A geological disaster emergency early warning method based on multi-source sensor collaboration

Abstract

This invention discloses a geological disaster emergency early warning method based on multi-source sensor collaboration, relating to the field of geological disaster early warning. The method includes: deploying sensors on the slope to be monitored and recording their spatial coordinates; determining the online status of the sensors based on whether they transmit data; performing connectivity analysis on disconnected sensors based on sensor adjacency relationships to form a set of disconnected zones; tracking the changing characteristics of the disconnected zones over time to identify those with a downward progression characteristic as key disconnected zones; acquiring online sensor data transmitted around the boundary area of ​​the key disconnected zones, identifying whether data indicating geological disasters exists within these data, and marking the key disconnected zones as geological disaster early warning areas if such data is present. This invention solves the problems of difficulty in effectively utilizing early-stage disconnection behavior, difficulty in identifying spatial progression trends, and the inability to coordinate online monitoring anomalies with disconnected structures for judgment. It can capture key signs of geological disasters earlier and improve the reliability of early warning.

Description

Technical Field

[0001] This invention relates to the field of geological disaster early warning, and more specifically, to a geological disaster emergency early warning method based on multi-source sensor collaboration. Background Technology

[0002] Geological disasters are characterized by their sudden onset, insidious evolution, and significant influence from multiple coupled factors. Especially under external influences such as rainfall, surface disturbance, or changes in groundwater, the internal structure of slopes can experience loosening, tensile cracking, and shear failure, with the damage often expanding rapidly within a short period. With the development of monitoring methods, deploying dense arrays of various sensors on slopes has become an important disaster monitoring approach. By collecting data on deformation, vibration, pore water pressure, and rainfall in real time, the condition of the slope can be reflected to some extent. However, when a disaster is in its gestation or acceleration phase, sensors in certain areas may lose connection due to burial, excessive displacement, or power or communication disruptions, forming localized, continuous sensing voids. This change is often a key indicator of the disaster's progression.

[0003] Existing monitoring systems typically treat sensor disconnection as equipment malfunction, rarely combining the spatial relationships and temporal characteristics of the disconnection with the disaster evolution trend for analysis. This can lead to misjudgments or omissions at critical moments. Furthermore, the morphology of the disconnected area may exhibit continuous progression over time, but traditional early warning methods fail to effectively utilize the spatial clustering behavior and temporal evolution trends of the disconnected area, nor do they coordinate the analysis of abnormal data transmitted from online sensors with the evolution of the disconnected area. Simultaneously, there is a lack of analytical methods to predict future risk areas based on historical spatiotemporal changes in disaster patterns, making it difficult to meet the lead time requirements in emergency response.

[0004] In this context, it would be beneficial if a method existed that could collaboratively analyze multi-source sensor data, sensor connectivity, spatial evolution characteristics of disconnected areas, and real-time observation information from online sensors, in order to capture risk signs at critical stages of a disaster and improve the timeliness and reliability of early warnings. Summary of the Invention

[0005] The technical problem to be solved by the present invention is to provide a geological disaster emergency early warning method based on multi-source sensor collaboration, which uses data from online sensors and disconnected sensors to perform dual-source collaborative judgment and make full use of sensor data.

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

[0007] A geological disaster emergency early warning method based on multi-source sensor collaboration includes the following steps:

[0008] Sensors are deployed on the slope to be monitored, and the spatial coordinates and number of each sensor are recorded in advance.

[0009] The sensor is marked as online or offline based on whether it transmits data back;

[0010] Based on the sensor adjacency relationship, the connectivity analysis of the sensors in the sensor disconnection state is performed. When more than a preset threshold number of adjacent sensors are simultaneously disconnected, the corresponding disconnected sensors are connected to form a disconnection zone, and all disconnection zones are combined into a disconnection zone set.

[0011] By tracking the characteristics of the loss zones over time, the loss zones with the characteristic of advancing from top to bottom are identified as key loss zones.

[0012] Data transmitted from online sensors surrounding the boundary area of ​​the key disconnection zone is acquired, and the presence of data indicating geological hazards is identified. If such data is found, the key disconnection zone is marked as a geological hazard early warning area.

