sensor, a smoke sensor, a gas sensor and a sensor data A2 are adopted to collect temperature data, smoke data and combustible gas leakage data in a carriage;
and collecting historical fire record report data which is published by related departments of the subway system through big data, wherein the historical fire record report data comprises fire reasons, fire spread conditions, alarm times and emergency treatment data, and is marked as historical fire data A3.
Extracting fire features from a fire alarm system of a training set B, calculating and obtaining a safety important task coefficient rwx, a data set complex index fz and a resource depth coefficient sd, fitting the safety important task coefficient rwx, the data set complex index fz and the resource depth coefficient sd to obtain a complex network structure coefficient fzwx, selecting a corresponding network structure and layer number model according to the complex network structure coefficient fzwx, and setting a definition loss function to construct a deep learning model;
the complex network structure coefficient fzwx is obtained by the following formula:
wherein Xn is expressed as recall, zq is expressed as accuracy, F1 is expressed as F1 score evaluation index, ypzl is expressed as total number of sample data in the set;
training the deep learning model by using a training set B, and adopting an optimization algorithm comprising one of SGD and Adam to continuously update parameters of the deep learning model to gradually converge, wherein in the training process, a verification set C is used for verifying the deep learning model; calculating recall rate Xn and accuracy zq of the deep learning model on the verification set C; calculating a monitoring performance coefficient Jkx according to the recall rate Xn and the accuracy rate zq, and adjusting the super-parameter configuration according to the monitoring performance coefficient Jkx; the recall rate Xn and the accuracy rate zq on the verification set C are calculated, and the monitoring performance coefficient Jkx is calculated according to the recall rate Xn and the accuracy rate zq, so that the super parameters can be adjusted and optimized according to actual conditions, and the performance of the deep learning model is improved;
evaluating the performance of the deep learning model using the test set D; the specific method comprises the steps of calculating recall rate Xn, accuracy zq and F1 fraction evaluation indexes of the deep learning model on a test set D, obtaining a test performance coefficient Csx, comparing the test performance coefficient Csx with a monitoring performance coefficient Jkx, obtaining a difference value Diff, and debugging a corresponding scheme aiming at the difference value Diff.
In the embodiment, by using the deep learning model, the method can learn complex fire characteristics from a large amount of data and accurately identify fire signs such as flame, smoke, abnormal temperature and the like, thereby reducing false alarm rate and false missing report rate and improving the accuracy rate of fire detection; by combining the image data and the sensor data, the deep learning model can comprehensively consider various fire characteristics such as flame intensity, smoke density, temperature change and the like, so that the fire risk is more comprehensively estimated, and the accuracy and reliability of fire early warning are improved;
the deep learning model has a faster reasoning speed, can perform rapid fire detection and early warning under the condition of higher real-time requirement, and timely gives an alarm, so that subway management departments and passengers can make emergency response more rapidly, and the minimization of fire accidents is ensured;
the deep learning model has automatic learning and adaptation capability, can adjust model parameters in real time according to different subway lines, carriages, weather conditions and the like, adapts to fire detection requirements of different environments, and improves the stability and reliability of the system; according to the method, an optimization algorithm such as SGD and Adam is used for training a deep learning model, so that the deep learning model gradually converges, and meanwhile, a proper network structure and layer number model are selected, so that efficient model training and reasoning can be realized under relatively less computing resources;
the debugging method of the subway FAS fire alarm system based on deep learning can effectively solve the problems that in the prior art, fire detection accuracy is not high, instantaneity is not enough, various fire characteristics cannot be comprehensively considered, and the like, improves performance and stability of the subway fire alarm system, ensures subway operation safety, and plays a positive promotion role in life safety of passengers and staff.
By dividing the data set a proportionally into training set B, validation set C and test set D, it is ensured that sufficient fire and normal samples are contained in each subset. Through reasonable division proportion, the balance of the data set can be maintained, and better learning and prediction of the model are facilitated. When defining the loss function, the weighted loss function is adopted to give different weights to the fire sample and the normal sample. A high weight is given to a few classes (fire samples), so that the model is more concerned about fire conditions, and the prediction effect on the fire is improved.
