Inspection method of important equipment control mode in automatic fire alarm system

Abstract

The invention provides a method for checking the control mode of important equipment in an automatic fire alarm system, and the method comprises the steps: carrying out the wavelet transformation of the time series data of a key control node, extracting the time-frequency domain features, classifying the time-frequency domain features, and analyzing the operation conditions of a roller shutter door, the operation conditions including speed sudden change and position deviation; inputting the operating parameter feature vector into a pre-trained convolutional neural network model, evaluating the operating state of the multi-line control mode of the roller shutter door, and judging whether the operating state accords with a preset fire-fighting specification or not; sending the adjustment suggestion to a fire control center, generating an adjustment instruction of a multi-line control mode of the roller shutter door, and issuing the adjustment instruction to a roller shutter door controller on site through a communication module of the automatic fire alarm system; and the roller shutter door controller dynamically adjusts the operation parameters of the control circuit according to the adjustment instruction, and the closing process of the roller shutter door is optimized.

Description

Technical Field

[0001] The present invention relates to the field of information technology, and in particular to a method for checking the control mode of important equipment in an automatic fire alarm system. Background Art

[0002] In the automatic fire alarm system, the closing process of the rolling shutter door adopts a multi-line control method. In order to ensure that the rolling shutter door can be closed quickly and reliably when a fire occurs, it is necessary to analyze the velocity stress curve during the closing process of the rolling shutter door. By drawing the velocity stress curve, the key control nodes in the closing process of the rolling shutter door, such as segmented landing and secondary reset, can be identified. At these key nodes, the speed and position of the rolling shutter door may have sudden changes or deviations, and these abnormal conditions may affect the normal closing of the rolling shutter door. In order to determine whether the multi-line control method of the rolling shutter door meets the preset fire protection specifications, it is necessary to perform dynamic characteristic analysis on the abnormal conditions of speed mutation and position deviation, and check the operating parameters of each control circuit in the rolling shutter door. However, in actual applications, due to factors such as the complex structure of the rolling shutter door and the changeable operating environment, there are certain technical challenges in accurately analyzing the velocity stress curve and identifying abnormal conditions. How to achieve real-time monitoring and dynamic adjustment of the closing process while ensuring the reliable closing of the rolling shutter door is a technical problem that needs to be solved urgently. Summary of the invention

[0003] The present invention provides a method for checking the control mode of important equipment in an automatic fire alarm system, which mainly includes:

[0004] The automatic fire alarm system collects speed and stress data during the closing process of the rolling shutter door under multi-line control mode, draws speed stress curves, and obtains time series data of key control nodes of segmented landing and secondary reset;

[0005] Perform wavelet transform on the time series data of key control nodes, extract time-frequency domain features, classify the time-frequency domain features, and analyze the operation of the rolling door, including speed mutation and position deviation;

[0006] Obtain the operating parameters of the rolling door control line in the time period corresponding to the speed mutation and position deviation, the operating parameters including current, voltage, and power, and construct a multi-dimensional operating parameter feature vector based on the operating parameters;

[0007] The operating parameter feature vector is input into the pre-trained convolutional neural network model to evaluate the operating status of the rolling door multi-line control mode and determine whether it meets the preset fire protection regulations;

[0008] If the judgment result is not in line, perform an interpretability analysis on the convolutional neural network model to obtain key operating parameters, retrieve a parameter optimization scheme corresponding to the key operating parameters in a preset rule library, and evaluate the effects of each scheme on improving the model judgment result through simulation of the parameter optimization scheme, so as to select a target parameter adjustment scheme;

[0009] Send the adjustment suggestion to the fire control center, generate an adjustment instruction for the multi-wire control mode of the rolling shutter door, and send it to the rolling shutter door controller on site through the communication module of the fire alarm system;

[0010] The rolling shutter door controller dynamically adjusts the operating parameters of the control circuit according to the adjustment instruction to optimize the closing process of the rolling shutter door.

[0011] The technical solution provided by the embodiment of the present invention may include the following beneficial effects:

[0012] The present invention discloses a method for inspecting the control mode of important equipment in a fire alarm system. The method collects the speed and stress data during the closing process of the rolling shutter door through the fire alarm system, and uses wavelet transform and support vector machine algorithms to identify abnormal operations. Further obtain the operating parameters of the control circuit during the abnormal period, construct a feature vector and input it into a pre-trained convolutional neural network model to evaluate the operating state of the rolling shutter door. If it does not meet the fire protection specifications, the present invention performs an interpretability analysis, retrieves an optimization scheme from the rule library and conducts a simulation evaluation, and selects the best adjustment scheme. Finally, the adjustment instruction is sent to the on-site controller through the fire alarm system to realize the dynamic optimization of the closing process of the rolling shutter door. The present invention realizes the real-time monitoring, evaluation and optimization of the operating state of the rolling shutter door through intelligent algorithms and multi-dimensional data analysis, and improves the fire safety performance. Description of the Drawings

[0013] Figure 1 It is a flowchart of a method for inspecting the control mode of important equipment in a fire alarm system of the present invention. Detailed Embodiments

[0014] In order to make the purpose, technical solution and advantages of the present invention clearer, the present invention will be described in detail below with reference to the drawings and specific embodiments.

[0015] As Figure 1 , a method for inspecting the control mode of important equipment in a fire alarm system in this embodiment may specifically include:

[0016] S101. The fire alarm system collects the speed and stress data during the closing process of the rolling shutter door in the multi-wire control mode, draws a speed-stress curve, and obtains the time series data of the key control nodes of segmented descent and secondary reset.

[0017] Collect the rolling shutter position data through a displacement sensor, the motion speed data through a speed sensor, and the force data through a stress sensor to form a rolling shutter motion state data packet; for the motion state data packet, the multi-threaded controller establishes a motion characteristic curve and a force characteristic curve to form a speed-stress correlation data set; use a self-learning convolutional neural network to train the speed-stress correlation data set, obtain the inflection point of the rolling shutter speed change as the segmented descent threshold point, and obtain the inflection point of the stress change as the stress threshold point during the descent process; perform time series analysis on the segmented descent threshold point and the stress threshold point, calculate the time interval data between the segmented descent threshold point and the stress threshold point, and obtain the rolling shutter segmented descent control data set.

