Intelligent management and control method and system of tunnel telephone broadcast system

Abstract

The present application relates to the technical field of tunnel information, in particular to an intelligent management and control method and system of a tunnel telephone broadcast system; through collecting on-site audio and video data, using an event classification model to identify event labels and urgency, and combining with section dynamic weight calculation to comprehensively determine priority, multiple emergency calls are intelligently sorted to ensure that high-risk events are handled first and are not missed. The audio and video data are automatically uploaded to the cloud, and the management personnel can remotely listen to the on-site environment sound of any telephone terminal to discover abnormalities (such as quarrels and equipment abnormal sound) in advance and improve the early warning capability. Voice recognition is performed on the connected call, event entity combinations are extracted, the most matched broadcast template is automatically recommended based on a broadcast template prediction model, the cumbersome process of management personnel switching between multiple systems is reduced, and the emergency response efficiency is significantly improved. The present application realizes the cooperation of multiple systems such as emergency telephone, broadcast and monitoring, and improves the intelligent level and disposal speed of tunnel emergency management.

Description

Technical Field

[0001] This invention relates to the field of tunnel information technology, specifically to an intelligent management and control method and system for a tunnel telephone broadcasting system. Background Technology

[0002] The tunnel's telephone broadcasting system, also known as the tunnel emergency broadcasting system or tunnel emergency broadcasting system, is an important facility used by tunnel management agencies to disseminate emergency voice information to vehicles and personnel within the tunnel. It consists of a core control system, transmission network, hardware equipment (transmitters, receivers, speakers, microphones, etc.), and software platform. Its main function is to enable the system to quickly activate in the event of emergencies such as traffic accidents or fires within the tunnel, issuing escape instructions to vehicles and personnel and guiding them to safe evacuation.

[0003] Currently, the telephone, broadcasting, monitoring, and sensor systems within the tunnel are mostly independently operating emergency telephone and broadcasting systems, lacking a unified management platform. This results in the absence of critical functions in the telephone broadcasting system, such as:

[0004] (1) When multiple alarms come in at the same time, they cannot be intelligently sorted by urgency or location. Operators can only answer them one by one, which may cause high-priority events to be missed. (2) It is not possible to remotely monitor the ambient sound of any telephone terminal in the tunnel, making it difficult to detect abnormalities in advance (such as arguments or abnormal equipment noises). (3) Operators need to select broadcast content and broadcast location based on the results of telephone calls, which requires operators to switch between different systems, making it cumbersome. Summary of the Invention

[0005] In view of this, the purpose of the present invention is to provide an intelligent management and control method and system for a tunnel telephone broadcasting system to solve the above-mentioned technical problems.

[0006] To achieve the above objectives, the present invention adopts the following technical solution:

[0007] The present invention provides an intelligent control method for a tunnel telephone broadcasting system, comprising the following steps:

[0008] When receiving emergency calls from multiple tunnel extensions simultaneously, the system acquires audio and video data collected by monitoring equipment within the group to which the tunnel extension is located. The collected audio and video data is added to an upload queue and uploaded to the cloud for backup.

[0009] The audio and video data are identified based on a pre-built event classification model to obtain the event tag of the location of the tunnel unit and the urgency of the event tag. The event tag includes fire, traffic accident, personnel distress call, equipment failure and no failure.

[0010] The urgency of the event tags of each tunnel extension is weighted based on the segment weight of the segment where the multiple tunnel extensions are located, so as to obtain the comprehensive priority of multiple emergency calls.

[0011] Based on the comprehensive priority, multiple emergency calls are sorted, and voice recognition and event entity extraction are performed on the connected emergency calls to obtain the event entity combination of the current emergency call;

[0012] Based on a pre-built broadcast template prediction probability model and the event entity combination, the probability of various broadcast templates required by the administrator is predicted, and the one or more broadcast modules with the highest probability are sent to the administrator.

[0013] In one embodiment of this application, the network structure of the event classification model includes:

[0014] The video coding branch is used to extract spatiotemporal features from video data based on pre-trained 3DCNN modules, thus obtaining video features. ;

[0015] The audio coding branch is used to extract the spectrogram of the audio data based on the pre-trained VGGish module, thereby obtaining audio features. ;

[0016] A multimodal fusion layer is used to combine the video features. and the audio features By fusing the features, we obtain the joint characteristics. ; and based on the fully connected layer, the joint features Dimensionality reduction is performed to obtain dimensionality-reduced features. ;

[0017] Multi-task output header, used for dimensionality reduction features Predict event labels and based on the dimensionality reduction features Output the urgency level.

[0018] In one embodiment of this application, the process of constructing the event classification model includes:

[0019] S1, Obtain audio and video data samples and labels for the audio and video data, wherein the labels include event labels and urgency labels;

[0020] S2, the audio and video data samples are split into audio data samples and video data samples, and the audio data samples and video data samples are respectively input into the audio encoding branch and the video encoding branch to obtain the predicted event label and predicted urgency output by the multi-task output head;

[0021] S3, calculate the classification loss between the predicted event label and the labeled event label based on the pre-built classification loss function, and calculate the regression loss between the predicted urgency and the labeled urgency based on the pre-built regression loss function;

[0022] S4, Based on the backpropagation of the classification loss and the regression loss, the internal parameters of the network structure of the event classification model are adjusted;

[0023] S5. Repeat steps S2-S4 until training is complete and the event classification model is obtained.

[0024] In one embodiment of this application, the method for determining the segment weight of the section where multiple tunnel extension units are located includes:

[0025] The static and dynamic parameters of multiple sections are obtained. The static parameters include slope, curve radius and historical accident rate. The dynamic parameters include average vehicle speed, traffic flow, time occupancy, CO concentration, smoke concentration and rainfall intensity. The dynamic parameters are the statistical average values ​​of the target duration before the current time point.

[0026] The static parameters and the dynamic parameters are normalized to obtain normalized static parameters and normalized dynamic parameters, wherein the values ​​of the normalized static parameters and the normalized dynamic parameters are positively correlated with the risk.

[0027] Static weights are obtained by weighting multiple normalized static parameters based on pre-built static parameter weights; and dynamic weights are obtained by weighting multiple normalized dynamic parameters based on pre-built dynamic parameter weights.

[0028] The static weights and dynamic weights are fused to obtain the segment weights for the sections where multiple tunnel extension units are located. .

[0029] In one embodiment of this application, the static parameter weights are determined based on expert evaluation, and the method for determining the dynamic parameter weights includes:

[0030] Calculate the first The section in the first The proportion of each dynamic parameter , wherein the specific gravity The mathematical expression is:

[0031]

[0032] In the formula, Indicates the first A normalized dynamic parameter, Indicates the number of segments;

[0033] Based on the specific gravity Calculate the first Information entropy of each normalized dynamic parameter And calculate the information entropy. Calculate the first The difference coefficient of each normalized dynamic parameter , where the first Information entropy of each normalized dynamic parameter and coefficient of difference The mathematical expressions are as follows:

[0034]

[0035]

[0036] Difference coefficients for multiple dynamic parameters Normalization is performed to obtain the dynamic parameter weights of multiple dynamic parameters. .