Example 2, which is an explanation made in example 1, specifically, the data set a is divided into the training set B, the validation set C, and the test set D in proportions of 70%, 15%, and 15%; the data set A is divided into the training set B, the verification set C and the test set D according to the proportion of 70%, 15% and 15%, so that the data can be fully utilized, data leakage is avoided, model optimization and parameter adjustment are carried out, and the performance of the model is comprehensively evaluated, so that the detection accuracy and performance of the subway FAS fire alarm system are improved.
And extracting fire features from the training set B fire alarm system, wherein the fire features comprise fire wire and dense smoke area intensity features, fire light intensity change features, target carriage internal object deformation features, carriage internal temperature change features, smoke concentration change features, combustible gas leakage features and fire spread change features.
In this embodiment, by dividing the data set a into different subsets, samples of the training set B, the verification set C, and the test set D, which are not overlapped, are ensured, the problem of data leakage is avoided, and accuracy of model evaluation is ensured. Training of the deep learning model is performed using training set B, and model verification and tuning are performed using verification set C. The partitioning mode allows the model to adjust the super parameters and the model structure according to the performance of the verification set C in the training process, so that the performance of the model is optimized; and performing performance evaluation on the deep learning model by using the test set D, and calculating recall rate Xn, accuracy zq and F1 score evaluation indexes of the model on the test set D to obtain a test performance coefficient Csx. The performance of the model can be more fully evaluated by comparison with the monitored performance coefficients Jkx on validation set C.
Example 3, this example is an explanation made in example 1, specifically, the specific method of defining the loss function is:
s1, for a fire alarm system, setting the probability of fire occurrence or the classification result of fire occurrence as the output of a deep learning model; the real tag is marked according to the information in the historical fire data A3; the output of the deep learning model is set to be the probability or classification result of fire occurrence, and the real labels are marked according to the historical fire data A3, so that the loss function can adapt to the characteristics of fire alarm tasks, and the model output is ensured to be matched with task requirements.
S2, selecting a set loss function to measure the difference between the predicted value and the true value according to the output of the deep learning model and the label type; for the second classification, including fire and non-fire classification, outputting by adopting a binary cross entropy loss function, and if the output is a probability value of fire occurrence, setting and calculating by using a mean square error loss function; for the two-classification problem, a binary cross entropy loss function is adopted for output, so that fire and non-fire situations can be effectively distinguished, and the model is helped to learn and correctly predict the probability or classification result of fire occurrence.
S3, inputting the output of the deep learning model and the corresponding real label into a loss function, and calculating to obtain a loss value of the deep learning model on a current sample; for all samples in the training set B, calculating average loss or total loss as the loss of the deep learning model on the whole training set B; depending on the type of output of the deep learning model, an appropriate loss function is selected to measure the difference between the predicted value and the actual value. If the output is a probability value of a fire occurrence, calculated using a mean square error loss function, this flexible choice enables the loss function to be matched to the task type, facilitating optimization of the model.
And S4, taking the loss function as an optimization target, and adjusting parameters of the deep learning model by using a gradient descent optimization algorithm in the training process, so that the loss function value is gradually reduced, the deep learning model is better fitted with training data, and a better fire disaster prediction effect is achieved. The loss function is used as an optimization target, and the gradient descent optimization algorithm is used for adjusting parameters of the deep learning model, so that the loss function value is gradually reduced, the model is helped to better fit training data, and the fire prediction effect is improved.
In this embodiment, the specific method for defining the loss function can adapt the deep learning model to the fire alarm task, distinguish fire and non-fire situations, flexibly select a suitable loss function, and help the model optimize parameters, thereby realizing a better fire prediction effect.