[0018] Exemplarily, from the generation of the fire alarm signal to the control point of the rolling shutter door, the displacement sensor collects the real-time position data of the rolling shutter door, the motor speed sensor collects the movement speed data of the rolling shutter door, and the stress sensor collects the force data of the rolling shutter door. Through the data collector, a movement state data packet of the rolling shutter door is formed. For the movement state data packet of the rolling shutter door, the multi-threaded controller reads the displacement data and speed data to establish the movement characteristic curve of the rolling shutter door, reads the stress data to establish the force characteristic curve of the rolling shutter door, and forms the speed-stress correlation data set of the rolling shutter door according to the characteristic curve. During the segmented descent process of the rolling shutter door, the self-learning convolutional neural network trains the speed-stress correlation data set, obtains the inflection point of the speed change of the rolling shutter door as the segmented descent threshold point, and obtains the inflection point of the stress change as the stress threshold point during the descent process. Through the random forest algorithm, time series analysis is performed on the segmented descent threshold point and the stress threshold point, and the time interval data between the threshold points is calculated to form the segmented descent control data set of the rolling shutter door. For the secondary reset process of the rolling shutter door, the multi-threaded controller reads the time interval data of the threshold points in the control data set, establishes the reset control curve of the rolling shutter door, and calculates the speed threshold and stress threshold during the reset process. According to the reset control curve, the correlation degree between the real-time movement data and the force data of the rolling shutter door is calculated. If the correlation degree is less than the preset threshold, the movement parameters of the rolling shutter door are adjusted until the correlation degree is greater than the preset threshold. In the actual application scenario, the displacement sensor senses the movement position of the rolling shutter door through a Hall element, the sampling frequency is set to 50Hz, and the accuracy of obtaining the position data of the rolling shutter door can reach 0.1mm. The motor speed sensor uses an incremental encoder with a resolution of 1024 lines / revolution, which can achieve high-precision detection of the movement speed of the rolling shutter door. The stress sensor uses a strain gauge force sensor with a measurement range of 0-1000N and a sensitivity of 2mV / V. The data collector synchronously samples the data of the three sensors, and the sampling time window is 100ms to form a data packet containing position, speed, and stress. The multi-threaded controller constructs the movement characteristic curve of the rolling shutter door based on the real-time data packet, reflects the movement state of the door body through the speed-time curve, and characterizes the force change through the stress-time curve. The speed-stress correlation data set records the corresponding relationship between speed and stress during the operation of the rolling shutter door, providing data support for subsequent segmented descent control. The self-learning convolutional neural network adopts a 3-layer convolutional layer structure, with 128 nodes in the input layer, 64 nodes in the hidden layer, and 32 nodes in the output layer, which is used to extract the features in the speed-stress correlation data set. By analyzing the change of the slope of the speed curve, the inflection point where the speed drops from 25cm / s to 15cm / s is identified as the segmented descent threshold point. Similarly, the inflection point where the stress drops from 800N to 500N in the stress curve is used as the stress threshold point. The random forest algorithm constructs 50 decision trees to perform classification and regression analysis on the time series data of the threshold points. Practice shows that the first segment descent threshold point appears 2.5s after startup, the second segment descent threshold point appears at 5s, and the total descent time is 7.5s. The time interval data of the threshold points guides the controller to adjust the motor output to achieve smooth segmented descent of the rolling shutter door.The reset control curve adopts the piecewise linear interpolation method, which subdivides the rolling shutter door movement speed and the acting force into multiple control intervals. The correlation calculation is based on the Pearson correlation coefficient. When the correlation coefficient of the speed and stress data is lower than the correlation coefficient threshold of 0.85, it indicates that the movement is unstable. The controller corrects the movement parameters by adjusting the motor output torque until the correlation coefficient is greater than the correlation coefficient threshold. The measured data shows that the adaptive control method with correlation feedback can control the speed fluctuation within the speed fluctuation requirement range and the acting force fluctuation within the acting force fluctuation requirement range during the secondary reset process of the rolling shutter door.

[0019] S102. Perform wavelet transform on the time series data of key control nodes, extract the time-frequency domain features, classify the time-frequency domain features, and analyze the operation conditions of the rolling shutter door. The operation conditions include speed mutation and position deviation.

[0020] Obtain the rolling shutter door movement displacement and speed change data through a Hall sensor, perform time marking on the movement displacement and speed change data in a data collector to obtain a time series state data set; perform wavelet transform according to the time series state data set, extract high-frequency coefficients and low-frequency coefficients from the wavelet coefficient matrix, and reconstruct the high-frequency coefficients and low-frequency coefficients to obtain the time-frequency domain basic features; calculate the statistical quantities of signal segment variance, mean, kurtosis, and skewness for the time-frequency domain basic features, perform principal component analysis on the statistical quantities, and extract the eigenvector with the largest contribution rate as the dimensionality reduction feature; use the support vector machine algorithm to construct a feature classifier, classify the dimensionality reduction feature, and obtain the speed change rate feature and the position offset amount feature. If the speed change rate is greater than the speed threshold or the position offset amount is greater than the position threshold, it is determined that the rolling shutter door is abnormal.

[0021] Exemplarily, the operating status data is collected from the rolling shutter control node, and the displacement and speed change data of the rolling shutter are obtained through a Hall sensor. The operating status data is marked with the control node time in the data collector to form a time series status data set. The wavelet transform is performed on the time series status data set in the data collector to generate a wavelet coefficient matrix. The high-frequency coefficients and low-frequency coefficients are extracted from the wavelet coefficient matrix, and the coefficients are reconstructed to obtain the time-frequency domain basic features. For the time-frequency domain basic features, a fixed-length signal segment is intercepted through a sliding window, and statistics such as variance, mean, kurtosis, and skewness of the signal segment are calculated to form the time-frequency domain statistical features. The principal component analysis is performed on the time-frequency domain statistical features, and the eigenvector with the largest contribution rate is extracted as the dimensionality reduction feature. A feature classifier is constructed through the support vector machine algorithm. The feature classifier is used to classify the time-frequency domain dimensionality reduction features, and the speed change rate feature and the position offset feature are extracted. The speed change rate threshold and the position offset threshold are set. According to the feature classification result, if the speed change rate is greater than the speed threshold, it is determined that the rolling shutter has a speed mutation. If the position offset is greater than the position threshold, it is determined that the rolling shutter has a position deviation. The monitoring of the rolling shutter operating status requires real-time data collection and analysis of multiple key control nodes. Through the Hall sensor installed on the rolling shutter track, the door body displacement data is recorded at the sampling frequency, and the motor speed signal is collected simultaneously to obtain the motion speed data. The timestamp accuracy at each control node reaches the millisecond level, and a multi-dimensional state data sequence including position, speed, and time is constructed. The db4 wavelet basis function is applied to the collected time series data for 4-layer decomposition to obtain the approximation coefficients reflecting the low-frequency motion trend of the rolling shutter and the detail coefficients characterizing the high-frequency disturbance characteristics respectively. The wavelet transform has the ability of multi-resolution analysis in the time-frequency domain. By observing the coefficient changes at different scales, the abnormal fluctuations in the operation process of the rolling shutter can be identified. The actual operation data shows that the fluctuation range of the detail coefficients is within ±0.5 during normal operation, while sharp peaks exceeding 1.0 will appear in the detail coefficients during speed mutations. The sliding window is used to segment the time-frequency domain signal, and the statistical feature quantities of each segment of the signal are calculated. The variance reflects the speed fluctuation amplitude, the mean represents the average operating status, the kurtosis measures the severity of the speed mutation, and the skewness describes the asymmetry of the speed distribution. After analysis, the variance is less than 0.3 and the kurtosis is between 2.8 and 3.2 during normal operation, while the variance increases to more than 0.8 and the kurtosis exceeds 4.0 during a failure. The dimensionality reduction of the 12-dimensional statistical features is performed by the principal component analysis method, and the first 3 principal components with a cumulative contribution rate reaching 85% are selected as the input features of the support vector machine. The support vector machine uses the radial basis kernel function, the penalty factor is set to 10, and the kernel parameters are optimized using cross-validation. The training samples include 500 groups of normal operation data and 100 groups of failure data, and the accuracy of the validation set reaches 95%.In practical applications, when the detected rate of change of speed exceeds the speed change rate threshold, it indicates that the rolling shutter door may be stuck or blocked; when the position offset exceeds the position offset threshold, it indicates that the guide rail is deformed or the door body is installed with deviation. During a certain fire drill, the system successfully identified a sudden speed change caused by dust accumulation on the guide rail, with the deviation value exceeding the speed change rate threshold, and timely issued a warning signal. By comparing and analyzing the data of multiple drills, it was found that the position deviation often appears prior to the sudden speed change and can be used as an important indicator for predictive maintenance.