[0037] In one embodiment of this application, speech recognition and event entity extraction are performed on the connected emergency call to obtain the event entity combination of the current emergency call, including:

[0038] When an emergency call is connected, real-time voice data is collected, and the real-time voice data is recognized to obtain the recognized text.

[0039] The identified text is segmented into multiple words;

[0040] The multiple words are converted into word vectors, and the cosine similarity between the word vectors and any event entity in the pre-built event entity library is calculated.

[0041] When the cosine similarity between the word vector and any event entity is greater than or equal to a preset similarity threshold, the word vector is used as an event entity, and an event entity combination is constructed based on all event entities extracted during the dialogue process, wherein each event entity combination includes two event entities.

[0042] In one embodiment of this application, the method for constructing the broadcast template prediction probability model includes:

[0043] Acquire multiple historical emergency call event records, wherein the historical emergency call event records include event entity combination samples extracted from voice samples. And the broadcast template ultimately selected by the management;

[0044] Statistical analysis of each event entity combination sample from multiple historical emergency call event records. The frequency of occurrence was used to determine the M most frequent event entity combinations, which were then selected as key event entity combinations. And merge other event entity combination samples as low-frequency entity combinations. ;

[0045] Constructing a portfolio of key event entities conditional probability and low-frequency physical combinations conditional probability , wherein the conditional probability and the conditional probability The mathematical expression is:

[0046]

[0047]

[0048] In the formula, This indicates that when event entities are combined as Time broadcast template The probability of being selected Indicates a broadcast template. Represents a combination of key event entities Total number of occurrences For smoothing coefficients, Represents a template set. Indicates the number of broadcast templates. This indicates that when event entities are combined as Time broadcast template The number of times it was selected; This indicates that when event entities are combined as Time broadcast template The probability of being selected Represents a combination of low-frequency event entities Total number of occurrences This indicates that when event entities are combined as Time broadcast template The number of times it was selected;

[0049] A broadcast template prediction probability model is constructed based on the conditional probability of multiple event entity combinations.

[0050] In one embodiment of this application, the probability of predicting multiple broadcast templates required by the administrator is based on a pre-built broadcast template prediction probability model and the event entity combination prediction, including:

[0051] Substitute the combination of the multiple event entities into the pre-built broadcast template prediction probability model to obtain the conditional probability of the combination of multiple event entities;

[0052] The probability of each broadcast module is extracted from the conditional probability of multiple event entity combinations, and the probabilities of each broadcast module are summed to obtain the total probability of each broadcast module.

[0053] In one embodiment of this application, it further includes:

[0054] The system acquires the broadcast status of multiple sections within the tunnel and feeds back the broadcast status and real-time playback content to the management personnel. The broadcast status includes directional broadcast, FM broadcast, or no broadcast.

[0055] This application also provides an intelligent management and control system for a tunnel telephone broadcasting system, including:

[0056] The acquisition module is used to acquire audio and video data collected by the monitoring equipment in the group where the tunnel unit is located when receiving emergency calls from multiple tunnel units at the same time. The acquired audio and video data is added to the upload queue and uploaded to the cloud for backup.

[0057] The identification module is used to identify the audio and video data based on a pre-built event classification model to obtain the event tag of the location of the tunnel unit and the urgency of the event tag. The event tag includes fire, traffic accident, personnel distress call, equipment failure and no failure.

[0058] The priority calculation module is used to weight the urgency of the event tags of each tunnel extension based on the segment weight of the segment where the multiple tunnel extensions are located, so as to obtain the comprehensive priority of multiple emergency calls.

[0059] The entity extraction module is used to sort multiple emergency calls based on the comprehensive priority, and to perform speech recognition and event entity extraction on the connected emergency calls to obtain the event entity combination of the current emergency call.

[0060] The broadcast mapping module is used to predict the probability of multiple broadcast templates required by the administrator based on a pre-built broadcast template prediction probability model and the combination of event entities, and send the one or more broadcast templates with the highest probability to the administrator.

[0061] The beneficial effects of this invention are as follows: This invention provides an intelligent management and control method and system for a tunnel telephone broadcasting system. Addressing the shortcomings of existing tunnel telephone broadcasting systems that operate independently and lack unified management, this method collects on-site audio and video data, uses an event classification model to identify event tags and urgency levels, and calculates comprehensive priority based on dynamic weights for different sections. This intelligently sorts multiple emergency calls, ensuring high-risk events are handled first and avoiding omissions. Audio and video data is automatically uploaded to the cloud, allowing managers to remotely monitor ambient sounds at any telephone terminal, detecting anomalies (such as arguments or equipment noises) in advance and improving early warning capabilities. Voice recognition is performed on connected calls to extract event entity combinations. Based on a broadcast template prediction model, the most suitable broadcast template is automatically recommended, reducing the cumbersome process of switching between multiple systems for managers and significantly improving emergency response efficiency. This achieves collaboration among multiple systems, including emergency telephones, broadcasts, and monitoring, enhancing the intelligence level and speed of emergency management in tunnels. Attached Figure Description

[0062] Figure 1 This is a functional information architecture diagram of an intelligent management and control system for a tunnel telephone broadcasting system, as shown in one embodiment of this application.

[0063] Figure 2 This is a flowchart illustrating the process of a single emergency call in one embodiment of this application;

[0064] Figure 3 This is a flowchart illustrating an intelligent control method for a tunnel telephone broadcasting system in one embodiment of this application;

[0065] Figure 4 This is a structural diagram of an intelligent control system for a tunnel telephone broadcasting system, as shown in one embodiment of this application.

[0066] Figure 5 A schematic diagram of the structure of a computer system suitable for implementing the electronic device of the present application is shown. Detailed Implementation

[0067] The following specific examples illustrate the implementation of the present invention. Those skilled in the art can easily understand other advantages and effects of the present invention from the content disclosed in this specification. The present invention can also be implemented or applied through other different specific embodiments, and various details in this specification can also be modified or changed based on different viewpoints and applications without departing from the spirit of the present invention. It should be noted that, unless otherwise specified, the following embodiments and features described therein can be combined with each other.

[0068] It should be noted that the illustrations provided in the following embodiments are only schematic representations of the basic concept of the present invention. Therefore, the drawings only show the layers related to the present invention and are not drawn according to the actual number, shape and size ratio of the layers in the actual implementation. In the actual implementation, the form and number of each layer can be arbitrarily changed, and the layer layout may also be more complex.

[0069] Numerous details are explored in the following description to provide a more thorough explanation of embodiments of the invention; however, it will be apparent to those skilled in the art that embodiments of the invention may be practiced without these specific details.