[0013] Preferably, the method further includes: making the following judgment for each missing contact zone in the set of missing contact zones, excluding the key missing contact zones:

[0014] The characteristics of the lost contact zone changing over time are tracked to form the time-series characteristic data of the lost contact zone's advancement;

[0015] Data transmitted back from online sensors surrounding the boundary region of the lost contact zone is acquired and superimposed with the propagation time-series characteristic data of the lost contact zone to obtain comprehensive observation data;

[0016] The geological hazard identification model is trained based on historical data from the comprehensive observation data.

[0017] The current data of the comprehensive observation data is input into the geological disaster identification model, and the geological disaster identification model outputs a determination of whether the disconnected zone belongs to the geological disaster early warning area.

[0018] Preferably, the geological hazard identification model also outputs a classification of geological hazard categories.

[0019] Preferably, the method further includes predicting the extension location of the geological disaster warning area within a future time window based on the advancing characteristics of the disconnected zone determined to be a geological disaster warning area.

[0020] Preferably, the identification of missing contact zones within the set of missing contact zones that exhibit a top-down progression characteristic includes:

[0021] Construct the boundary polygon region of the disconnection zone over multiple consecutive time steps;

[0022] For each time step, calculate the geometric centroid coordinates of the boundary polygon region, and arrange the geometric centroid coordinates corresponding to each time step in chronological order to form a centroid trajectory sequence in which the centroid position of the lost zone changes over time.

[0023] When the trajectory sequence of the center of gravity presents a shape that is continuously downward in the vertical direction, the corresponding loss zone is identified as a loss zone with the characteristic of advancing from top to bottom, and is regarded as the key loss zone.

[0024] Preferably, the extended locations of the geological disaster early warning area within the predicted future time window include:

[0025] Within multiple historical time steps, based on the spatial coordinate information of sensors that are out of contact within the out-of-contact zone identified as a geological disaster early warning area, corresponding boundary polygon regions are constructed, and the boundary polygon sequence is obtained in chronological order.

[0026] Based on historical training data consisting of boundary polygon sequences, an extended prediction model for geological disaster early warning areas is trained.

[0027] During the online early warning phase, the boundary polygon sequence of the current and recent several time steps of the boundary polygon region is input into the extended prediction model. The extended prediction model outputs the predicted boundary polygon region within the future time window as the extended location of the geological disaster early warning area within the future time window.

[0028] Preferably, identifying whether there is data indicating geological hazards in the feedback data from online sensors surrounding the boundary area of ​​the key disconnection zone includes:

[0029] Extract the data transmitted back from online sensors within the target time window when the geological disaster occurred from historical geological disaster samples to construct a historical geological disaster sample library;

[0030] Acquire the data transmitted back from the online sensors surrounding the boundary region of the key disconnection zone within the target time window to form the current observation data;

[0031] The similarity between the current observation data and the sample data in the historical geological disaster sample database is calculated. When the similarity meets the preset similarity threshold condition, it is determined that there is data representing geological disasters in the current observation data.

[0032] Preferably, the sensor includes any one or more of the following: displacement sensor, acceleration sensor, pore water pressure sensor, and rainfall sensor.

[0033] Preferably, the geological hazard identification model is a neural network model.

[0034] Preferably, the extended prediction model is a Transformer-based prediction model.

[0035] The advantage of this invention over existing technologies lies in its construction of a multi-source sensor collaborative emergency early warning method, transforming sensor disconnection behavior from a traditional fault into a usable disaster evolution signal. By constructing disconnection zones based on sensor adjacency relationships and forming a set of disconnection zones, and then identifying disconnection zones with downward progression characteristics, potential destructive trends aligned with the slope's gravity direction can be captured, enabling timely perception of early disaster signs. Furthermore, by identifying the presence of disaster-related anomalies through data transmitted from online sensors, and combining spatial structural anomalies with multi-source monitoring anomalies, the accuracy of early warning decisions is significantly improved.

[0036] In further analysis, this invention tracks the temporal characteristics of the lost-connection zone, transforming the continuous changes in spatial morphology over time into quantifiable temporal data. It then overlays the observation information from online sensors with the temporal characteristics of the advance to form comprehensive observation data. A geological hazard identification model trained on this data can automatically determine whether the lost-connection zone belongs to a geological hazard early warning area, and simultaneously output possible hazard types, making the early warning process more intelligent and precise.