[0022] S103. Obtain the operating parameters of the rolling shutter door control circuit during the time period corresponding to the sudden speed change and position deviation. The operating parameters include current, voltage, and power, and construct a multi-dimensional operating parameter feature vector based on the operating parameters.

[0023] Collect current signals, voltage signals, and power signals from the rolling shutter door control circuit through a circuit monitor, and obtain the original operating parameter dataset after filtering the collected signals; intercept data segments according to the timestamps of the sudden speed change points and position deviation points in the original operating parameter dataset, and perform normalization processing on the data segments to obtain a standardized operating parameter dataset; calculate the current change rate, voltage change rate, and power change rate for the standardized operating parameter dataset, and perform smoothing processing through a Gaussian filter to obtain a parameter change feature curve; extract the fluctuation frequency feature, peak feature, and root mean square feature from the parameter change feature curve using a deep neural network, and then perform principal component analysis for dimensionality reduction and fusion to obtain a multi-dimensional feature vector of the operating parameters.

[0024] Exemplarily, current signals, voltage signals, and power signals in the rolling shutter control circuit are obtained through a circuit monitor, and the collected signals are sampled and filtered. The sampling time points of the operating parameters are marked in the data collector to generate an original dataset of the operating parameters. According to the timestamps of the speed mutation points and position deviation points, data segments of 10 sampling points before and after are intercepted from the original dataset of the operating parameters, and the data segments are normalized to generate a standardized dataset of the operating parameters. For the standardized dataset of the operating parameters, the current change rate, voltage change rate, and power change rate are calculated, and the change rate curve is smoothed through a Gaussian filter to generate a parameter change characteristic curve. The piecewise integral operation is performed on the parameter change characteristic curve to calculate the current integral value, voltage integral value, and power integral value, and the first group of components of the operating parameter characteristic sequence is generated. Through a deep neural network, the fluctuation frequency characteristics, peak characteristics, and root mean square characteristics are extracted from the parameter change characteristic curve to generate the second group of components of the operating parameter characteristic sequence. The principal component analysis method is used to reduce the dimension and fuse the two groups of characteristic sequence components, and the characteristic dimension with the largest contribution rate is selected to construct a multi-dimensional characteristic vector of the operating parameters. The monitoring of the operating parameters of the rolling shutter control circuit involves the real-time acquisition and processing of three key physical quantities: current, voltage, and power. The circuit monitor uses high-precision sensors, with a current sampling accuracy of 0.1 A, a voltage sampling accuracy of 0.1 V, and a power calculation accuracy of 1 W. The sampling frequency is set to 1000 Hz to ensure the capture of rapidly changing electrical parameters. The original data passes through a 50 Hz notch filter to eliminate power frequency interference, and the sampling points carry timestamp information accurate to milliseconds. When a speed mutation or position deviation occurs, 10 sampling points before and after the abnormal point are located from the timestamp to form a 21-point data segment. For a 380 V industrial power supply, the voltage data is normalized to the 0-1 interval, the current data is normalized according to the rated current of the motor of 15 A, and the power data is normalized according to the rated power of the motor of 5.5 kW to eliminate the dimension difference. The measured data shows that the fluctuation range of the normalized parameters during normal operation is within ±0.1. The parameter change rate reflects the dynamic change of the electrical characteristics and is obtained by calculating the difference between adjacent sampling points. The current change rate reaches a peak of 0.8 A / ms at the start moment and remains below 0.1 A / ms during normal operation. The voltage change rate is affected by the power grid fluctuation and usually does not exceed 0.5 V / ms. The power change rate comprehensively reflects the load characteristics, and the normal value range is ±100 W / ms. The standard deviation of the Gaussian filter takes 5 sampling points, and the smoothed curve is more suitable for feature extraction. The piecewise integral is performed on the smoothed curve, and the integral interval takes the same number of sampling points on both sides of the abnormal point to reflect the cumulative change of the parameters. The current integral value characterizes the change of the motor output torque and is at the order of 0.5 A·s during normal operation. The voltage integral value reflects the power grid stability, and the fluctuation range is ±2 V·s. The power integral value indicates the energy conversion characteristics, and the typical value is 0.2 kW·s. The deep neural network contains 3 convolutional layers and 2 fully connected layers to extract the fluctuation characteristics of the parameter curve.The frequency feature reflects the periodic change of parameters, the peak feature depicts the instantaneous mutation, and the root mean square feature characterizes the steady-state characteristics. Through principal component analysis, the cumulative contribution rate of the first three principal components reaches 90%, constituting the final feature vector. In particular, current mutations often precede mechanical failures and can be used as early warning indicators.

[0025] S104. Input the operation parameter feature vector into a pre-trained convolutional neural network model to evaluate the operation state of the multi-wire control mode of the rolling shutter door and determine whether it meets the preset fire protection specifications.

[0026] Read the speed threshold, position deviation threshold, and electrical parameter threshold from the fire protection specification database according to the operation parameter feature vector of the rolling shutter door, and generate standardized feature data through a data pre-processor; for the standardized feature data, use a convolutional neural network to extract the speed feature, position feature, and electrical feature of the rolling shutter door to generate an operation state feature matrix; calculate the compliance scores of the speed parameter, position parameter, and electrical parameter through a fully connected layer for the operation state feature matrix; set the speed parameter threshold interval, position parameter threshold interval, and electrical parameter threshold interval according to the compliance scores. If the speed parameter, position parameter, and electrical parameter are all within the preset threshold intervals, it is determined that the operation state of the rolling shutter door meets the requirements of the fire protection specifications.