[0070] This application consists of two parts: system architecture and control methods. Figure 1 This is a functional information architecture diagram of an intelligent management and control system for a tunnel telephone broadcasting system, as shown in one embodiment of this application. Figure 1 As shown, the system architecture includes:

[0071] Intelligent Management and Control Platform: This platform integrates multiple subsystems such as emergency telephones, directional broadcasts, and FM broadcasts into a unified management and control platform, enabling centralized management and control of the tunnel's communication and broadcasting systems. Operators can operate and monitor all systems from a single interface, improving work efficiency and reducing the possibility of operational errors.

[0072] I. Emergency Telephone Module:

[0073] 1. Phone List: The phone icons are distributed according to the tunnel phone numbers, showing whether the phone has incoming or connected status.

[0074] 2. Incoming call alarm: When an emergency call is received, the system will issue a global notification, highlighting the alarm information, and simultaneously connect the video monitoring of two nearby cameras, supporting functions such as PTZ control.

[0075] 3. Incoming call queue: When multiple alarm calls are received simultaneously, the calls are displayed in a queue, and you can switch between them to view them.

[0076] 4. Audio recording and monitoring: The tunnel management station can monitor the ambient sound of any terminal in the tunnel.

[0077] 5. Call Logs: An audio recording is generated after the call ends. You can view historical call records, and the recordings can be played back and downloaded. The audio recordings are stored for at least 3 months.

[0078] 6. Intercom function: It enables one-click calling from the emergency telephone box in the tunnel to the tunnel management station for two-way communication; the monitoring center can also call the emergency intercom in the tunnel and automatically answer the two-way call.

[0079] II. Directional Broadcast Module:

[0080] 1. Broadcast Terminal List: The corresponding broadcast device icons are distributed according to the tunnel broadcast terminals, indicating whether content is currently playing.

[0081] 2. Public address function: The tunnel management station can directly dial the broadcast number to broadcast messages to all emergency telephone broadcast terminals in the area.

[0082] 3. Scheduled Tasks: Broadcast supports scheduled task management, dividing into schemes and tasks (scheme-task is one-to-many); you can choose to execute or stop a scheme / task; and you can set priority, sequential / random playback, single / loop playback, volume control, start and end time, playback content (text TTS / media URL / radio station) and other options for individual tasks.

[0083] 4. Real-time tasks: Supports selecting broadcast playback tasks, and supports setting sequential / random playback, single / loop playback, text TTS / media URL / radio / real-time voice-over playback, volume control, etc.

[0084] Media Management: For text-based TTS / audio file sources, you can preset text and music files and support adding, editing, deleting, listening, and downloading operations.

[0085] 5. Partition / Group Management: Supports refining tunnels into partitions / groups and associating them with devices. When issuing commands, partitions of the devices can be selected. Partitions support adding, editing, and deleting operations.

[0086] 6. Terminal Management: Supports adding, editing, and deleting broadcast terminals and acquisition terminals, and viewing terminal status, etc.

[0087] III. FM Full-Frequency Coverage Emergency Broadcast Module:

[0088] 1. FM Broadcast Management: By using the scheduled or real-time task function of the broadcast module, selecting the "FM Broadcast Insertion Host" of the broadcast terminal enables the output signal from the IP module to the FM broadcast insertion host, which transmits the signal to the FM power amplifier and transmitting antenna inside the tunnel, ultimately outputting an FM broadcast signal for the vehicle radio to listen to.

[0089] 2. Equipment function management and parameter settings: The playback status of the "FM radio frequency insertion host" can be modified in real time, and its daily operation strategy can be set and managed through the "timed tasks" of the broadcast module, the "event management" and "contingency plan management" at the system level.

[0090] IV. New software configuration function:

[0091] 1. Video surveillance module: Provides real-time video streaming for tunnel monitoring, and supports camera switching and PTZ control (if the device supports it).

[0092] 2. System Management Module: Through the supplier equipment list, add information such as tunnel direction and station number, and enter information on emergency telephone, directional broadcast, and FM broadcast equipment; record the object, action, operator, and time of equipment control operations.

[0093] 3. Contingency Plan Management Module: Configure vehicle collision contingency plans and illegal parking contingency plans, and can link with the directional broadcast system to automatically play preset text and music files.

[0094] Figure 2 This is a flowchart of a single emergency call in one embodiment of this application, as shown below. Figure 2 As shown, the workflow for an incoming call from a single tunnel extension includes:

[0095] Call Initiation: The emergency extension initiates a call to the monitoring center. The emergency call platform receives the request, updates the extension status to "ringing", and automatically calls the duty phone (monitoring extension).

[0096] Ringing and video linkage: When the duty phone rings, the platform simultaneously retrieves real-time surveillance footage from the vicinity of the incident location and begins monitoring the scene.

[0097] Call handling:

[0098] Successful call: The on-duty staff picks up the phone, the platform updates the extension status to "in call," recording begins, and both parties begin the conversation. After the call ends, the staff hangs up, the platform updates the status to "hanged up," the live video feed stops, and the recorded audio file is generated and stored.

[0099] Call failed: If the on-duty staff does not answer the phone (timeout or rejection), the platform updates the extension status to "not connected", stops the live video feed, and does not generate a recording.

[0100] End: Regardless of whether the call is successfully answered or not, the platform records the call status and corresponding data for later tracking.

[0101] Figure 3 This is a flowchart illustrating an intelligent control method for a tunnel telephone broadcasting system in one embodiment of this application, as shown below. Figure 3 This embodiment of an intelligent management and control method for a tunnel telephone broadcasting system may include steps S310 to S350:

[0102] S310, when receiving emergency calls from multiple tunnel extensions simultaneously, acquires audio and video data collected by monitoring equipment within the group to which the tunnel extension is located, wherein the collected audio and video data is added to an upload queue and uploaded to the cloud for backup;

[0103] When the intelligent management platform connects a call but multiple calls are still waiting to be connected, a call queue is formed. This application visualizes the call queue for management personnel.

[0104] When multiple emergency calls come in simultaneously, it indicates that there may be an abnormal event inside the tunnel. Therefore, this application uses a pre-set camera and microphone to collect audio and video data of the area where the extension is located as the basic data for abnormal event identification.

[0105] S320, Based on a pre-built event classification model, the audio and video data are identified to obtain the event tag of the location of the tunnel unit and the urgency of the event tag. The event tag includes fire, traffic accident, personnel distress call, equipment failure and no failure.

[0106] This application classifies audio and video data by pre-constructing an event classification model to obtain time tags and urgency levels. The network structure of the event classification model in this application includes:

[0107] (1) Video coding branch, used to extract spatiotemporal features of video data based on pre-trained 3DCNN modules, to obtain video features. ;

[0108] Specifically, 3DCNNs (such as I3D (Inflated3DConvNets)) are used to extract spatiotemporal features. I3D is pre-trained on the Kinetics dataset, with an input of 16×224×224×3 and an output feature vector. .

[0109] (2) Audio coding branch, used to extract the spectrogram of audio data based on the pre-trained VGGish module to obtain audio features. ;

[0110] Specifically, VGGish is used to process the Mel spectrogram, treating the 2D spectrogram as an image. VGGish is pre-trained on AudioSet, taking a 96×64 spectrogram as input and outputting a feature vector. Furthermore, if the spectrogram has a long time dimension, it can be compressed first by average pooling, or further processed using a time series model (such as GRU).