[0037] For lost contact zones identified as geological disaster early warning areas, this invention constructs a boundary polygon sequence and uses an extended prediction model to predict the boundary polygon regions within future time windows, enabling the early warning range to be extrapolated in advance, thereby supporting early deployment and scheduling. Furthermore, this invention is scalable in terms of sensor types, anomaly identification methods, and prediction models, adapting to monitoring needs in different scenarios. Overall, this invention improves the reliability of disaster confirmation, enhances the lead time for early warnings, and maintains effectiveness in complex environments. Attached Figure Description

[0038] Figure 1 This is a flowchart of the method of the present invention;

[0039] Figure 2 This is a schematic diagram of the sensor installation according to the present invention;

[0040] Figure 3 This is a schematic diagram illustrating the downward movement of the disconnected belt in this invention. Detailed Implementation

[0041] The specific embodiments of the present invention will now be described with reference to the accompanying drawings.

[0042] As shown in Figure 1, this invention provides a geological disaster emergency early warning method that utilizes sensor failure itself as a signal feature and combines it with multi-source data collaborative analysis. Traditional monitoring methods often rely on changes in the continuously transmitted data values ​​from sensors to determine disasters. However, during severe geological disasters, sensors in the core area are easily physically damaged and cease functioning, leading to the loss of crucial data. This embodiment shifts the monitoring approach, treating the sensor's disconnection state as a spatial feature of geological deformation. By analyzing the spatiotemporal evolution of the disconnected area and combining it with monitoring data from surviving sensors in the surrounding area, a highly reliable disaster early warning system is achieved.

[0043] As shown in Figure 2, the method in this embodiment first requires basic hardware deployment and initialization. On the slope to be monitored or other areas prone to geological hazards, staff need to deploy a large number of sensor nodes according to the terrain and geological structure. These sensors constitute a high-density sensor network. To achieve comprehensive physical quantity monitoring, the deployed sensors cover various types, specifically including displacement sensors for monitoring surface and deep displacement changes, accelerometers for capturing ground vibration and impact signals, pore water pressure sensors for monitoring changes in pore water pressure within the soil, and rainfall sensors for monitoring precipitation-induced factors. After the sensor deployment is completed, the system pre-records the spatial coordinate information and unique identification number of each sensor. The spatial coordinate information establishes a digital three-dimensional terrain model, making subsequent analysis of sensor positional relationships possible.

[0044] During system operation, the status marking step is executed first. The data acquisition center continuously receives data packets or heartbeat signals from each sensor. Based on whether the sensor successfully transmits data within a specified time window, the system marks the sensor as online or offline. Typically, if a sensor does not respond to any signal for several consecutive communication cycles, or explicitly returns a hardware fault code, it is considered offline. After ruling out common faults such as battery depletion or communication congestion, this offline status highly likely indicates that the sensor itself or its connecting cables have been damaged by external geological forces such as shearing, compression, or burial.

[0045] Subsequently, the system performs connectivity analysis on the sensors in a disconnected state based on their adjacency relationships. While the disconnection of a single sensor may be a random failure, the simultaneous disconnection of a large number of sensors strongly indicates a geological hazard. Therefore, the system searches for the spatial neighbor nodes of the disconnected sensors to determine if they are also disconnected. When more than a preset threshold number of adjacent sensors are disconnected within the same time period, the system connects these spatially connected disconnected sensors to form a disconnection zone. In some embodiments, the threshold number can be selected as 2 to 6. By traversing the entire monitoring area, the system summarizes all identified disconnection zones into a disconnection zone set. This set dynamically reflects all areas on the slope that may be undergoing severe deformation.

[0046] To differentiate between different types of anomalies and accurately pinpoint the sources of geological hazards, the system continuously tracks the characteristics of lost contact zones over time. Geological hazards such as landslides and debris flows often exhibit the physical characteristic of moving from high to low altitudes driven by gravity. Therefore, the system filters out lost contact zones from the set that exhibit downward-moving characteristics and marks them as key lost contact zones.