[0027] Exemplarily, according to the operating parameter feature vector, the rolling shutter door operating speed threshold, position deviation threshold, and electrical parameter threshold are read from the fire protection specification database. The feature vector is normalized by a data preprocessor to generate standardized feature data. The standardized feature data is enhanced by adding Gaussian noise and random offsets to generate an augmented feature dataset. The training data and validation data are divided by a cross-validation selector. A convolutional neural network is used to extract features from the training data. The convolutional kernel parameters are optimized by backpropagation to extract the speed feature, position feature, and electrical feature of the rolling shutter door, generating an operating state feature matrix. For the operating state feature matrix, the scores of each parameter are calculated through a fully connected layer, and the compliance with the specifications of the speed parameter, position parameter, and electrical parameter is calculated respectively. According to the compliance score, the threshold value intervals of the speed parameter, position parameter, and electrical parameter are set to generate multi-dimensional evaluation indicators. Comparing with the multi-dimensional evaluation indicators, if the speed parameter, position parameter, and electrical parameter are all within the preset threshold value intervals, it is determined that the operating state of the rolling shutter door meets the fire protection specification requirements. If any parameter exceeds the preset threshold value interval, it is determined that the operating state of the rolling shutter door does not meet the fire protection specification requirements. The fire protection specification database contains the standard operating parameters of different types of rolling shutter doors. For conventional fireproof rolling shutter doors, the operating speed threshold is set in the range of 150 - 200 mm / s, the position deviation threshold is controlled within ±30 mm, and in terms of electrical parameters, the voltage fluctuation is required not to exceed ±7% of the rated value, the current fluctuation does not exceed ±10% of the rated value, and the power factor is not less than 0.85. The feature vector normalization uses the min-max scaling method to map all parameters to the 0 - 1 interval, eliminating the influence of dimension. During the data enhancement process, Gaussian noise is used to simulate sensor noise, and random offsets are used to simulate installation errors and mechanical wear. The number of training samples is expanded from 1000 groups to 5000 groups, and the 5-fold cross-validation method is adopted. Each fold contains 4000 groups of training data and 1000 groups of validation data. When the accuracy of the validation set reaches above the preset accuracy, it is considered that the model training is sufficient. The convolutional neural network adopts a multi-branch structure to process the speed feature, position feature, and electrical feature respectively. The speed branch uses 32 3x3 convolutional kernels to extract motion features, the position branch uses 16 5x5 convolutional kernels to extract spatial features, and the electrical branch uses 64 2x2 convolutional kernels to extract electrical parameter features. The convolutional kernel parameters are optimized by the backpropagation algorithm, and the initial value of the learning rate is set to 0.01, and it decays to 0.1 times the original value every 50 iterations. The operating state feature matrix contains three feature layers: the speed layer, the position layer, and the electrical layer, and the matrix dimension is 16x16x3. The fully connected layer flattens the feature matrix and calculates the compliance scores of the three types of parameters respectively, and the score range is 0 - 100 points. In practical applications, a speed parameter score lower than 60 points indicates abnormal operating speed, a position parameter score lower than 70 points indicates guide rail deformation or installation deviation, and an electrical parameter score lower than 80 points indicates a power supply failure.Through the identification and analysis of different types of faults, the effectiveness of multi-dimensional evaluation indicators is verified. Especially in preventive maintenance, the changing trend of parameter scores can early warn potential faults.

[0028] Analyze the input characteristic vectors of operating parameters through a convolutional neural network model, comprehensively evaluate the motion speed, position accuracy, current parameters and operating stability status indicators, output the evaluation results of each indicator, and compare the evaluation results with the preset speed range, position accuracy, current limit and stability requirements in the fire protection code. Calculate the overall compliance score through weighted calculation to determine whether the preset fire protection code standards are met.

[0029] Extract the speed waveform characteristics, position deviation characteristics, current change characteristics and operating stability characteristics from the operating parameter characteristic vectors through a convolutional neural network to obtain a speed evaluation data matrix, a position evaluation data matrix, a current evaluation data matrix and a stability evaluation data matrix; calculate the corresponding evaluation index values according to the speed evaluation data matrix, position evaluation data matrix, current evaluation data matrix and stability evaluation data matrix, and the evaluation index values include the speed mean and standard deviation, position deviation mean and maximum deviation, current fluctuation amplitude and frequency, fluctuation period and amplitude; calculate the corresponding compliance scores according to the preset standards of the fire protection code for the evaluation index values, and the compliance scores include speed compliance score, position compliance score, current compliance score and stability compliance score; use a feedforward neural network to assign benchmark weights to the speed compliance score, position compliance score, current compliance score and stability compliance score, and obtain the overall compliance score of the rolling door operating state through arithmetic mean operation.

[0030] Exemplarily, speed waveform features, position deviation features, current change features, and operation stability features are extracted from the operating parameter feature vectors through a convolutional neural network, and a speed evaluation data matrix, a position evaluation data matrix, a current evaluation data matrix, and a stability evaluation data matrix are respectively constructed. Calculate the mean and standard deviation of the motion speed for the speed evaluation data matrix, calculate the mean and maximum deviation of the position deviation for the position evaluation data matrix, calculate the amplitude and frequency of the current fluctuation for the current evaluation data matrix, and calculate the fluctuation period and amplitude for the stability evaluation data matrix. According to the preset standards of the fire protection code, calculate the speed compliance score for the mean and standard deviation of the speed, calculate the position compliance score for the mean and maximum deviation of the position deviation, calculate the current compliance score for the amplitude and frequency of the current fluctuation, and calculate the stability compliance score for the fluctuation period and amplitude. Use a feedforward neural network to construct a multi-layer weight processor, assign a reference weight to the speed compliance score, assign a reference weight to the position compliance score, assign a reference weight to the current compliance score, and assign a reference weight to the stability compliance score. Perform an arithmetic mean operation on the weighted scores to generate an overall compliance score for the operation status of the rolling shutter door. If the overall compliance score is greater than the preset qualified threshold, it is determined that the operation status of the rolling shutter door meets the requirements of the fire protection code. The evaluation of the operation status of the rolling shutter door involves the comprehensive analysis of multiple key parameters, and the convolutional neural network realizes the in-depth analysis of the operating parameters through multi-layer feature extraction. Taking a conventional fireproof rolling shutter door as an example, the speed waveform feature reflects the motion characteristics of the door body, the sampling frequency is set to 100 Hz, and the mean and standard deviation of the speed are extracted to evaluate the operation stability. The mean of the motion speed within the range of 150 - 200 mm / s indicates normal operation, and a standard deviation exceeding 20 mm / s indicates the presence of mechanical faults. In terms of position evaluation, the motion trajectory of the door body is recorded through a displacement sensor. The mean of the position deviation reflects the installation accuracy of the guide rail, and the maximum deviation reflects the degree of deformation of the guide rail. The measured data shows that the mean of the position deviation does not exceed 5 mm during normal operation, and the maximum deviation is controlled within 15 mm. The deformation or installation deviation of the guide rail will cause a sharp increase in the position deviation, and there is a risk of jamming when the maximum deviation exceeds 30 mm. The current evaluation focuses on the working state of the motor, and records the change characteristics of the starting current, running current, and braking current. The amplitude of the current fluctuation characterizes the load change, and the fluctuation frequency reflects the mechanical resistance. Under normal operating conditions, the peak value of the starting current does not exceed 2.5 times the rated current, the amplitude of the running current fluctuation is within ±10% of the rated value, and the fluctuation frequency is lower than 2 Hz. The increase in motor aging or mechanical friction will cause the running current to increase and the fluctuation frequency to increase. The evaluation of operation stability comprehensively considers the periodic characteristics of speed fluctuation, position deviation, and current change. The fluctuation period reflects the fault characteristic frequency, and the fluctuation amplitude characterizes the severity of the fault. Through the feedforward neural network to assign weights, the speed compliance weight is 0.35, the position compliance weight is 0.25, the current compliance weight is 0.25, and the stability compliance weight is 0.15.The overall compliance score is obtained by weighted averaging each score according to the weight, and the preset passing threshold is 85 points. Through the analysis of the evaluation data, it is found that mechanical failures often first reflect in position deviation and current characteristics, while the speed fluctuation significantly increases only in the later stage of the fault development.