[0111] (3) Multimodal fusion layer, used to fuse the video features and the audio features By fusing the features, we can obtain the joint characteristics. ; and based on the fully connected layer, the joint features Dimensionality reduction is performed to obtain dimensionality-reduced features. ;

[0112] Specifically, the video features can be directly... and the audio features By splicing the features together, we can obtain the joint features. Alternatively, attention fusion can be introduced to calculate cross-modal attention weights and then weighted fusion to obtain joint features. .

[0113] Then dimensionality reduction is achieved through a fully connected layer (FC), for example... The dimension is set to 512, where, This represents the weight matrix of the dimension-reduced fully connected layer. This represents the bias vector of the dimension-reduced fully connected layer.

[0114] (4) Multi-task output head, used for dimensionality reduction features Predict event labels and based on the dimensionality reduction features Output the urgency level.

[0115] The multi-task output header includes a classification header and a regression header:

[0116] Category Header: The dimension is equal to the number of event categories C=5 (fire, traffic accident, people seeking help, equipment failure and no failure). This represents the weight matrix of the fully connected layer in the classification head. This represents the bias vector of the fully connected layer in the classification head;

[0117] Return to the head: ,in, This represents the Sigmoid function, which outputs an urgency level of [0,1]. This represents the weight matrix of the fully connected layer in the regression head. This represents the bias vector of the fully connected layer in the regression head.

[0118] Based on the above network structure, the process of training the event classification model includes:

[0119] S1, Obtain audio and video data samples and labels for the audio and video data, wherein the labels include event labels and urgency labels;

[0120] Specifically, the audio and video data samples must ensure strict temporal alignment between the audio and video (usually based on the video frame rate, extracting segments of the same length, such as 2-5 seconds). The video needs to be decoded into a continuous frame sequence and normalized, and the audio needs to be converted into a Mel spectrogram or other acoustic features to ensure that the input format matches the model branches.

[0121] S2, the audio and video data samples are split into audio data samples and video data samples, and the audio data samples and video data samples are respectively input into the audio encoding branch and the video encoding branch to obtain the predicted event label and predicted urgency output by the multi-task output head;

[0122] S3, calculate the classification loss between the predicted event label and the labeled event label based on the pre-built classification loss function, and calculate the regression loss between the predicted urgency and the labeled urgency based on the pre-built regression loss function;

[0123] The model employs a joint loss function to balance classification and regression tasks, specifically including:

[0124]

[0125]

[0126]

[0127] In the formula, Indicates the total loss. Represents the classification loss weights. Represents classification loss. Indicates the regression loss weights. Indicates regression loss, This refers to the number of sample batches. For sample index, Indexed by event category, Indicates the first Event category labels for each sample Indicates the first The predicted probability of a sample. Indicates the first The urgency label for each sample Indicates the first The urgency of prediction for each sample.

[0128] S4, Based on the backpropagation of the classification loss and the regression loss, the internal parameters of the network structure of the event classification model are adjusted;

[0129] S5. Repeat steps S2-S4 until training is complete and the event classification model is obtained.

[0130] During training, the optimizer (e.g., Adam), initial learning rate (e.g., 10⁻⁴ / 10⁻⁴), learning rate decay strategy, batch size (limited by GPU memory, e.g., 16~32), number of training epochs (e.g., 50~100), and early stopping condition (stopping when the validation loss no longer decreases) must be specified. Event labels should be discrete categories (e.g., fire, traffic accident, etc.), and urgency levels should be normalized to continuous values ​​[0,1]. These can be assigned by experts based on the severity of the event or obtained statistically from historical data. Independent validation and test sets need to be created to monitor overfitting and evaluate the model's generalization ability. The model with the best performance on the validation set can be saved during training.

[0131] After training is complete, the current audio and video data is input into the model to obtain the audio and video data corresponding to the current call zone, which may include event tags and urgency levels. Event tags and urgency levels are used for display to management personnel, and the urgency level is also used in subsequent emergency call priority calculations.

[0132] S330, the urgency of the event tags of each tunnel extension is weighted based on the segment weight of the segment where the multiple tunnel extensions are located, to obtain the comprehensive priority of multiple emergency calls;

[0133] In this application, the segment weights are based on historical and real-time data to describe the probability of risk events occurring in different segments. The core purpose of setting segment weights for multiple segments is to enable the system to dynamically adjust the processing priority of emergency calls based on the real-time risk status of different locations within the tunnel, thereby prioritizing the allocation of limited emergency response resources to the most dangerous and urgent areas. Specifically, the calculation process for the segment weights is as follows:

[0134] (1) Obtain static and dynamic parameters of multiple sections, wherein the static parameters include slope, curve radius and historical accident rate, and the dynamic parameters include average vehicle speed, traffic flow, time occupancy, CO concentration, smoke concentration and rainfall intensity, and the dynamic parameters are statistical average values ​​within the target time period before the current time point;

[0135] Specifically, static parameters are based on the inherent properties of the tunnel section (static parameters include gradient, curve radius, historical accident rate, etc.); dynamic parameters are selected from multiple indicators that best reflect the current risk, including average vehicle speed, traffic volume, time occupancy, CO concentration, smoke concentration and rainfall intensity (only for the tunnel entrance section, 0 for no rainfall).

[0136] The collection of the above data relies on sensors pre-deployed inside the tunnel or meteorological data collected from the cloud, specifically including:

[0137] Environmental sensors: CO / VI sensors, smoke detectors, temperature sensors, etc. are deployed at regular intervals (e.g., 100 meters) inside the tunnel to collect data in real time.

[0138] Traffic detectors: acquire instantaneous traffic flow, average speed, and time occupancy of each section through radar or geomagnetic coils.

[0139] Meteorological data: The weather station at the tunnel entrance provides information such as wind speed, wind direction, humidity, and rainfall, and can also connect to third-party weather APIs.

[0140] (2) Normalize the static parameters and the dynamic parameters to obtain normalized static parameters and normalized dynamic parameters, wherein the values ​​of the normalized static parameters and the normalized dynamic parameters are positively correlated with the risk;

[0141] Specifically, for each static or dynamic parameter index By using Min-Max normalization, its values ​​are mapped to the [0,1] interval to obtain the normalized parameter index. If the indicator is negative (e.g., lower speed means higher risk), then take... This is to ensure that the value is positively correlated with the risk.

[0142] (3) Weighting multiple normalized static parameters based on pre-constructed static parameter weights to obtain static weights; and weighting multiple normalized dynamic parameters based on pre-constructed dynamic parameter weights to obtain dynamic weights;

[0143] For static parameter weights, which are mainly indicators such as slope, curve radius, and historical accident rate that change little or not at all, they can be determined by expert scoring or by the analytic hierarchy process.