Example 4, this example is an explanation made in example 1, specifically, the safety-critical task coefficient rwx is calculated by the following formula:
where rws represents the number of important tasks that the fire alarm system needs to perform, including the number of tasks that detect flame, smoke fire characteristics; aqx is expressed as the degree of safety of the fire, including the coverage of the fire precautions and the number of emergency responses; mygl is expressed as a spread probability value; α, β, γ are weight coefficients of rws, aqx, and mygl, and are set by the user, and α+β+γ=1.0.
In this embodiment, by introducing the important task number rws as the weight coefficient α, the number of tasks that the fire alarm system needs to complete can be taken into consideration. Important tasks often cover the detection of critical features such as flames, smoke, etc., and the greater the number of these tasks, the more functions and capabilities the fire alarm system needs to have. The safety degree aqx of fire occurrence is introduced as a weight coefficient beta, and factors such as the coverage rate of preventive measures of the fire and the emergency response times can be considered. These factors can affect the processing and control effects after a fire has occurred, and by weighting considerations, the performance of the fire alarm system can be more fully assessed. By introducing the spread probability value mygl as the weight coefficient γ, the degree of spread of the fire can be considered. The speed and the range of the fire spread are related to the emergency of the fire early warning, and the actual application value of the fire alarm system can be estimated more accurately after the factor is considered. By calculating the safety important task coefficient rwx, the important task number of the fire alarm system, the safety degree of the fire and the spreading probability value can be comprehensively considered, and the performance and the early warning capability of the fire alarm system can be optimized, so that the accuracy and the reliability of the subway FAS fire alarm system are improved.
Example 5, which is an illustration of example 1, specifically, the dataset complexity index fz is calculated by the following formula:
fz＝(N*Ce)/(C1*C2)
where N represents the total number of samples in the dataset; ce represents a class balance index of the data set, and is used for measuring the balance degree of different class samples; metrics are measured using statistical indicators, including Gini coefficients and information entropy; c1 and C2 are adjustment coefficients representing the number of samples of the data set and the data distribution, respectively.
In this embodiment: the total number of samples N in a data set is an important indicator that reflects the size and capacity of the data set. Larger data sets can generally provide more rich information and more comprehensive sample coverage, helping to train a more robust and accurate deep learning model; the class-balancing index Ce of the dataset can measure the degree of balance of different class samples. In a subway FAS fire alarm system, fire data are often in a few categories, and normal operation data are in a plurality of categories, so that unbalanced distribution of data sets is caused. The quantitative difference of samples of different categories can be quantified through the Ce index, so that the problem of unbalanced data is solved, and the prediction effect of the model on fire samples is improved. The data distribution adjustment coefficients C1 and C2 of the data set are used to adjust the number of samples and the data distribution of the data set. The values of C1 and C2 are reasonably set, so that the data set can be more in line with the actual scene, and the generalization capability of the deep learning model on the actual data is improved; by comprehensively considering N, ce, C1 and C2, the complexity index fz of the data set can be used for comprehensively evaluating the complexity of the data set. A data set with moderate complexity and relatively balanced sample distribution is beneficial to training a deep learning model and is beneficial to improving the performance and generalization capability of the model.
The complex index fz of the data set is calculated, so that the data set of the subway FAS fire alarm system can be comprehensively evaluated, the sample distribution of the data set is optimized, the quality of the data set is improved, support is provided for training and predicting effects of a deep learning model, and the accuracy and reliability of the subway FAS fire alarm system are further improved.
Embodiment 6, which is an explanation made in embodiment 1, specifically, the resource depth coefficient sd is obtained by calculating the following formula:
wherein ccL is expressed as a parameter quantity value of a deep learning model, and in the neural network, parameters refer to weight and bias, and the greater the parameter quantity, the higher the complexity of the deep learning model; xql represents the computational resource memory requirements required for training and deep learning models; jjxl is expressed as computational efficiency, which includes computational time and computational power ratio.