[0031] S105. If the judgment result is non-compliant, perform an interpretability analysis on the convolutional neural network model to obtain the key operating parameters, retrieve the parameter optimization scheme corresponding to the key operating parameters in the preset rule base, evaluate the effects of each scheme on improving the model judgment result through simulation of the parameter optimization scheme, so as to select the target parameter adjustment scheme.

[0032] Perform a parameter sensitivity analysis on the convolutional neural network through the gradient backpropagation algorithm to obtain the neuron contribution value ranking table, and select the operating parameters corresponding to the key neurons from the contribution value ranking table; query the preset rule base according to the operating parameters corresponding to the key neurons to obtain the optimization rule set containing the operating parameters, and calculate the deviation degree between the rule value range and the current parameter value for the optimization rule set; sort the optimization rules according to the deviation degree, select the rule with the largest deviation degree from the sorting result as the candidate rule, and obtain the parameter adjustment interval in the candidate rule; use the Monte Carlo method to generate a test data set within the parameter adjustment interval, perform simulation verification on the test data set, and select the optimal parameter adjustment scheme according to the Euclidean distance of the verification result.

[0033] Exemplarily, the parameter sensitivity analysis of the convolutional neural network is carried out by the gradient backpropagation algorithm, the contribution value of each neuron to the judgment result is calculated, the top twenty percent of the neurons are selected as key neurons after sorting the contribution values, and the operating parameters corresponding to the key neurons are extracted to generate a key parameter set. For the operating parameters in the key parameter set, the optimization rules containing the parameter are retrieved from the preset rule library, the deviation degree between the parameter value range in the rule and the current parameter value is calculated, and a parameter - rule association table is generated. According to the deviation degree in the parameter - rule association table, the optimization rules are sorted by priority, the top five rules with the largest deviation degree are selected as candidate rules, and the parameter adjustment intervals included in the rules are recorded. The Monte Carlo sampling method is used to generate multiple groups of parameter adjustment values within the parameter adjustment intervals, a parameter update sequence is constructed for each group of adjustment values, and a simulation test data set is generated. The test data set is verified through the simulation environment, the Euclidean distance between the judgment result output value and the target value after each parameter adjustment is recorded, and an effect evaluation data set is generated. According to the effect evaluation data set, the combination of parameter adjustment values with the smallest Euclidean distance is selected as the optimal adjustment plan, and a target parameter adjustment plan is generated. The interpretability analysis of the convolutional neural network reveals the influence mechanism of the parameters on the judgment result through gradient backpropagation. In the operation state evaluation of a shopping mall's fire - proof rolling shutter door, the sensitivity analysis of the network output layer to the input parameters shows that among 500 neurons, 125 have a significant impact on the judgment result, and the contribution value exceeds 0.1. These key neurons are mainly distributed in the speed detection layer, position detection layer, and current detection layer, corresponding to three core parameters: running speed, position deviation, and current fluctuation. The rule library stores the optimization rules for different parameters. The rule for the speed parameter includes a speed range of 150 - 200 mm / s, an acceleration limit of 0.5 m / s², and a speed volatility less than 10%. The position parameter rule stipulates that the maximum deviation does not exceed 30 mm, the guide rail deformation is less than 5 mm, and the position repeatability accuracy is better than 2 mm. The current parameter rule includes a starting current multiple less than 2.5, an operating current volatility lower than 15%, and a power factor greater than 0.85. The deviation degree between the current parameter and the rule requirement is quantified through normalization, and the rules with a deviation degree greater than 0.3 are marked as high - priority. For the parameter adjustment intervals specified by the high - priority rules, the Monte Carlo method is used to generate 100 groups of candidate parameter values. The speed adjustment values are uniformly sampled in the range of 140 - 160 mm / s, the position compensation values are generated in the interval of - 10 mm to + 10 mm, and the current limit is adjusted in the range of 80% - 120% of the rated value. Each group of parameters runs for 50 cycles in the simulation environment, and the running trajectory data is recorded. The simulation test results are evaluated by the Euclidean distance to measure the closeness to the target state. In a certain test case, the original speed exceeds the specification requirements, the position deviation exceeds the limit, and the starting current peak reaches 3 times the rated value.Through parameter optimization, the running speed is adjusted to 155 mm / s, the position compensation is increased by 8 mm, the starting current multiple is reduced to 2.3 times, and the Euclidean distance is reduced from the initial 0.45 to 0.12. The selected parameter combination is verified to be effective in actual operation, and the speed, position, and current parameters all return to the specified range. The implementation process of the optimization rules also reveals the coupling relationship between parameters. The reduction in speed leads to a natural decrease in the starting current, and the improvement in position compensation reduces mechanical friction, thereby affecting the current fluctuation characteristics. Through the intelligent matching of the rule base and simulation verification, not only the compliance of individual parameters is ensured, but also the coordinated optimization of multiple parameters is achieved. Subsequent operation data shows that the optimized rolling shutter door moves more smoothly, the position accuracy is improved, the electrical characteristics are improved, and all indicators continuously meet the requirements of the fire protection code.

[0034] S106. Send the adjustment suggestion to the fire control center, generate an adjustment instruction for the multi-line control mode of the rolling shutter door, and send it to the rolling shutter door controller on-site through the communication module of the fire alarm system.