[0144] For the weights of dynamic parameters, this application uses the entropy weight method to determine them, specifically including:

[0145] (3-1) Assume the tunnel has m sections, and select n dynamic indicators (such as vehicle speed, traffic flow, CO concentration, etc.). Collect the average or typical values ​​of each indicator for each section within a historical period (such as one month) to form the original data matrix:

[0146]

[0147] In the formula, Indicates the first The section in the first The value of each dynamic parameter;

[0148] Then, normalization is performed. Since the dimensions and orders of magnitude of the indicators are different, normalization is necessary. For positive indicators (the larger the value, the higher the risk), the following method is used:

[0149]

[0150] In the formula, Indicates the first The minimum value of each dynamic parameter. Indicates the first The maximum value of each dynamic parameter. Indicates the first One dynamic parameter;

[0151] For negative indicators (the smaller the value, the higher the risk), then:

[0152]

[0153] After normalization, Furthermore, all indicators follow the principle that "the higher the value, the higher the risk," to facilitate subsequent entropy weight analysis.

[0154] (3-2) Calculate the first The section in the first The proportion of each dynamic parameter , wherein the specific gravity The mathematical expression is:

[0155]

[0156] In the formula, Indicates the first A normalized dynamic parameter, Indicates the number of segments;

[0157] (3-3) Based on the aforementioned specific gravity Calculate the first Information entropy of each normalized dynamic parameter And calculate the information entropy. Calculate the first The difference coefficient of each normalized dynamic parameter , where the first Information entropy of each normalized dynamic parameter and coefficient of difference The mathematical expressions are as follows:

[0158]

[0159]

[0160] (3-4) Difference coefficients for multiple dynamic parameters Normalization is performed to obtain the dynamic parameter weights of multiple dynamic parameters. .

[0161] (3-5) The static weights are obtained by weighting the multiple static parameters of each segment based on multiple static parameter weights, and by weighting the multiple dynamic parameters of each segment based on multiple dynamic parameter weights. and dynamic weights .

[0162] The principle behind the dynamic parameter weighting analysis is as follows: the greater the data variation of a certain indicator (i.e., the more significant the differences between different segments on that indicator), the greater the amount of information contained in that indicator, and therefore it should be given a higher weight in the comprehensive evaluation; conversely, if a certain indicator is similar across all segments (small variation), then that indicator does not contribute much to distinguishing risks, and its weight is small. Through the above analysis process, the dynamic parameters of different segments can be evaluated based on real data, making the assessment more scientific and reasonable.

[0163] (4) The static weights and the dynamic weights are fused to obtain the segment weights of the sections where multiple tunnel substations are located. .

[0164] Finally, the static weights of each segment are also combined using a weighted fusion method. and dynamic weights The fusion is represented as:

[0165]

[0166] The weighting here can be set according to needs. If more emphasis is placed on the risks brought about by the dynamic parameters of each section within the tunnel, it can be reduced. The value of is increased if the value of is decreased, and vice versa. The value of .

[0167] S340, sort multiple emergency calls based on the comprehensive priority, and perform speech recognition and event entity extraction on the connected emergency calls to obtain the event entity combination of the current emergency call;

[0168] After sorting, the incoming call queue is displayed as described above, with a recommendation to prioritize high-priority alarm calls, as these may indicate more serious incidents, more complex on-site environments, or higher risks of accidents. Therefore, this prioritization system effectively helps managers streamline the call connection sequence and resolve serious, high-risk incidents as early as possible.

[0169] After a call is connected, the system automatically records the audio and performs voice recognition. Based on the voice recognition content, it recommends broadcasts to help administrators manage the broadcast content.

[0170] In this application, broadcast templates are pre-recorded and stored in a broadcast template library. Through historical data analysis, a probabilistic mapping relationship between event entity combinations and broadcast templates is constructed. During a call, by analyzing the event entity combinations that appear, the most likely directional / FM broadcast template for management personnel is output. This avoids the need for management personnel to switch to the broadcast system to search for and play the desired content. In the event of an abnormal event, this significantly reduces the broadcast preparation time for management personnel.

[0171] In this application, the prediction of the broadcast template depends on speech-to-text conversion and key entity extraction, specifically including:

[0172] S341, when an emergency call is connected, collect real-time speech data, recognize the real-time speech data to obtain a recognized text;

[0173] After a certain amount of speech data is collected, use automatic speech recognition (ASR) technology to convert the real-time speech signal into a text string and convert the unstructured audio into structured text data. Then process it in a loop until the call ends, avoiding the need for a large amount of time for conversion and subsequent matching after the call ends.

[0174] S342, perform word segmentation on the recognized text to obtain multiple words;

[0175] Use a Chinese word segmentation tool (such as jieba) to split the continuous text into an independent word sequence and identify the word boundaries based on the dictionary and statistical model. For example: the text "There is a fire in the tunnel. Hurry and save people" is segmented into ["tunnel", "in", "fire", "了", "hurry", "save people"].

[0176] S343, convert the multiple words into word vectors, and calculate the cosine similarity between the word vectors and any one event entity in the pre-constructed event entity library;

[0177] Convert each word segmentation result into a vector representation through a pre-trained word vector model (such as Word2Vec), and calculate the cosine similarity with the pre-constructed event entity library (each entity corresponds to a standard word vector) to measure the semantic proximity.

[0178] S344, when the cosine similarity between the word vector and any one event entity is greater than or equal to the preset similarity threshold, regard the word vector as an event entity, and construct an event entity combination based on all the event entities extracted during the conversation process, where each event entity combination includes two event entities.

[0179] Record the event entities corresponding to the words whose similarity exceeds the preset threshold, and form an event entity combination (set) after de-duplication based on all the event entities extracted during the entire conversation process. Thus, extract the key information in the call to form a structured emergency event description, providing accurate input for subsequent broadcast template recommendation.

[0180] For example: extract two entities "fire" and "personnel求救" from the call and construct an event entity combination {fire, personnel求救}.

[0181] S350, based on the pre-built broadcast template prediction probability model and the event entity combination, predicts the probability of multiple broadcast templates required by the administrator, and sends the one or more broadcast modules with the highest probability to the administrator.

[0182] The broadcast template prediction probability model in this application maps the probability mapping relationship between various combinations of event entities and broadcast templates. The specific construction process and principle include:

[0183] (1) Obtain multiple historical emergency call event records, wherein the historical emergency call event records include event entity combination samples extracted from voice samples. And the broadcast template ultimately selected by the management;

[0184] Historical data accumulated during system operation is collected. Each record contains a combination of event entities extracted from voice calls (e.g., {fire, trapped}) and the broadcast template ultimately selected by the management personnel at that time (e.g., "fire evacuation template"). This data forms the basis for model learning. By utilizing real historical operational data, the model can learn the decision-making patterns of management personnel in different event scenarios, providing a statistical basis for subsequent probability predictions.