In this embodiment, ccL in the resource depth coefficient sd represents the parameter number value of the deep learning model, including the weight and bias. The greater the number of parameters, the greater the complexity of the model, which can be fitted to more complex functional relationships, thereby improving the expressive power of the model. By calculating sd, the complexity of the model can be quantified, the proper model structure and layer number can be selected, the problem of overfitting is avoided, and the generalization capability of the model is improved. Xql in the resource depth coefficient sd represents the computational resource memory requirement values required for training and deep learning models. Computing resources are an important consideration in training deep learning models. Excessive computing resource requirements may result in excessive training time, excessive hardware costs, and may not even be able to run on a particular computing platform. By calculating sd, the demand of the computing resources of the model can be evaluated, which is helpful for selecting proper computing platform and hardware configuration and improving the training efficiency of the model. Jjxl in the resource depth coefficient sd represents computational efficiency, including computational time and computational power ratio. Computational efficiency means that models can be inferred and predicted quickly and efficiently given computing resources. The higher calculation efficiency means that the model can complete the prediction task in a shorter time, and is suitable for a real-time application scene. Through sd calculation, the calculation efficiency of the model can be evaluated, the model with higher calculation speed can be selected, and the performance of the model in practical application can be improved.
By calculating the resource depth coefficient sd, the complexity and calculation performance of the deep learning model can be comprehensively evaluated, the model suitable for the actual application scene can be selected, the prediction accuracy, calculation efficiency and instantaneity of the subway FAS fire alarm system can be improved, and the performance and reliability of the system can be further improved.
Example 7, which is an explanation made in example 1, specifically, the recall Xn, the accuracy zq, and the monitoring coefficient of performance Jkx are calculated by the following formula:
Xn＝TP/(TP+FN)
zq＝TP/(TP+FP)
wherein TP represents the number of samples for which the deep learning model correctly predicts as fire, and FN represents the number of samples for which the deep learning model incorrectly predicts as non-fire; FP represents the number of samples that the deep learning model mispredicts as a fire; f1 and f2 are coefficients for balancing weights between recall Xn and accuracy zq; the meaning of the parameters is that the recall rate Xn and the accuracy rate zq are balanced, if the higher recall rate Xn is more important, namely the fire disaster situation is not leaked as much as possible, the increase f1 is set, and if the higher accuracy rate zq is more important, namely the false alarm is avoided as much as possible, the increase f2 is set.
In the embodiment, the higher the recall rate Xn is, the better the model can capture fire samples, the lower the report missing rate is, and the sensitivity and the reliability of fire detection are enhanced; the higher the accuracy zq is, the more accurate the prediction result of the model is, the false alarm rate is reduced, and the reliability of the fire alarm system is improved; f1 and f2 are coefficients for balancing weights between recall Xn and accuracy zq. Increasing the value of f1 when the model requires more attention to the recall of fire detection; the value of f2 is increased when the model requires more attention to accuracy. The higher the monitoring performance coefficient Jkx is, the better balance between the recall rate and the accuracy rate is achieved by the model, and the better fire detection performance is achieved.
Example 8, this example is an illustration made in example 1, specifically, the test coefficient of performance Csx is calculated by the following formula:
wherein F1 is expressed as F1 fraction on the test set D, and is used for comprehensively considering the balance of recall rate and accuracy rate; the F1 score is a harmonic mean value of the recall rate Xn and the accuracy rate zq and is used for measuring the comprehensive performance of the deep learning model; w1, w2, w3 are represented as weight coefficient values of F1 score, recall Xn, and accuracy zq, and w1+w2+w3=1.0; r represents a correction constant.
In this embodiment, by calculating the test performance coefficient Csx, the recall rate, the accuracy and the F1 score of the model on the test set D may be comprehensively considered, so as to evaluate the overall performance of the deep learning model. The higher Csx is, the better the performance of the model on test data is, the fire situation can be detected more accurately, and more reliable early warning and guarantee are provided for a subway FAS fire alarm system. By setting reasonable weight coefficients, the trade-off relation between the recall rate and the accuracy rate of the model can be flexibly balanced, and the performance of the fire detection system is further optimized.