[0047] To address the specific identification process of missing contact zones exhibiting a downward-progressing characteristic, this invention employs a trajectory analysis method based on the geometric center of gravity. The system acquires the coordinates of all sensors constituting the missing contact zone in real time across multiple consecutive time steps and constructs the boundary polygonal region of the missing contact zone. For each boundary polygonal region generated at each time step, the system calculates its geometric center of gravity coordinates. As time progresses, the geometric center of gravity coordinates corresponding to each time step are arranged in chronological order, thus forming a sequence of center of gravity trajectories showing the change in the center of gravity position of the missing contact zone over time. The system analyzes the vector direction of this trajectory sequence in three-dimensional space. When the center of gravity trajectory sequence exhibits a shape that continuously descends vertically and conforms to a downward slope trend horizontally, it indicates that the area of ​​destruction is expanding downwards under the influence of gravity. The system then identifies the corresponding missing contact zone as a missing contact zone exhibiting a downward-progressing characteristic, i.e., a key missing contact zone. Figure 3 The embodiment shown is a schematic diagram of the centers of gravity G1, G2, and G3 moving vertically downwards under the conditions of T1, T2, and T3. This loss zone can be considered as the key loss zone.

[0048] In this embodiment, to quickly obtain the geometric centroid of the lost-connection boundary polygon, existing mature analytical geometry methods can be used for calculation. For vertices arranged in sequence... The centroid coordinates of a simple polygon can be directly obtained using the following analytical formula:

[0049] ;

[0050] ;

[0051] ;

[0052] Where A is the area of ​​the polygon. Let be the geometric centroid coordinates of the polygon. The successor point of vertex i takes the value i+1, and the successor point of the last vertex returns to the first vertex. n is the number of vertices. This formula only involves addition and multiplication operations, and the computational complexity is linear. It can quickly obtain the centroid coordinates of the lost zone at each time step in continuous time steps.

[0053] Using the geometric centroid as a representation of the advancing trend of the lost contact zone is more effective in reflecting the overall spatial movement trend of the zone compared to directly tracking changes in individual sensors or local boundary points. Single-point methods may be affected by factors such as individual sensor anomalies, abrupt changes in local displacement, or occlusion of individual points, while centroid determination is equivalent to a consistent integration of the spatial distribution of the entire lost contact zone, effectively filtering out local noise and making the determination of the advancing direction more stable. Furthermore, the centroid trajectory naturally possesses continuity and directionality, and its changing trend in three-dimensional space can directly reflect the overall downward expansion behavior of the region. Therefore, compared to methods based on local geometric quantities, local boundary directions, or discrete point sets, the centroid-based determination method is more suitable as a core indicator for identifying the downward advancing characteristics of the lost contact zone.

[0054] After identifying the key loss-of-connection zone, the system performs multi-source collaborative verification to further confirm whether a geological disaster has actually occurred. The system acquires data transmitted from sensors still online surrounding the boundary of the key loss-of-connection zone. Although these boundary sensors are not yet damaged, they are highly likely to have captured precursory or accompanying signals of a disaster, such as severe vibration acceleration, a surge in pore water pressure, or abnormal displacement. The system identifies whether these transmitted data contain data characteristics indicative of a geological disaster. If so, the key loss-of-connection zone is officially marked as a geological disaster warning area, and a corresponding alarm is triggered.

[0055] To identify whether data representing geological hazards exists in the online sensor data transmitted from boundary areas, this embodiment employs a similarity matching scheme based on a historical sample database. The system pre-extracts data segments transmitted from online sensors located in the hazard-edge area within the target time window of a geological hazard event from a large number of historical geological hazard event records, constructing a historical geological hazard sample database. During the real-time monitoring phase, the system acquires the transmitted data from online sensors surrounding the boundary area of ​​the key loss-of-connection zone within the corresponding target time window, forming the current observation data. Subsequently, time series similarity algorithms, such as Dynamic Time Warping (DTW) or cosine similarity calculation, are used to compare the current observation data with the sample data in the historical geological hazard sample database. When the calculated similarity value meets a preset similarity threshold, it indicates that the signal pattern recorded by the current boundary sensor is highly consistent with historically confirmed hazard signal patterns, thus confirming the presence of data representing geological hazards in the current observation data.