[0035] Generate control parameters in bytecode format according to the parameter adjustment scheme dataset. The control parameters include the running speed parameter value, the position deviation parameter value, and the current limit parameter value; use a cyclic redundancy checker to generate a check code for the control parameters in bytecode format, and generate a standard data frame sequence through a sharding processor. The standard data frame sequence is marked with a timestamp and a sequence number identifier; obtain the bus communication status data from the fire alarm communication module, calculate the communication reliability score for the bus communication status data through a status scorer, and select the bus with the highest communication reliability score as the main transmission channel; for the standard data frame sequence, perform fire protection equipment communication protocol encoding on the main transmission channel. If no controller feedback signal is received within the preset response time limit, resend the standard data frame sequence through the backup channel.

[0036] Exemplarily, according to the parameter adjustment scheme data set, the running speed parameter value, position deviation parameter value, and current limit parameter value are converted into byte code format in the instruction generator of the fire control center. The data frame header and data field are organized according to the fire controller communication specification to generate a rolling shutter control instruction data packet. For the control instruction data packet, a cyclic redundancy checker is used to generate a check code, a timestamp mark and a sequence number identifier are added, and the data packet is fragmented according to the communication specification requirements to generate a standard data frame sequence. The bus communication status is read from the fire alarm communication module, and the signal strength, bit error rate, and delay parameters of multiple control buses are sampled and statistically analyzed to generate communication quality evaluation data. According to the communication quality evaluation data, the communication reliability score of each bus is calculated by the status scorer, and the bus with the highest reliability score is selected as the main transmission channel, and the bus with the second highest score is selected as the backup channel. On the main transmission channel, the standard data frame sequence is encoded according to the fire equipment communication protocol, and the control instruction is sent to the on-site rolling shutter controller through the fire alarm communication module. The response monitoring is started for the sent control instruction. If the controller feedback signal is not received after exceeding the preset response time limit, the instruction is re-sent by switching to the backup channel, and the communication exception alarm is triggered after three consecutive transmission failures. The parameter adjustment instruction of the rolling shutter by the fire control center is encoded and transmitted using the standard communication specification. In the fire protection system of a high-rise building, the running speed parameter is converted into a 4-byte floating-point number, the position deviation threshold occupies 2 bytes of integer, and the current limit is represented by 2 bytes. The data frame header contains an 8-byte device address code, a 2-byte function code, and a 4-byte timestamp. The data field length is 32 bytes, and a 4-byte CRC check code is appended at the end. For the generated control instruction data packet, the CRC32 check algorithm calculates the check code 0xA5B6C7D8, and the timestamp accuracy reaches the millisecond level. Considering the transmission characteristics of the on-site communication bus, the data packet is divided into 4 data frames, and a 2-byte sequence number identifier is added to each frame to indicate the transmission order. This fragmentation strategy not only ensures the real-time performance of single-frame data but also avoids the risk of data loss during long-frame transmission. The fire alarm system usually adopts a dual-bus redundant configuration, and the main and backup buses run in parallel. Through the communication quality monitoring module, the bus parameters are sampled once. The signal strength is represented by the relative level, and a signal above -10dB is a high-quality signal; the bit error rate is based on the transmission statistics of the previous minute, and the threshold is set to 10^-6; the end-to-end delay reflects the signal transmission time, and it should be less than 20ms under normal circumstances. In practical applications, the signal strength of the main bus is -8dB, the bit error rate is 3×10^-7, and the delay is 15ms, and the reliability score is 92 points. The parameters of the backup bus are -12dB, 8×10^-7, and 18ms respectively, and the score is 85 points. The system preferentially selects the main bus for instruction transmission while maintaining real-time monitoring of the backup bus. When a communication exception occurs, the instruction response timeout is set, and after the timeout, it immediately switches to the backup bus to re-send, and at most three retries are allowed.During the transmission of a certain adjustment instruction, the main bus was interfered by the external environment, resulting in a sudden drop in signal strength. The instruction sent for the first time did not receive a response from the controller. The system detected the timeout and automatically switched to the standby bus to retransmit the instruction. After one retransmission, the standby bus successfully received the response signal from the controller, confirming that the parameter adjustment had been executed. The time consumed for the switching and retransmission processes met the real-time requirements of the control instruction. Through this dual-bus redundant transmission mechanism, even if a single bus fails, the reliable delivery of the control instruction can be ensured.

[0037] S107. The rolling door controller dynamically adjusts the operating parameters of the control circuit according to the adjustment instruction to optimize the closing process of the rolling door.

[0038] The rolling door controller parses the received adjustment instruction, extracts the operating speed parameter value, position deviation parameter value, and current limit parameter value from the adjustment instruction, and generates a parameter adjustment sequence in the data buffer; establishes a speed adjustment subroutine, a position adjustment subroutine, and a current adjustment subroutine according to the parameter adjustment sequence, and optimizes the adjustment step size and adjustment period of the adjustment subroutine by using a fuzzy controller to obtain an optimized control signal; for the optimized control signal, modulates the amplitude, frequency, and phase of the motor drive voltage through a pulse width modulator to form a motor drive waveform, and uses a neural network predictor to generate predicted motion trajectory data for the motor drive waveform; collects the actual operating parameters of the rolling door through a position sensor, a speed sensor, and a current sensor, compares the actual operating parameters with the predicted motion trajectory data, and compensates and corrects the motor drive waveform according to the parameter deviation value obtained from the comparison.