Example 9, which is an explanation made in example 1, specifically, a difference Diff between the test coefficient of performance Csx and the monitor coefficient of performance Jkx, i.e., diff=csx-Jkx, is calculated; according to the magnitude and direction of the difference value Diff, judging whether the performance of the deep learning model on the test set D is better than the performance of the deep learning model on the verification set C or not; a large positive difference indicates that the deep learning model performs better on test set D, and a large negative difference may indicate that the deep learning model performs worse on test set D;
according to the result of the difference value Diff, corresponding scheme debugging is carried out; if the test performance coefficient Csx is better, namely Csx > Jkx, the performance of the deep learning model on the test set D is better than that on the verification set C, and the deep learning model is considered to be continuously used; if the test performance coefficient Csx is poor, i.e., csx < Jkx, then it is necessary to re-optimize the super-parameters, adjust the deep learning model structure, or add more data to improve the deep learning model.
In this embodiment, by calculating and comparing the difference Diff between the test performance coefficient Csx and the monitor performance coefficient Jkx, objective evaluation can be made on the performance of the deep learning model, and a corresponding debugging scheme can be adopted accordingly. The method is favorable for optimizing the deep learning model, improving the fire detection capability of the subway FAS fire alarm system and effectively guaranteeing the safety of passengers and subways.
Although embodiments of the present invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made therein without departing from the principles and spirit of the invention, the scope of which is defined in the appended claims and their equivalents.
Claims (10)
1. A debugging method of a subway FAS fire alarm system based on deep learning is characterized by comprising the following steps: comprises the steps of,
collecting subway line, carriage, different time periods and weather condition sample data of different subway lines in a time axis mode, and collecting video and sensor sample data when the subway normally operates and when fire records occur; recorded as dataset a;
marking the data set A, marking the data of normal conditions and historical fire conditions, and marking the data including the time stamp, the position, the fire scale and related influence information of the occurrence of the fire;
after preprocessing the data set A, dividing the data set A into a training set B, a verification set C and a test set D;
extracting fire features from a fire alarm system of a training set B, calculating and obtaining a safety important task coefficient rwx, a data set complex index fz and a resource depth coefficient sd, fitting the safety important task coefficient rwx, the data set complex index fz and the resource depth coefficient sd to obtain a complex network structure coefficient fzwx, selecting a corresponding network structure and layer number model according to the complex network structure coefficient fzwx, and setting a definition loss function to construct a deep learning model;
the complex network structure coefficient fzwx is obtained by the following formula:
wherein Xn is expressed as recall, zq is expressed as accuracy, F1 is expressed as F1 score evaluation index, ypzl is expressed as total number of sample data in the set;
training the deep learning model by using a training set B, and adopting an optimization algorithm comprising one of SGD and Adam to continuously update parameters of the deep learning model to gradually converge, wherein in the training process, a verification set C is used for verifying the deep learning model; calculating recall rate Xn and accuracy zq of the deep learning model on the verification set C; calculating a monitoring performance coefficient Jkx according to the recall rate Xn and the accuracy rate zq, and adjusting the super-parameter configuration according to the monitoring performance coefficient Jkx;
evaluating the performance of the deep learning model using the test set D; the specific method comprises the steps of calculating recall rate Xn, accuracy zq and F1 fraction evaluation indexes of the deep learning model on a test set D, obtaining a test performance coefficient Csx, comparing the test performance coefficient Csx with a monitoring performance coefficient Jkx, obtaining a difference value Diff, and debugging a corresponding scheme aiming at the difference value Diff.