[0056] In addition to analyzing key missing-connection zones with obvious characteristics, this embodiment also uses an artificial intelligence model to conduct in-depth analysis of other missing-connection zones besides the key ones, to prevent the omission of atypical geological disaster patterns. For these non-key missing-connection zones, the system also tracks their characteristics over time, forming temporal characteristic data of the missing-connection zone's advancement. In a specific embodiment, the advancement temporal characteristic data can be represented in the most direct way, that is, recording the unique number, spatial coordinates, and time information of the time when each missing sensor constituting the missing-connection zone entered the missing-connection state in multiple consecutive time steps. Arranging the above data of newly missing sensors in different time steps in chronological order can form advancement temporal characteristic data reflecting the outward expansion of the missing-connection zone over time. This data can intuitively describe the evolution process of the missing-connection zone in the time dimension, providing a reliable basis for subsequent identification of the advancing direction of the missing-connection zone and determining whether there is an overall upward advancing trend. At the same time, the feedback data of online sensors surrounding the boundary region of the missing-connection zone is acquired. The system overlays and fuses the propulsion timing characteristic data of the lost contact zone with the data transmitted back from the boundary online sensors to obtain multi-dimensional comprehensive observation data.

[0057] Based on this comprehensive observation data, the system introduces a geological hazard identification model. This model is a fully trained neural network whose architecture is designed to simultaneously handle spatial and temporal features. During the training phase, the model is iteratively trained using labeled historical comprehensive observation data, allowing it to learn the data patterns corresponding to different types of geological activities. In practical applications, the currently collected and fused comprehensive observation data is input into the geological hazard identification model. After processing, the model outputs a binary classification judgment indicating whether the lost-connection zone belongs to a geological hazard warning area. Furthermore, the output layer of this geological hazard identification model not only includes a binary judgment of whether it is a hazard, but also incorporates multi-classification output nodes, capable of outputting specific classification results for the geological hazard category, such as determining whether the hazard is a deep landslide, a shallow collapse, or a debris flow, thereby providing more accurate decision-making basis for emergency rescue.

[0058] After identifying the geological disaster warning area, this embodiment also includes a function to predict the future development trend of the disaster in order to buy valuable evacuation time for downstream threatened areas. Based on the advancing characteristics of the disconnected zone identified as the geological disaster warning area, the system predicts the extension location of the geological disaster warning area within a future time window. The specific prediction is achieved through a specially designed extension prediction model.

[0059] In some embodiments, the extended prediction model employs a deep learning network based on the Transformer architecture. This is because the self-attention mechanism unique to the Transformer architecture can extremely effectively capture long-range dependencies in time-series data, making it highly suitable for handling complex spatiotemporal processes such as the evolution of geological hazards. During the model building and training phases, the system collects data from multiple historical time steps. Based on the spatial coordinates of sensors that are out of service within the out-of-connection zones already identified as geological hazard warning areas, corresponding boundary polygon regions are constructed and arranged chronologically to obtain a sequence of boundary polygons. These historical boundary polygon sequences constitute the training dataset, used to train the extended prediction model, enabling it to learn how to infer future shape changes based on a series of past shape changes.

[0060] In a specific embodiment, the Transformer prediction model comprises an encoder and a decoder. The encoder receives the sequence of boundary polygon vertex coordinates from historical time steps, embeds temporal sequence information into the data through a positional encoding layer, and then extracts the correlation features of shape evolution between different time steps through a multi-head self-attention layer. The decoder uses the feature vector output by the encoder, combined with a query mechanism, to progressively generate the boundary polygon coordinates predicted for future time steps.

[0061] During the online early warning phase, the system inputs a sequence of boundary polygon regions, consisting of the current time and several recent historical time steps, into a trained extended prediction model. The model performs inference calculations and outputs the predicted boundary polygon region within a future time window, such as the next ten minutes or half an hour. This predicted polygon region represents the extended location of the geological disaster early warning area within the future time window. This result can be directly mapped onto a map, visually displaying the potential scope of the disaster, thereby guiding personnel evacuation and the deployment of emergency resources.

[0062] In summary, this embodiment constructs a comprehensive, multi-layered geological disaster monitoring and early warning system by fusing the on / off status characteristics of sensors with real-time transmitted physical quantity data, combined with geometric morphology analysis, historical sample matching, and advanced deep learning models. This system can not only use sensor failure—a phenomenon usually considered an adverse factor—to reverse-engineer the disaster process, but also predict the future trajectory of disasters through models, greatly improving the timeliness and accuracy of geological disaster early warnings.