[0039] Exemplarily, the rolling shutter door controller parses the received adjustment instruction, extracts the running speed parameter value, position deviation parameter value, and current limit parameter value from the instruction data frame, records the difference between the current parameter and the target parameter in the data buffer, and generates a parameter adjustment sequence. According to the parameter adjustment sequence, a speed adjustment subroutine, a position adjustment subroutine, and a current adjustment subroutine are established in the motor driver, and the adjustment step and adjustment period of each parameter are calculated respectively. A fuzzy controller is used to optimize the speed adjustment step and adjustment period online, and the control gain coefficient is dynamically adjusted according to the response characteristics during the parameter adjustment process to generate an optimized control signal. For the optimized control signal, the amplitude, frequency, and phase of the motor drive voltage are modulated in real time through a pulse width modulator to form a motor drive waveform. A neural network predictor is used to predict the motion response after the motor drive waveform is applied to the rolling shutter door, and generate predicted motion trajectory data. The actual operating parameters of the rolling shutter door are collected through a position sensor, a speed sensor, and a current sensor, and the measured data is compared with the predicted trajectory data to calculate the parameter deviation value. According to the parameter deviation value, the amplitude coefficient, frequency coefficient, and phase coefficient of the motor drive waveform are compensated and corrected, and the motor control parameters are updated to form a new drive waveform. The dynamic adjustment of the rolling shutter door control instruction involves parameter optimization at multiple levels. In the control of the fire shutter door in a certain high-rise building, after receiving the adjustment instruction, the controller extracts key parameters such as the target running speed of 175 mm / s, the position deviation threshold of ±20 mm, and the current limit of 15 A. The current speed is 220 mm / s, the position deviation reaches 35 mm, and the operating current fluctuates between 16 - 18 A. The parameter adjustment sequence records these difference data. The motor driver sets the adjustment step according to the parameter difference. The speed is adjusted by 5 mm / s each time, and the adjustment period is 200 ms; the position compensation amount is 2 mm, and the compensation period is 100 ms; the current adjustment step is 0.5 A, and the adjustment period is 50 ms. This hierarchical adjustment strategy avoids the mutual interference between parameters and ensures the smoothness of the adjustment process. The fuzzy controller uses 7 fuzzy rules to optimize the adjustment parameters. When the speed deviation is greater than 30 mm / s, the control gain is set to 0.8; when the deviation is in the range of 10 - 30 mm / s, the gain is set to 0.5; when the deviation is less than 10 mm / s, the gain is reduced to 0.3. Through the dynamic adjustment of the gain coefficient, the balance between the parameter convergence speed and the adjustment smoothness is achieved. The pulse width modulation uses a carrier frequency of 4 kHz, a voltage modulation depth of 0.85, and a phase lead angle of 15 degrees. During the speed regulation process, the voltage amplitude is adjusted by changing the duty cycle. The duty cycle gradually decreases from 0.8 to 0.65, so that the motor output torque decreases smoothly. At the same time, the frequency decreases from 50 Hz to 42 Hz to achieve dynamic speed adjustment.The neural network predictor trains a motion response model based on historical operation data. The prediction shows that it takes 3 adjustment cycles for the display speed to drop from 220 mm / s to 175 mm / s, and the overshoot is about 3%; it takes 5 cycles for the position deviation to converge from 35 mm to within 20 mm; the current setting process takes about 4 cycles to stabilize below 15 A. The real-time monitoring data of multiple sensors shows that during the actual adjustment process, the speed drop curve is flatter than the predicted value and takes 4 cycles; the position compensation effect is better than the prediction, and the deviation is reduced to 18 mm after 3 cycles; the current setting meets the expectations and stabilizes at about 14.5 A after 4 cycles. Based on the measured data, the controller compensates the drive waveform, fine-tunes the duty cycle to 0.67, adjusts the frequency to 43 Hz, and keeps the phase angle unchanged, making the operating parameters closer to the target values. After dynamic optimization, the running speed of the rolling shutter door is stable within the range of 172 - 178 mm / s, the position deviation is controlled within ±15 mm, and the fluctuation range of the running current does not exceed 0.8 A. All indicators meet the requirements of the fire protection code. Multiple test verifications show that this parameter optimization method based on multi-layer feedback has good adaptability and reliability.

[0040] The above description is only a preferred embodiment of the present application and an explanation of the applied technical principles. Those skilled in the art should understand that the scope of the invention involved in the present application is not limited to the technical solution formed by the specific combination of the above technical features, but should also cover other technical solutions formed by any combination of the above technical features or their equivalent features without departing from the concept of the present application. For example, the technical solutions formed by mutually replacing the above features with the (but not limited to) technical features with similar functions disclosed in the present application.

Claims

1. A method for checking the control mode of important equipment in an automatic fire alarm system, characterized in that: The method comprises: The automatic fire alarm system collects speed and stress data during the closing process of the rolling shutter door under multi-line control mode, draws speed stress curves, and obtains time series data of key control nodes of segmented landing and secondary reset; Perform wavelet transform on the time series data of key control nodes, extract time-frequency domain features, classify the time-frequency domain features, and analyze the operation of the rolling door, including speed mutation and position deviation; Obtain the operating parameters of the rolling door control line in the time period corresponding to the speed mutation and position deviation, the operating parameters including current, voltage, and power, and construct a multi-dimensional operating parameter feature vector based on the operating parameters; The operating parameter feature vector is input into the pre-trained convolutional neural network model to evaluate the operating status of the rolling door multi-line control mode and determine whether it meets the preset fire protection regulations; If the judgment result is not in compliance, the convolutional neural network model is analyzed for interpretability to obtain key operating parameters, and the parameter optimization scheme corresponding to the key operating parameters is retrieved from the preset rule library. The parameter optimization scheme is simulated to evaluate the effect of each scheme on improving the model judgment result, so as to select the target parameter adjustment scheme; The adjustment suggestions are sent to the fire control center, and the adjustment instructions for the rolling shutter door multi-line control mode are generated and sent to the rolling shutter door controller on site through the communication module of the automatic fire alarm system; The rolling door controller dynamically adjusts the operating parameters of the control circuit according to the adjustment instructions to optimize the closing process of the rolling door.

2. The method according to claim 1, characterized in that: The automatic fire alarm system collects speed and stress data during the closing process of the rolling door under the multi-line control mode, draws a speed stress curve, and obtains time series data of key control nodes of segmented landing and secondary reset, including: The position data of the rolling shutter door is collected by the displacement sensor, the movement speed data is collected by the speed sensor, and the force data is collected by the stress sensor to form a data packet of the movement state of the rolling shutter door; For the motion state data packet, the multi-threaded controller establishes a motion characteristic curve and a force characteristic curve to form a velocity-stress correlation data set; The speed-stress correlation data set is trained by using a self-learning convolutional neural network to obtain the inflection point of the rolling shutter door speed change as the segmented landing threshold point, and the inflection point of the stress change as the stress threshold point of the landing process; A time series analysis is performed on the segmented descent threshold points and the stress threshold points, and the time interval data between the segmented descent threshold points and the stress threshold points are calculated to obtain a segmented descent control data set for the rolling door.

3. The method according to claim 1, characterized in that The time series data of the key control nodes are subjected to wavelet transformation, time-frequency domain features are extracted, the time-frequency domain features are classified, and the operation of the rolling door is analyzed, the operation including speed mutation and position deviation, including: The motion displacement and speed change data of the rolling shutter door are obtained through the Hall sensor, and the motion displacement and speed change data are time-stamped in the data collector to obtain a time series state data set; Performing wavelet transform on the time series state data set, extracting high-frequency coefficients and low-frequency coefficients from the wavelet coefficient matrix, and reconstructing the high-frequency coefficients and low-frequency coefficients to obtain basic features in the time-frequency domain; Calculate the statistics of signal segment variance, mean, kurtosis and skewness based on the basic features in the time-frequency domain, perform principal component analysis on the statistics, and extract the eigenvector with the largest contribution rate as the dimensionality reduction feature; A feature classifier is constructed using the support vector machine algorithm to classify the dimensionality reduction features to obtain speed change rate features and position offset features. If the speed change rate is greater than the speed threshold or the position offset is greater than the position threshold, it is determined that the rolling door is abnormal.

4. The method according to claim 1, characterized in that: The operation parameters of the rolling door control circuit in the time period corresponding to the speed mutation and the position deviation are obtained, and the operation parameters include current, voltage, and power. A multi-dimensional operation parameter feature vector is constructed according to the operation parameters, including: The current signal, voltage signal and power signal are collected from the rolling door control circuit through the circuit monitor, and the collected signals are filtered to obtain the original data set of the operating parameters; According to the timestamps of the speed mutation points and the position deviation points in the original operating parameter data set, data segments are intercepted, and the data segments are normalized to obtain a standardized operating parameter data set; Calculating the current change rate, voltage change rate and power change rate for the standardized operating parameter data set, and obtaining a parameter change characteristic curve by smoothing through a Gaussian filter; After using a deep neural network to extract the fluctuation frequency characteristics, peak characteristics and root mean square characteristics from the parameter change characteristic curve, principal component analysis and dimensionality reduction fusion are performed to obtain a multidimensional feature vector of the operating parameters.

5. The method according to claim 1, characterized in that: The operating parameter feature vector is input into the pre-trained convolutional neural network model to evaluate the operating status of the rolling door multi-line control mode to determine whether it meets the preset fire protection regulations, including: According to the rolling door operation parameter feature vector, the speed threshold, position deviation threshold and electrical parameter threshold are read from the fire protection specification database, and standardized feature data is generated through a data preprocessor; Based on the standardized feature data, a convolutional neural network is used to extract the speed feature, position feature and electrical feature of the rolling door to generate an operating state feature matrix; Calculating the specification compliance scores of speed parameters, position parameters and electrical parameters for the operating state feature matrix through a fully connected layer; According to the compliance score, a speed parameter threshold interval, a position parameter threshold interval and an electrical parameter threshold interval are set. If the speed parameter, the position parameter and the electrical parameter are all within the preset threshold intervals, it is determined that the operating state of the rolling door meets the requirements of the fire protection regulations. It also includes: analyzing the input operating parameter feature vector through a convolutional neural network model, comprehensively evaluating the movement speed, position accuracy, current parameters and operating stability status indicators, outputting the evaluation results of each indicator, and comparing the evaluation results with the speed range, position accuracy, current limit and stability requirements preset in the fire protection regulations, and obtaining an overall compliance score through weighted calculation to determine whether the preset fire protection regulations are met.

6. The method according to claim 5, characterized in that The convolutional neural network model analyzes the input operating parameter feature vector, comprehensively evaluates the motion speed, position accuracy, current parameters and operating stability status indicators, outputs the evaluation results of each indicator, and compares the evaluation results with the speed range, position accuracy, current limit and stability requirements preset in the fire protection regulations, and obtains the overall compliance score through weighted calculation to determine whether the preset fire protection regulations are met, including: The speed waveform features, position deviation features, current change features and operation stability features are extracted from the operating parameter feature vector by using a convolutional neural network to obtain a speed evaluation data matrix, a position evaluation data matrix, a current evaluation data matrix and a stability evaluation data matrix; Calculate corresponding evaluation index values ​​according to the speed evaluation data matrix, the position evaluation data matrix, the current evaluation data matrix and the stability evaluation data matrix, wherein the evaluation index values ​​include speed mean and standard deviation, position deviation mean and maximum deviation, current fluctuation amplitude and frequency, and fluctuation period and amplitude; Calculate the corresponding compliance score for the evaluation index value according to the preset standard of the fire protection code, and the compliance score includes the speed compliance score, the position compliance score, the current compliance score and the stability compliance score; A feedforward neural network is used to assign reference weights to the speed compliance score, position compliance score, current compliance score and stability compliance score, and the overall compliance score of the rolling door operation state is obtained through arithmetic average operation.

7. The method according to claim 1, characterized in that If the judgment result is not in compliance, the convolutional neural network model is subjected to interpretability analysis to obtain key operating parameters, and a parameter optimization scheme corresponding to the key operating parameters is retrieved from a preset rule base. The parameter optimization scheme is simulated to evaluate the effect of each scheme on improving the model judgment result, so as to select a target parameter adjustment scheme, including: Performing parameter sensitivity analysis on the convolutional neural network through the gradient back propagation algorithm to obtain a neuron contribution value ranking table, and selecting operating parameters corresponding to key neurons from the contribution value ranking table; Querying a preset rule library according to the operating parameters corresponding to the key neurons, obtaining an optimization rule set including the operating parameters, and calculating the degree of deviation between the rule value range and the current parameter value for the optimization rule set; Sorting the optimization rules according to the degree of deviation, selecting the rule with the largest degree of deviation from the sorting result as the candidate rule, and obtaining the parameter adjustment interval in the candidate rule; A Monte Carlo method is used to generate a test data set within the parameter adjustment interval, a simulation verification is performed on the test data set, and an optimal parameter adjustment scheme is selected according to the Euclidean distance of the verification result.

8. The method according to claim 1, characterized in that The adjustment suggestion is sent to the fire control center, and the adjustment instruction of the rolling shutter door multi-line control mode is generated, and sent to the rolling shutter door controller on site through the communication module of the automatic fire alarm system, including: Generate control parameters in bytecode format according to the parameter adjustment scheme data set, wherein the control parameters include a running speed parameter value, a position deviation parameter value, and a current limit parameter value; A cyclic redundancy checker is used to generate a check code for the control parameter in the bytecode format, and a standard data frame sequence is generated through a slice processor, wherein the standard data frame sequence has a timestamp mark and a sequence number identifier; Obtain bus communication status data from the automatic fire alarm communication module, calculate the communication reliability score of the bus communication status data through a status scorer, and select the bus with the highest communication reliability score as the main transmission channel; For the standard data frame sequence, the fire equipment communication protocol encoding is performed on the main transmission channel. If the controller feedback signal is not received within the preset response time limit, the standard data frame sequence is resent through the backup channel.

9. The method according to claim 1, characterized in that: The rolling door controller dynamically adjusts the operating parameters of the control circuit according to the adjustment instruction to optimize the closing process of the rolling door, including: The rolling door controller parses the received adjustment instruction, extracts the operating speed parameter value, the position deviation parameter value and the current limit parameter value from the adjustment instruction, and generates a parameter adjustment sequence in a data buffer; According to the parameter adjustment sequence, a speed adjustment subroutine, a position adjustment subroutine and a current adjustment subroutine are established, and a fuzzy controller is used to optimize the adjustment step length and adjustment cycle of the adjustment subroutine to obtain an optimized control signal; For the optimized control signal, the amplitude, frequency and phase of the motor drive voltage are modulated by a pulse width modulator to form a motor drive waveform, and a neural network predictor is used to generate predicted motion trajectory data for the motor drive waveform; The actual operating parameters of the rolling door are collected by the position sensor, the speed sensor and the current sensor, the actual operating parameters are compared with the predicted motion trajectory data, and the motor drive waveform is compensated and corrected according to the parameter deviation value obtained by the comparison.

Images