2. The debugging method of the deep learning-based subway FAS fire alarm system according to claim 1, wherein the debugging method comprises the following steps: the data set A comprises an image data set A1, a sensor data set A2 and historical fire data A3, and a camera and an infrared sensor are installed to collect image data in a subway carriage and marked as image data A1;
a temperature sensor, a smoke sensor, a gas sensor and a sensor data A2 are adopted to collect temperature data, smoke data and combustible gas leakage data in a carriage;
and collecting historical fire record report data which is published by related departments of the subway system through big data, wherein the historical fire record report data comprises fire reasons, fire spread conditions, alarm times and emergency treatment data, and is marked as historical fire data A3.
3. The debugging method of the deep learning-based subway FAS fire alarm system according to claim 2, wherein the debugging method comprises the following steps: dividing the data set A into a training set B, a verification set C and a test set D according to the proportion of 70%, 15% and 15%;
and extracting fire features from the training set B fire alarm system, wherein the fire features comprise fire wire and dense smoke area intensity features, fire light intensity change features, target carriage internal object deformation features, carriage internal temperature change features, smoke concentration change features, combustible gas leakage features and fire spread change features.
4. The debugging method of the deep learning-based subway FAS fire alarm system according to claim 1, wherein the debugging method comprises the following steps: the specific method for defining the loss function is as follows:
s1, for a fire alarm system, setting the probability of fire occurrence or the classification result of fire occurrence as the output of a deep learning model; the real tag is marked according to the information in the historical fire data A3;
s2, selecting a set loss function to measure the difference between the predicted value and the true value according to the output of the deep learning model and the label type; for the second classification, including fire and non-fire classification, outputting by adopting a binary cross entropy loss function, and if the output is a probability value of fire occurrence, setting and calculating by using a mean square error loss function;
s3, inputting the output of the deep learning model and the corresponding real label into a loss function, and calculating to obtain a loss value of the deep learning model on a current sample; for all samples in the training set B, calculating average loss or total loss as the loss of the deep learning model on the whole training set B;
and S4, taking the loss function as an optimization target, and adjusting parameters of the deep learning model by using a gradient descent optimization algorithm in the training process, so that the loss function value is gradually reduced, the deep learning model is better fitted with training data, and a better fire disaster prediction effect is achieved.
5. The debugging method of the deep learning-based subway FAS fire alarm system according to claim 1, wherein the debugging method comprises the following steps: the safety important task coefficient rwx is obtained by calculation through the following formula:
where rws represents the number of important tasks that the fire alarm system needs to perform, including the number of tasks that detect flame, smoke fire characteristics; aqx is expressed as the degree of safety of the fire, including the coverage of the fire precautions and the number of emergency responses; mygl is expressed as a spread probability value; α, β, γ are weight coefficients of rws, aqx, and mygl, and are set by the user, and α+β+γ=1.0.
6. The debugging method of the deep learning-based subway FAS fire alarm system according to claim 1, wherein the debugging method comprises the following steps: the dataset complexity index fz is calculated by the following formula:
fz＝(N*Ce)/(C1*C2)
where N represents the total number of samples in the dataset; ce represents a class balance index of the data set, and is used for measuring the balance degree of different class samples; metrics are measured using statistical indicators, including Gini coefficients and information entropy; c1 and C2 are adjustment coefficients representing the number of samples of the data set and the data distribution, respectively.
7. The debugging method of the deep learning-based subway FAS fire alarm system according to claim 1, wherein the debugging method comprises the following steps: the resource depth coefficient sd is obtained by calculation according to the following formula:
wherein ccL is expressed as a parameter quantity value of a deep learning model, and in the neural network, parameters refer to weight and bias, and the greater the parameter quantity, the higher the complexity of the deep learning model; xql represents the computational resource memory requirements required for training and deep learning models; jjxl is expressed as computational efficiency, which includes computational time and computational power ratio.