[0185] (2) Statistically analyze the combination of each event entity from multiple historical emergency call event records. The frequency of occurrence was used to determine the M most frequent event entity combinations, which were then selected as key event entity combinations. And merge other event entity combination samples as low-frequency entity combinations. ;

[0186] Count the number of times each different event entity combination appears in the historical records (frequency), and sort them from high to low frequency. Define the top M combinations with the most occurrences as "critical event entity combinations" (high-frequency combinations), and merge all other combinations with fewer occurrences into a unified "low-frequency entity combination" category.

[0187] Low-frequency combinations occur very infrequently, and modeling them individually can lead to unreliable probability estimates (e.g., combinations that occur only once might overfit the chosen option). Merging these combinations and utilizing collective statistics yields a stable probability distribution. This avoids storing probabilities for thousands of rare combinations separately, significantly reducing computational and storage overhead while ensuring accurate modeling of mainstream scenarios. For unseen combinations that may appear in the future, the general probability of low-frequency combinations can be used for reasonable recommendations, avoiding situations that cannot be handled.

[0188] For example: Suppose there are 200 different combinations of event entities in the historical data. Among them, "{fire}" appears 300 times, "{traffic accident}" appears 250 times, "{fire, trapped}" appears 150 times, and so on. The top 50 combinations cover 95% of the total occurrences, and these 50 are considered the key combinations; the remaining 150 combinations (such as {equipment failure, smoke} which only appears twice) are merged into "low-frequency entity combinations".

[0189] (3) Construct a combination of key event entities conditional probability and low-frequency physical combinations conditional probability , wherein the conditional probability and the conditional probability The mathematical expression is:

[0190]

[0191]

[0192] In the formula, This indicates that when event entities are combined as Time broadcast template The probability of being selected Indicates a broadcast template. Represents a combination of key event entities Total number of occurrences For smoothing coefficients, Represents a template set. Indicates the number of broadcast templates. This indicates that when event entities are combined as Time broadcast template The number of times it was selected; This indicates that when event entities are combined as Time broadcast template The probability of being selected Represents a combination of low-frequency event entities Total number of occurrences This indicates that when event entities are combined as Time broadcast template The number of times it was selected;

[0193] In the two calculation formulas above, the probability of each broadcast template being selected under each key combination and low-frequency combination is calculated respectively. The formulas use Laplace smoothing (with μ smoothing) to avoid the probability of some templates having never appeared under a specific combination being zero. For key combinations, the numerator is the number of times template t is selected under that combination plus a smoothing coefficient; the denominator is the total number of selections for that combination plus a smoothing coefficient multiplied by the total number of templates. For low-frequency combinations, the numerator is the total number of times template t is selected under all low-frequency combinations plus a smoothing coefficient; the denominator is the total number of selections for all low-frequency combinations plus a smoothing coefficient multiplied by the total number of templates.

[0194] The probabilities of key combinations are based on their own data, accurately reflecting template preferences in a specific scenario. The probabilities of low-frequency combinations are based on collective data, avoiding the random influence of rare individual combinations and providing a general recommendation benchmark. In addition, Laplace smoothing is introduced to ensure that all templates have non-zero probabilities, enhancing the model's robustness.

[0195] For example, the key combination "{fire}" appears 300 times, of which template t1 is selected 200 times, t2 is selected 100 times, and other templates are selected 0 times. After smoothing, P(t1|{fire}) = (200+1) / (300+1*T), where T is the total number of templates.

[0196] The total number of low-frequency combinations is 50, with template t1 selected 10 times, t2 selected 30 times, and t3 selected 10 times. After smoothing, P(t1|low frequency) = (10+1) / (50+1*T).

[0197] (4) Construct a broadcast template prediction probability model based on the conditional probability of multiple event entity combinations.

[0198] All conditional probabilities calculated in step (3) are stored as lookup tables or model parameters to form a complete broadcast template prediction probability model. When a new event occurs, the corresponding probability is directly looked up based on the extracted event entity combination if it belongs to a key combination; otherwise, the probability of the low-frequency combination is used.

[0199] The probability calculation process based on the above probability model includes:

[0200] S351, Substitute the combination of the multiple event entities into the pre-built broadcast template prediction probability model to obtain the conditional probability of the combination of the multiple event entities;

[0201] During a call, multiple combinations of event entities may be extracted at different times or in different statements (e.g., mentioning "fire" first, then "trapped"). These combinations are then input into a pre-built broadcast template prediction probability model, which, based on historical statistics, provides the conditional probability of each broadcast template for each combination. .

[0202] S352, extract the probability of each broadcast module from the conditional probability of multiple event entity combinations, and sum the probabilities of each broadcast module to obtain the total probability of each broadcast module.

[0203] The conditional probabilities of the same broadcast template under different combinations are summed to obtain the comprehensive score of the template, which is expressed as:

[0204]

[0205] The contributions of different information fragments to template selection are cumulative; a higher overall score indicates that the template better matches the overall call context. Multi-source information can also be integrated to make the recommendation results more comprehensive and robust, reducing the bias or randomness that may arise from a single combination and improving the accuracy of the recommendations. Finally, the broadcast template with the highest score is recommended to the administrator, or the top N broadcast templates with the highest scores are output in order for the administrator to choose from.

[0206] In addition, it is necessary to obtain the broadcast status of multiple sections within the tunnel in real time, including directional broadcast, FM broadcast, or no broadcast, and to feed back the broadcast status and real-time playback content to the management personnel. This is to avoid errors or omissions in the broadcast content of different sections during directional broadcasting.

[0207] This invention provides an intelligent management and control method for a tunnel telephone broadcasting system, addressing the shortcomings of existing tunnel telephone broadcasting systems that operate independently and lack unified management. By collecting on-site audio and video data, using an event classification model to identify event tags and urgency levels, and combining this with dynamic weighting of sections to calculate comprehensive priority, multiple emergency calls are intelligently sorted to ensure high-risk events are handled first and avoid omissions. Audio and video data is automatically uploaded to the cloud, allowing managers to remotely monitor ambient sounds at any telephone terminal, detecting anomalies (such as arguments or equipment noises) in advance, thus improving early warning capabilities. Voice recognition is performed on connected calls to extract event entity combinations, and a broadcast template prediction model automatically recommends the most suitable broadcast template, reducing the cumbersome process of switching between multiple systems for managers and significantly improving emergency response efficiency. This method achieves collaboration among multiple systems such as emergency telephones, broadcasts, and monitoring, enhancing the intelligence level and speed of tunnel emergency management.

[0208] like Figure 4 As shown, this application also provides an intelligent management and control system for a tunnel telephone broadcasting system, comprising:

[0209] The acquisition module is used to acquire audio and video data collected by the monitoring equipment in the group where the tunnel unit is located when receiving emergency calls from multiple tunnel units at the same time. The acquired audio and video data is added to the upload queue and uploaded to the cloud for backup.

[0210] The identification module is used to identify the audio and video data based on a pre-built event classification model to obtain the event tag of the location of the tunnel unit and the urgency of the event tag. The event tag includes fire, traffic accident, personnel distress call, equipment failure and no failure.

[0211] The priority calculation module is used to weight the urgency of the event tags of each tunnel extension based on the segment weight of the segment where the multiple tunnel extensions are located, so as to obtain the comprehensive priority of multiple emergency calls.

[0212] The entity extraction module is used to sort multiple emergency calls based on the comprehensive priority, and to perform speech recognition and event entity extraction on the connected emergency calls to obtain the event entity combination of the current emergency call.

[0213] The broadcast mapping module is used to predict the probability of multiple broadcast templates required by the administrator based on a pre-built broadcast template prediction probability model and the combination of event entities, and send the one or more broadcast templates with the highest probability to the administrator.

[0214] This invention discloses an intelligent management and control system for a tunnel telephone broadcasting system, addressing the shortcomings of existing tunnel telephone broadcasting systems that operate independently and lack unified management. By collecting on-site audio and video data, using an event classification model to identify event tags and urgency levels, and combining this with dynamic weighting of sections to calculate comprehensive priority, multiple emergency calls are intelligently sorted to ensure high-risk events are handled first and avoid omissions. Audio and video data is automatically uploaded to the cloud, allowing managers to remotely monitor ambient sounds at any telephone terminal, detecting anomalies (such as arguments or equipment noises) in advance, thus improving early warning capabilities. Voice recognition is performed on connected calls to extract event entity combinations, and a broadcast template prediction model automatically recommends the most suitable broadcast template, reducing the cumbersome process of switching between multiple systems for managers and significantly improving emergency response efficiency. This system achieves collaboration among emergency telephone, broadcast, and monitoring systems, enhancing the intelligence level and speed of tunnel emergency management.

[0215] Figure 5 A schematic diagram of a computer system suitable for implementing the embodiments of this application is shown. It should be noted that... Figure 5 The computer system of the electronic device shown is merely an example and should not impose any limitation on the functionality and scope of use of the embodiments of this application.

[0216] like Figure 5As shown, the computer system includes a Central Processing Unit (CPU) 501, which can perform various appropriate actions and processes, such as executing the methods described in the above embodiments, based on a program stored in Read-Only Memory (ROM) 502 or a program loaded from storage portion 508 into Random Access Memory (RAM) 503. The RAM 503 also stores various programs and data required for system operation. The CPU 501, ROM 502, and RAM 503 are interconnected via a bus 504. An Input / Output (I / O) interface 505 is also connected to the bus 504.

[0217] The following components are connected to I / O interface 505: an input section 506 including a keyboard, mouse, etc.; an output section 507 including a cathode ray tube (CRT), liquid crystal display (LCD), etc., and speakers, etc.; a storage section 508 including a hard disk, etc.; and a communication section 509 including a network interface card such as a LAN (Local Area Network) card, modem, etc. The communication section 509 performs communication processing via a network such as the Internet. A drive 510 is also connected to I / O interface 505 as needed. Removable media 511, such as a disk, optical disk, magneto-optical disk, semiconductor memory, etc., are installed on drive 510 as needed so that computer programs read from them can be installed into storage section 508 as needed.

[0218] Specifically, according to embodiments of this application, the processes described above with reference to the flowcharts can be implemented as computer software programs. For example, embodiments of this application include a computer program product comprising a computer program carried on a computer-readable medium, the computer program including a computer program for performing the methods shown in the flowcharts. In such embodiments, the computer program can be downloaded and installed from a network via communication section 509, and / or installed from removable medium 511. When the computer program is executed by central processing unit (CPU) 501, it performs various functions defined in the system of this application.

[0219] It should be noted that the computer-readable medium shown in the embodiments of this application can be a computer-readable signal medium or a computer-readable storage medium, or any combination of the two. A computer-readable storage medium can be, for example, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination thereof. More specific examples of a computer-readable storage medium may include, but are not limited to: an electrical connection having one or more wires, a portable computer disk, a hard disk, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM), flash memory, optical fiber, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination thereof. In this application, a computer-readable signal medium may include a data signal propagated in baseband or as part of a carrier wave, carrying a computer-readable computer program. Such propagated data signals can take various forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination thereof. Computer-readable signal media can also be any computer-readable medium other than computer-readable storage media, which can send, propagate, or transmit a program for use by or in connection with an instruction execution system, apparatus, or device. The computer program contained on the computer-readable medium can be transmitted using any suitable medium, including but not limited to wireless, wired, etc., or any suitable combination thereof.

[0220] The flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to various embodiments of this application. Each block in a flowchart or block diagram may represent a module, segment, or portion of code, which contains one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions indicated in the blocks may occur in a different order than those indicated in the drawings. For example, two consecutively indicated blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in a block diagram or flowchart, and combinations of blocks in a block diagram or flowchart, may be implemented using a dedicated hardware-based system that performs the specified function or operation, or using a combination of dedicated hardware and computer instructions.

[0221] The units described in the embodiments of this application can be implemented in software or hardware, and the described units can also be located in a processor. The names of these units do not necessarily limit the specific unit itself.

[0222] Another aspect of this application provides a computer-readable storage medium storing a computer program thereon, which, when executed by a computer's processor, causes the computer to perform the method as described above. This computer-readable storage medium may be included in the electronic device described in the above embodiments, or it may exist independently and not assembled into the electronic device.

[0223] Another aspect of this application provides a computer program product or computer program including computer instructions stored in a computer-readable storage medium. A processor of a computer device reads the computer instructions from the computer-readable storage medium and executes the computer instructions, causing the computer device to perform the methods provided in the various embodiments described above.

[0224] The above embodiments are merely preferred embodiments provided to fully illustrate this application, and the scope of protection of this application is not limited thereto. Equivalent substitutions or modifications made by those skilled in the art based on this application are all within the scope of protection of this application.

Claims

1. An intelligent control method for a tunnel telephone broadcasting system, characterized in that, Including the following steps: When receiving emergency calls from multiple tunnel extensions simultaneously, the system acquires audio and video data collected by monitoring equipment within the group to which the tunnel extension is located. The collected audio and video data is added to an upload queue and uploaded to the cloud for backup. The audio and video data are identified based on a pre-built event classification model to obtain the event tag of the location of the tunnel unit and the urgency of the event tag. The event tag includes fire, traffic accident, personnel distress call, equipment failure and no failure. The urgency of the event tags of each tunnel extension is weighted based on the segment weight of the segment where the multiple tunnel extensions are located, so as to obtain the comprehensive priority of multiple emergency calls. Based on the comprehensive priority, multiple emergency calls are sorted, and it is recommended to prioritize answering high-priority alarm calls. Then, voice recognition and event entity extraction are performed on the emergency calls that are being connected to obtain the event entity combination of the current emergency call. Based on a pre-built broadcast template prediction probability model and the event entity combination, the probability of multiple broadcast templates required by the administrator is predicted, and the one or more broadcast templates with the highest probability are sent to the administrator.

2. The intelligent control method for a tunnel telephone broadcasting system according to claim 1, characterized in that, The network structure of the event classification model includes: The video coding branch is used to extract spatiotemporal features from video data based on pre-trained 3DCNN modules, thus obtaining video features. ; The audio coding branch is used to extract the spectrogram of the audio data based on the pre-trained VGGish module, thereby obtaining audio features. ; A multimodal fusion layer is used to combine the video features. and the audio features By fusing the features, we obtain the joint characteristics. ; and based on the fully connected layer, the joint features Dimensionality reduction is performed to obtain dimensionality-reduced features. ; Multi-task output header, used for dimensionality reduction features Predict event labels and based on the dimensionality reduction features Output the urgency level.

3. The intelligent control method for a tunnel telephone broadcasting system according to claim 2, characterized in that, The process of constructing the event classification model includes: S1, Obtain audio and video data samples and labels for the audio and video data, wherein the labels include event labels and urgency labels; S2, the audio and video data samples are split into audio data samples and video data samples, and the audio data samples and video data samples are respectively input into the audio encoding branch and the video encoding branch to obtain the predicted event label and predicted urgency output by the multi-task output head; S3, calculate the classification loss between the predicted event label and the labeled event label based on the pre-built classification loss function, and calculate the regression loss between the predicted urgency and the labeled urgency based on the pre-built regression loss function; S4, Based on the backpropagation of the classification loss and the regression loss, the internal parameters of the network structure of the event classification model are adjusted; S5. Repeat steps S2-S4 until training is complete and the event classification model is obtained.

4. The intelligent control method for a tunnel telephone broadcasting system according to claim 1, characterized in that, The methods for determining the segment weights of multiple tunnel extension units include: The static and dynamic parameters of multiple sections are obtained. The static parameters include slope, curve radius and historical accident rate. The dynamic parameters include average vehicle speed, traffic flow, time occupancy, CO concentration, smoke concentration and rainfall intensity. The dynamic parameters are the statistical average values ​​of the target duration before the current time point. The static parameters and the dynamic parameters are normalized to obtain normalized static parameters and normalized dynamic parameters, wherein the values ​​of the normalized static parameters and the normalized dynamic parameters are positively correlated with the risk. Static weights are obtained by weighting multiple normalized static parameters based on pre-built static parameter weights; and dynamic weights are obtained by weighting multiple normalized dynamic parameters based on pre-built dynamic parameter weights. The static weights and dynamic weights are fused to obtain the segment weights for the sections where multiple tunnel extension units are located. .

5. The intelligent control method for a tunnel telephone broadcasting system according to claim 4, characterized in that, The static parameter weights are determined based on expert evaluation, and the method for determining the dynamic parameter weights includes: Calculate the first The section in the first The proportion of each dynamic parameter , wherein the specific gravity The mathematical expression is: In the formula, Indicates the first A normalized dynamic parameter, Indicates the number of segments; Based on the specific gravity Calculate the first Information entropy of each normalized dynamic parameter And calculate the information entropy. Calculate the first The difference coefficient of each normalized dynamic parameter , where the first Information entropy of each normalized dynamic parameter and coefficient of difference The mathematical expressions are as follows: Difference coefficients for multiple dynamic parameters Normalization is performed to obtain the dynamic parameter weights of multiple dynamic parameters. .

6. The intelligent control method for a tunnel telephone broadcasting system according to claim 1, characterized in that, The system performs speech recognition and event entity extraction on the connected emergency call to obtain the event entity combination of the current emergency call, including: When an emergency call is connected, real-time voice data is collected, and the real-time voice data is recognized to obtain the recognized text. The identified text is segmented into multiple words; The multiple words are converted into word vectors, and the cosine similarity between the word vectors and any event entity in the pre-built event entity library is calculated. When the cosine similarity between the word vector and any event entity is greater than or equal to a preset similarity threshold, the word vector is used as an event entity, and an event entity combination is constructed based on all event entities extracted during the dialogue process, wherein each event entity combination includes two event entities.

7. The intelligent control method for a tunnel telephone broadcasting system according to claim 6, characterized in that, The method for constructing the broadcast template prediction probability model includes: Acquire multiple historical emergency call event records, wherein the historical emergency call event records include event entity combination samples extracted from voice samples. And the broadcast template ultimately selected by the management; Statistical analysis of each event entity combination sample from multiple historical emergency call event records. The frequency of occurrence was used to determine the M most frequent event entity combinations, which were then selected as key event entity combinations. And merge other event entity combination samples as low-frequency entity combinations. ; Constructing a portfolio of key event entities conditional probability and low-frequency physical combinations conditional probability , wherein the conditional probability and the conditional probability The mathematical expression is: In the formula, This indicates that when event entities are combined as Time broadcast template The probability of being selected Indicates a broadcast template. Represents a combination of key event entities Total number of occurrences For smoothing coefficients, Represents a template set. Indicates the number of broadcast templates. This indicates that when event entities are combined as Time broadcast template The number of times it was selected; This indicates that when event entities are combined as Time broadcast template The probability of being selected Represents a combination of low-frequency event entities Total number of occurrences This indicates that when event entities are combined as Time broadcast template The number of times it was selected; A broadcast template prediction probability model is constructed based on the conditional probability of multiple event entity combinations.

8. The intelligent control method for a tunnel telephone broadcasting system according to claim 7, characterized in that, Based on a pre-built broadcast template prediction probability model and the event entity combination, the probability of predicting various broadcast templates required by the administrator includes: Substitute the combination of the multiple event entities into the pre-built broadcast template prediction probability model to obtain the conditional probability of the combination of multiple event entities; Extract the probability of each broadcast template from the conditional probability of multiple event entity combinations, and sum the probabilities of each broadcast template to obtain the total probability of each broadcast template.

9. The intelligent control method for a tunnel telephone broadcasting system according to claim 1, characterized in that, Also includes: The system acquires the broadcast status of multiple sections within the tunnel and feeds back the broadcast status and real-time playback content to management personnel.

10. An intelligent control system for a tunnel telephone broadcasting system, characterized in that, include: The acquisition module is used to acquire audio and video data collected by the monitoring equipment in the group where the tunnel unit is located when receiving emergency calls from multiple tunnel units at the same time. The acquired audio and video data is added to the upload queue and uploaded to the cloud for backup. The identification module is used to identify the audio and video data based on a pre-built event classification model to obtain the event tag of the location of the tunnel unit and the urgency of the event tag. The event tag includes fire, traffic accident, personnel distress call, equipment failure and no failure. The priority calculation module is used to weight the urgency of the event tags of each tunnel extension based on the segment weight of the segment where the multiple tunnel extensions are located, so as to obtain the comprehensive priority of multiple emergency calls. The entity extraction module is used to sort multiple emergency calls based on the comprehensive priority, recommend to connect high-priority alarm calls first, and perform voice recognition and event entity extraction on the connected emergency calls to obtain the event entity combination of the current emergency call. The broadcast mapping module is used to predict the probability of multiple broadcast templates required by the administrator based on a pre-built broadcast template prediction probability model and the combination of event entities, and send the one or more broadcast templates with the highest probability to the administrator.

Images