[0063] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. A geological disaster emergency early warning method based on multi-source sensor collaboration, characterized in that, Includes the following steps: Sensors are deployed on the slope to be monitored, and the spatial coordinates and number of each sensor are recorded in advance. The sensor is marked as online or offline based on whether it transmits data back; Based on the sensor adjacency relationship, the connectivity analysis of the sensors in the sensor disconnection state is performed. When more than a preset threshold number of adjacent sensors are simultaneously disconnected, the corresponding disconnected sensors are connected to form a disconnection zone, and all disconnection zones are combined into a disconnection zone set. The characteristics of the lost contact zones changing over time are tracked to identify those zones exhibiting a downward progression from top to bottom as key lost contact zones. The identification of lost contact zones exhibiting this downward progression includes: Construct the boundary polygon region of the disconnection zone over multiple consecutive time steps; For each time step, calculate the geometric centroid coordinates of the boundary polygon region, and arrange the geometric centroid coordinates corresponding to each time step in chronological order to form a centroid trajectory sequence in which the centroid position of the lost zone changes over time. When the trajectory sequence of the center of gravity presents a shape that is continuously downward in the vertical direction, the corresponding loss zone is identified as a loss zone with the characteristic of advancing from top to bottom, and is regarded as the key loss zone; Data transmitted from online sensors surrounding the boundary area of ​​the key disconnection zone is acquired, and the presence of data indicating geological hazards is identified. If such data is found, the key disconnection zone is marked as a geological hazard early warning area.

2. The method according to claim 1, characterized in that, The method further includes: making the following judgment on each missing contact zone in the set of missing contact zones, excluding the key missing contact zones: The characteristics of the lost contact zone changing over time are tracked to form the time-series characteristic data of the lost contact zone's advancement; Data transmitted back from online sensors surrounding the boundary region of the lost contact zone is acquired and superimposed with the propagation time-series characteristic data of the lost contact zone to obtain comprehensive observation data; The geological hazard identification model is trained based on historical data from the comprehensive observation data. The current data of the comprehensive observation data is input into the geological disaster identification model, and the geological disaster identification model outputs a determination of whether the disconnected zone belongs to the geological disaster early warning area.

3. The method according to claim 2, characterized in that, The geological hazard identification model also outputs a classification of geological hazard categories.

4. The method according to claim 1 or 2, characterized in that, The method also includes predicting the extension location of the geological disaster warning area within a future time window based on the advancing characteristics of the disconnected zone identified as a geological disaster warning area.

5. The method according to claim 4, characterized in that, The predicted extension of the geological disaster early warning area within the future time window includes: Within multiple historical time steps, based on the spatial coordinate information of sensors that are out of contact within the out-of-contact zone identified as a geological disaster early warning area, corresponding boundary polygon regions are constructed, and the boundary polygon sequence is obtained in chronological order. Based on historical training data consisting of boundary polygon sequences, an extended prediction model for geological disaster early warning areas is trained. During the online early warning phase, the boundary polygon sequence of the current and recent several time steps of the boundary polygon region is input into the extended prediction model. The extended prediction model outputs the predicted boundary polygon region within the future time window as the extended location of the geological disaster early warning area within the future time window.

6. The method according to claim 1, characterized in that, Identifying whether data indicating geological hazards exists in the data transmitted back from online sensors surrounding the boundary area of ​​the key disconnection zone includes: Extract the data transmitted back from online sensors within the target time window when the geological disaster occurred from historical geological disaster samples to construct a historical geological disaster sample library; Acquire the data transmitted back from the online sensors surrounding the boundary region of the key disconnection zone within the target time window to form the current observation data; The similarity between the current observation data and the sample data in the historical geological disaster sample database is calculated. When the similarity meets the preset similarity threshold condition, it is determined that there is data representing geological disasters in the current observation data.

7. The method according to claim 1, characterized in that, The sensor includes any one or more of the following: displacement sensor, acceleration sensor, pore water pressure sensor, and rainfall sensor.

8. The method according to claim 2, characterized in that, The geological hazard identification model is a neural network model.

9. The method according to claim 5, characterized in that, The extended prediction model is a Transformer-based prediction model.

Images