8. The debugging method of the deep learning-based subway FAS fire alarm system according to claim 1, wherein the debugging method comprises the following steps: the recall rate Xn, the accuracy zq and the monitoring performance coefficient Jkx are obtained by calculation according to the following formula:
Xn＝TP/(TP+FN)
zq＝TP/(TP+FP)
wherein TP represents the number of samples for which the deep learning model correctly predicts as fire, and FN represents the number of samples for which the deep learning model incorrectly predicts as non-fire; FP represents the number of samples that the deep learning model mispredicts as a fire; f1 and f2 are coefficients for balancing weights between recall Xn and accuracy zq; the meaning of the parameters is that the recall rate Xn and the accuracy rate zq are balanced, if the higher recall rate Xn is more important, namely the fire disaster situation is not leaked as much as possible, the increase f1 is set, and if the higher accuracy rate zq is more important, namely the false alarm is avoided as much as possible, the increase f2 is set.
9. The debugging method of the deep learning-based subway FAS fire alarm system according to claim 1, wherein the debugging method comprises the following steps: the test performance coefficient Csx is obtained by calculation according to the following formula:
wherein F1 is expressed as F1 fraction on the test set D, and is used for comprehensively considering the balance of recall rate and accuracy rate; the F1 score is a harmonic mean value of the recall rate Xn and the accuracy rate zq and is used for measuring the comprehensive performance of the deep learning model; w1, w2, w3 are represented as weight coefficient values of F1 score, recall Xn, and accuracy zq, and w1+w2+w3=1.0; r represents a correction constant.
10. The method for debugging the deep learning-based subway FAS fire alarm system according to claim 9, wherein: calculating the difference Diff between the test coefficient of performance Csx and the monitor coefficient of performance Jkx, i.e., diff=csx-Jkx; according to the magnitude and direction of the difference value Diff, judging whether the performance of the deep learning model on the test set D is better than the performance of the deep learning model on the verification set C or not; a large positive difference indicates that the deep learning model performs better on test set D, and a large negative difference may indicate that the deep learning model performs worse on test set D;
according to the result of the difference value Diff, corresponding scheme debugging is carried out; if the test performance coefficient Csx is better, namely Csx > Jkx, the performance of the deep learning model on the test set D is better than that on the verification set C, and the deep learning model is considered to be continuously used; if the test performance coefficient Csx is poor, i.e., csx < Jkx, then it is necessary to re-optimize the super-parameters, adjust the deep learning model structure, or add more data to improve the deep learning model.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202311122909.XA CN117150298B (en) | 2023-09-01 | 2023-09-01 | Deep learning-based subway FAS fire alarm system debugging method |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202311122909.XA CN117150298B (en) | 2023-09-01 | 2023-09-01 | Deep learning-based subway FAS fire alarm system debugging method |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN117150298A true CN117150298A (en) | 2023-12-01 |
| CN117150298B CN117150298B (en) | 2024-06-07 |
Family
ID=88883827
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202311122909.XA Active CN117150298B (en) | 2023-09-01 | 2023-09-01 | Deep learning-based subway FAS fire alarm system debugging method |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN117150298B (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117649659A (en) * | 2023-12-07 | 2024-03-05 | 国网福建省电力有限公司漳浦县供电公司 | Particle identification method based on deep learning and intelligent operation and maintenance system of power equipment |
Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109522819A (en) * | 2018-10-29 | 2019-03-26 | 西安交通大学 | A kind of fire image recognition methods based on deep learning |
| CN114283469A (en) * | 2021-12-14 | 2022-04-05 | 贵州大学 | Lightweight target detection method and system based on improved YOLOv4-tiny |
| CN114444386A (en) * | 2022-01-19 | 2022-05-06 | 中国能源建设集团安徽省电力设计院有限公司 | Fire early warning and post-disaster floor damage prediction method and system based on BIM and deep learning |
| CN115221964A (en) * | 2022-07-21 | 2022-10-21 | 武汉峰锐智能科技有限公司 | Fire alarm open set classification method based on deep learning |
| CN116108884A (en) * | 2023-02-24 | 2023-05-12 | 北京理工大学 | Improved fuzzy neural network fire detection method based on genetic algorithm |
| US20230206420A1 (en) * | 2021-01-28 | 2023-06-29 | Beijing Zhongxiangying Technology Co., Ltd. | Method for detecting defect and method for training model |
-
2023
- 2023-09-01 CN CN202311122909.XA patent/CN117150298B/en active Active
Patent Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109522819A (en) * | 2018-10-29 | 2019-03-26 | 西安交通大学 | A kind of fire image recognition methods based on deep learning |
| US20230206420A1 (en) * | 2021-01-28 | 2023-06-29 | Beijing Zhongxiangying Technology Co., Ltd. | Method for detecting defect and method for training model |
| CN114283469A (en) * | 2021-12-14 | 2022-04-05 | 贵州大学 | Lightweight target detection method and system based on improved YOLOv4-tiny |
| CN114444386A (en) * | 2022-01-19 | 2022-05-06 | 中国能源建设集团安徽省电力设计院有限公司 | Fire early warning and post-disaster floor damage prediction method and system based on BIM and deep learning |
| CN115221964A (en) * | 2022-07-21 | 2022-10-21 | 武汉峰锐智能科技有限公司 | Fire alarm open set classification method based on deep learning |
| CN116108884A (en) * | 2023-02-24 | 2023-05-12 | 北京理工大学 | Improved fuzzy neural network fire detection method based on genetic algorithm |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117649659A (en) * | 2023-12-07 | 2024-03-05 | 国网福建省电力有限公司漳浦县供电公司 | Particle identification method based on deep learning and intelligent operation and maintenance system of power equipment |
Also Published As
| Publication number | Publication date |
|---|---|
| CN117150298B (en) | 2024-06-07 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110647539B (en) | Prediction method and system for vehicle faults | |
| CN109186813B (en) | Temperature sensor self-checking device and method | |
| CN116625438B (en) | Gas pipe network safety on-line monitoring system and method thereof | |
| CN117150298B (en) | Deep learning-based subway FAS fire alarm system debugging method | |
| CN110119758A (en) | A kind of electricity consumption data abnormality detection and model training method, device | |
| WO2022147853A1 (en) | Complex equipment power pack fault prediction method based on hybrid prediction model | |
| CN117151649B (en) | Construction method management and control system and method based on big data analysis | |
| CN115865649A (en) | Intelligent operation and maintenance management control method, system and storage medium | |
| CN111191855B (en) | Water quality abnormal event identification and early warning method based on pipe network multi-element water quality time sequence data | |
| CN114023399A (en) | Air particulate matter analysis early warning method and device based on artificial intelligence | |
| CN117455745B (en) | Public safety event sensing method and system based on multidimensional fusion data analysis | |
| CN116311739A (en) | Multi-sensor fire detection method based on long-short-term memory network and environment information fusion | |
| CN117875946B (en) | Man-machine collaborative autonomous infrared inspection method for operation and maintenance of transformer substation equipment | |
| CN110598969B (en) | Offshore channel emergency risk early warning method | |
| CN117688464B (en) | Hidden danger analysis method and system based on multi-source sensor data | |
| CN116502156B (en) | Intelligent identification method and system for light CT abnormal state of converter station | |
| CN117113157B (en) | Platform district power consumption fault detection system based on artificial intelligence | |
| CN118092326A (en) | Digital twinning-based factory emergency management system | |
| KR20230071565A (en) | System for disaster observation monitoring and method thereof | |
| Santos et al. | Machine Learning in Failure Prediction in Breathalyzers | |
| Liu et al. | Sensor Fault Diagnosis of Air Conditioning System Based on Multi-dimension Clustering Algorithm | |
| CN117807507A (en) | Sensing layer detection system and detection method for ubiquitous power internet of things power equipment | |
| CN116307408A (en) | Internet enterprise management system based on data analysis | |
| CN118153948A (en) | Hydrogen energy storage equipment leakage fault early warning method and device based on deep learning | |
| CN118333411A (en) | Safety production accident hidden danger early warning system based on artificial intelligence |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant |