Danger warning methods

Abstract

A hazard warning method is implemented by a warning unit that is connected to multiple sensing units via a signal, comprising: (A) for each sensing unit, after receiving multiple data to be analyzed from the sensing unit, obtaining a first result including a first hazard level based on the data to be analyzed using a first model based on a recurrent neural network; (B) based on the data to be analyzed, the first hazard level, and multiple correlation data between any two sensing units, obtaining a second result corresponding to each sensing unit using a second model based on a graph neural network, wherein each second result includes a second hazard level; (C) for each sensing unit, obtaining a target hazard level indicating the future hazard level of the area where the sensing unit is set up based on the first hazard level and the second hazard level.

Description

Technical Field

[0001] This invention relates to a prediction method, and more particularly to a hazard warning method applied to predicting the probability of a hazard occurring. Prior Technology

[0002] Existing early warning systems primarily rely on changes in time-series data for analysis and alerts. These systems typically analyze historical data trends to determine the likelihood of future anomalies and then provide early warnings.

[0003] However, such methods only analyze time series characteristics and fail to comprehensively consider other influencing factors and potential correlations between data, which may lead to delayed or misjudged predictions, further affecting the accuracy and reliability of the warnings.

[0004] Therefore, overcoming the above limitations to improve the accuracy and reliability of early warning systems has become an important issue in this technical field. Summary of the Invention

[0005] Therefore, the object of the present invention is to provide a hazard warning method for predicting the probability of hazard occurrence.

[0006] Therefore, the hazard warning method of the present invention is implemented by a warning unit. The warning unit is signal connected to multiple sensing units set in multiple areas. Each sensing unit is used to detect and transmit multiple data to be analyzed corresponding to multiple time points to the warning unit. The warning unit stores a first model based on a recurrent neural network and a second model based on a graph neural network. The first model is used to obtain a first result including a first degree of danger, and the second model is used to obtain a second result including a second degree of danger. The hazard warning method includes a step (A), a step (B), and a step (C).

[0007] Step (A) is that for each sensing unit, after receiving the waiting analysis data from the sensing unit, the early warning unit uses the first model to obtain the first result corresponding to the sensing unit based on the waiting analysis data corresponding to the sensing unit. Each first result includes the first danger level indicating the future danger level of the corresponding area.

[0008] Step (B) is to use the early warning unit to obtain the second result corresponding to each sensing unit by using the second model based on the waiting analysis data, the first degree of danger, and multiple correlation data between any two sensing units. Each second result includes the second degree of danger indicating the future degree of danger of the corresponding area.

[0009] Step (C) is to obtain, for each sensing unit, a target danger level that corresponds to the sensing unit and indicates the future danger level of the corresponding area based on the first danger level and the second danger level corresponding to the sensing unit.

[0010] The advantages of this invention are as follows: The early warning unit uses a first model based on a recurrent neural network to analyze the time-series data received by each sensing unit to obtain a first result. Then, a second model based on a graph neural network is used to analyze the spatial relationships between the sensing units to obtain a second result. Finally, by combining the first and second results analyzed by the two models, a more accurate and reliable assessment of the target's danger level is obtained, thereby improving the accuracy and reliability of the early warning system. Simple Explanation of the Diagram

[0011] Other features and effects of the present invention will be clearly presented in the embodiments with reference to the drawings, wherein: Figure 1 is a block diagram illustrating an early warning system for implementing one embodiment of the hazard warning method of the present invention; Figure 2 is a flowchart illustrating steps 501-507 of this embodiment of the hazard warning method of the present invention; Figure 3 is a flowchart illustrating steps 508-514 of this embodiment of the hazard warning method of the present invention; and Figure 4 is a flowchart illustrating how this embodiment of the hazard warning method of the present invention obtains a second result corresponding to each sensing unit. Implementation

[0012] Before the invention is described in detail, it should be noted that similar elements are represented by the same numbers in the following description.

[0013] Referring to Figure 1, one embodiment of the hazard warning method of the present invention is implemented by a warning system 100. The warning system 100 includes a warning unit 1, a plurality of sensing units 2 connected to the warning unit 1 via a communication network 900 and disposed in a plurality of areas, a plurality of alarm units 3 connected to the warning unit 1 via the communication network 900 and disposed in the aforementioned areas, and a management unit 4 connected to the warning unit 1 via the communication network 900.

[0014] The early warning unit 1 includes an early warning communication module 11, an early warning storage module 12, and an early warning processing module 13 electrically connected to the early warning communication module 11 and the early warning storage module 12.

[0015] The warning terminal storage module 12 stores a first model that has been trained and is based on a recurrent neural network, a second model that has been trained and is based on a graph neural network, multiple sets of first training data for training the first model, multiple sets of second training data for training the second model, location information of the setting position of each sensing unit 2, information on various strain methods corresponding to various different gases, map information including such areas and at least one escape exit, and a warning threshold.

[0016] Further explanation of the first model: The first model is used to obtain a first result containing a first hazard level and multiple gas probability values ​​corresponding to the same gas based on multiple first input data. Each first training data point used to train the first model includes multiple training gas concentration information points corresponding to multiple training time points, multiple training gas particle information points corresponding to the same training time points, a manually labeled and known training gas, and a manually labeled and known first training hazard level. The training gas is one of these gases. It is worth noting that the recurrent neural network used to train the first model can be a recurrent neural network (RNN), a long short-term memory network (LSTM), a gated recurrent unit (GRU), or an attention-based long short-term memory network (LSTM-Attention), but is not limited thereto. In other embodiments, each first training data point may only contain the training gas concentration information, the manually labeled training gas, and the manually labeled first training hazard level, but is still not limited thereto.

[0017] Further explanation of the second model: The second model is used to obtain a second result containing a second hazard level based on multiple second input data. Each second training data point used to train the second model includes training gas concentration information, training gas particle information, the first hazard level, location information, multiple training correlation data between any two sensing units 2, and a manually labeled and known second training hazard level. It is worth noting that the graph neural network used to train the second model can be a basic graph neural network (GNN) or a graph convolutional neural network (GCN), but is not limited thereto. The method for obtaining the training correlation data is the same as that described in step 503, and will not be repeated here. In other embodiments, each second training data point may only include the training gas concentration information, the first hazard level, and the manually labeled second training hazard level, but is still not limited thereto.

[0018] The management unit 4 includes a management terminal communication module 41, a management terminal display module 42, and a management terminal processing module 43 electrically connected to the management terminal communication module 41 and the management terminal display module 42.

[0019] The early warning unit 1 can be a server, but is not limited to this.

[0020] Each sensing unit 2 may be a sensor, or a sensor for sensing gas concentration and for sensing gas particles, such as a combination of a metal-oxide semiconductor sensor, a photoionization detector (PID), a non-dispersive infrared gas sensor (NDIR), and an optical particle counter (OPC), but is not limited thereto.

[0021] Each alarm unit 3 can be an alarm device, but is not limited to this.

[0022] The management unit 4 can be a tablet computer, a laptop computer, a smartphone, or a personal computer, but is not limited to these.

[0023] Referring to Figures 2-4, the following will describe the operational details of each component in the hazard warning system 100 in conjunction with this embodiment of the hazard warning method of the present invention. This embodiment of the hazard warning method includes steps 501-514.

[0024] In step 501, for each sensing unit 2, the sensing unit 2 detects the target within the corresponding area, generates multiple data points corresponding to multiple time points, and transmits the data to be analyzed to the early warning unit 1. The data to be analyzed includes multiple data points on gas concentration within the corresponding area and corresponding time points, and multiple data points on gas particle count within the corresponding area and corresponding time points. It is worth noting that in this embodiment, the target detected by the sensing unit 2 within the corresponding area is a gas; however, in other embodiments, the target detected by the sensing unit 2 within the corresponding area may not be a gas.

[0025] In step 502, for each sensing unit 2, after receiving the waiting analysis data from the sensing unit 2 through the warning terminal communication module 11, the warning terminal processing module 13 obtains the first result corresponding to the sensing unit 2 based on the waiting analysis data corresponding to the sensing unit 2 and using the first model. Each first result includes a first hazard level indicating the future hazard level of the corresponding area, and the probability value of the corresponding gas. It is worth noting that in other embodiments, the first result may only include the first hazard level.

[0026] In step 503, the early warning processing module 13, based on the waiting analysis data, the first degree of danger, the location information, and multiple correlation data between any two sensing units 2, uses the second model to obtain the second result corresponding to each sensing unit 2. Each second result includes a second degree of danger indicating the future degree of danger of the corresponding area. It is worth noting that in this embodiment, the early warning processing module 13 uses the waiting analysis data, the first degree of danger, and the location information as multiple vertex features of the second model based on a graph neural network, and aggregates the correlation data as multiple edge features. Those skilled in the art can freely adjust the information contained in the vertex features and edge features.

[0027] Referring to Figure 4, step 503 also includes steps 503A to 503D.

[0028] In step 503A, the warning terminal processing module 13 calculates a motion correlation coefficient between any two sensing units 2 based on the gas concentration information using the Pearson product-moment correlation coefficient (PPMCC) algorithm, and uses it as the majority of the correlation data.

[0029] In step 503B, the early warning processing module 13 calculates a diffusion rate of the gas between any two sensing units 2 based on the time points, gas concentration information, and location information, and uses this rate as the most significant of the associated data. The method for calculating this diffusion rate is a known technique and will not be described in detail here.

[0030] In step 503C, the early warning processing module 13 calculates a settling velocity of the gas between any two sensing units 2 based on the time points, the gas particle information, and the position information, and uses this velocity as the majority of the associated data. The method for calculating this settling velocity is a known technique and will not be described in detail here.

[0031] In step 503D, the early warning processing module 13 uses the second model to obtain the second result corresponding to each sensing unit 2 based on the waiting analysis data, the first degree of danger, the location information, and the related data.

[0032] In step 504, for each sensing unit 2, the early warning processing module 13 obtains a target danger level corresponding to the sensing unit 2 and indicating the future danger level of the corresponding area based on the first danger level and the second danger level corresponding to the sensing unit 2.

[0033] In step 505, for each sensing unit 2, the early warning processing module 13 determines whether the danger level of the target corresponding to the sensing unit 2 is greater than the early warning threshold. When the early warning processing module 13 determines that the danger level of the target corresponding to the sensing unit 2 is not greater than the early warning threshold, no action is performed; when the early warning processing module 13 determines that the danger level of the target corresponding to the sensing unit 2 is greater than the early warning threshold, step 506 is performed.

[0034] In step 506, for each sensing unit 2 whose target danger level is greater than the warning threshold, the warning processing module 13 generates a warning signal and transmits the warning signal to the alarm unit 3 corresponding to the sensing unit 2 through the warning communication module 11.

[0035] In step 507, for each alarm unit 3 that receives the warning signal, the alarm unit 3 issues a warning.

[0036] In step 508, for each sensing unit 2 whose target danger level is greater than the warning threshold, the warning terminal processing module 13 obtains a target gas with the maximum gas probability value from the gases according to the gas probability value corresponding to the sensing unit 2, obtains the strain method information corresponding to the target gas, and transmits the strain method information corresponding to the target gas to the management unit 4 through the warning terminal communication module 11.

[0037] In step 509, after receiving the strain method information corresponding to the target gas through the management terminal communication module 41, the management terminal processing module 43 displays the strain method information corresponding to the target gas on the management terminal display module 42 for management personnel to refer to and process quickly.

[0038] In step 510, the warning processing module 13 designates the area corresponding to each sensing unit 2 where the danger level of the target is greater than the warning threshold as a warning area.

[0039] In step 511, the early warning processing module 13, based on the at least one warning area and the second model, uses the gradient descent inverse tracing algorithm to obtain a pollution source area from the at least one warning area, and transmits the map information indicating the pollution source area to the management unit 4 through the early warning communication module 11.

[0040] In step 512, after receiving the map information indicating the pollution source area through the management terminal communication module 41, the management terminal processing module 43 displays the map information indicating the pollution source area on the management terminal display module 42 for management personnel to refer to.

[0041] In step 513, the warning processing module 13, based on the at least one warning area and the map information, uses a path planning algorithm to obtain an escape route to avoid the at least one warning area, and transmits the map information indicating the escape route to the management unit 4 through the warning communication module 11. It is worth noting that, in this embodiment, the path planning algorithm can be an A* search algorithm or a Dijkstra algorithm, but is not limited to these.

[0042] In step 514, after receiving the map information indicating the escape route through the management terminal communication module 41, the management terminal processing module 43 displays the map information indicating the escape route on the management terminal display module 42 for management personnel to refer to.

[0043] In summary, the hazard warning method of the present invention, by periodically executing the warning method through the warning system 100, analyzes the currently received waiting analysis data using the first model and the second model to obtain a more accurate and reliable future target hazard level for each sensing unit 2 corresponding to the area, effectively improving the accuracy and reliability of the warning system's prediction. Furthermore, the warning system 100 can also quickly identify the corresponding emergency response information and the pollution source area after determining that a target gas leak (e.g., a gas leak that is dangerous to people or objects) has occurred in the corresponding area, and provide an escape route, thus effectively achieving the purpose of the present invention.

[0044] However, the above description is merely an embodiment of the present invention and should not be construed as limiting the scope of the present invention. Any simple equivalent changes and modifications made in accordance with the scope of the patent application and the contents of the patent specification of the present invention shall still fall within the scope of the patent of the present invention.

[0045] 900: Communication Network 100: Early Warning System 1: Early Warning Unit 11: Early warning terminal communication module 12: Early warning terminal storage module 13: Early Warning Processing Module 2: Sensing Unit 3: Alarm Unit 4: Management Unit 41: Management Communication Module 42: Management Terminal Display Module 43: Management Terminal Processing Module 501~514: Steps 503A~503D: Steps

Claims

1. A hazard warning method, implemented by a warning unit, the warning unit being signal-connected to multiple sensing units located in multiple areas, each sensing unit being used to detect and transmit multiple data points to be analyzed corresponding to multiple time points to the warning unit, the warning unit storing a first model based on a recurrent neural network, a second model based on a graph neural network, multiple first training data points for training the first model, and multiple second training data points for training the second model, the first model being used to obtain a first result including a first hazard level, the second model being used to obtain a second result including a second hazard level, each first training data point including multiple training gas concentration information corresponding to multiple training time points, a manually labeled and known training gas, and a manually labeled and known first training hazard level, each second training data point including the training gas concentration information, the first hazard level, multiple training correlation data points between any two sensing units, and a manually labeled and known second training hazard level, the hazard warning method comprising the following steps: (A) For each sensing unit, after receiving the waiting analysis data from the sensing unit, the early warning unit uses the first model to obtain the first result corresponding to the sensing unit based on the waiting analysis data corresponding to the sensing unit. Each first result includes a first hazard level indicating the future hazard level of the corresponding area, and multiple gas probability values ​​corresponding to various different gases. The waiting analysis data includes multiple gas concentration information corresponding to the corresponding time points; (B) Based on the waiting analysis data, the first hazard levels, and multiple correlation data between any two sensing units, the early warning unit uses the second model to obtain the second result corresponding to each sensing unit. Each second result includes a second hazard level indicating the future hazard level of the corresponding area. The correlation data includes a correlation coefficient between any two sensing units calculated by the early warning unit based on the gas concentration information using the Pearson product-momentum correlation coefficient algorithm; and (C) For each sensing unit, the warning unit obtains a target danger level corresponding to the sensing unit and indicates the future danger level of the corresponding area based on the first danger level and the second danger level corresponding to the sensing unit.

2. The hazard warning method as described in claim 1, wherein the warning unit stores a warning threshold, and the warning unit is signal-connected to multiple alarm units installed in the area, wherein, Following step (C), the system further includes the following steps: (D) For each sensing unit, the warning unit determines whether the danger level of the target corresponding to the sensing unit is greater than the warning threshold; and (E) For each sensing unit, when the warning unit determines that the danger level of the target corresponding to the sensing unit is greater than the warning threshold, the warning unit generates and transmits a warning signal to the alarm unit corresponding to the sensing unit so that the alarm unit corresponding to the sensing unit issues a warning.

3. The hazard warning method as described in claim 1, wherein the warning unit further stores location information for setting each sensing unit, wherein, In step (B), the warning unit calculates a diffusion rate of the gas between any two sensing units based on the time points, the gas concentration information, and the location information, as one of the associated data.

4. The hazard warning method as described in claim 1, wherein the warning unit further stores location information for setting each sensing unit, and the waiting analysis data transmitted by each sensing unit further includes multiple pieces of gas particle information corresponding to the corresponding time points, wherein, In step (B), the warning unit calculates a settling rate of the gas between any two sensing units based on the time points, the gas particle information, and the location information, as one of the associated data.

5. The hazard warning method as described in claim 2, wherein the warning unit further stores information on various strain methods corresponding to the same gas, wherein, After step (D), the following steps are also included: (F) For each sensing unit, when the warning unit determines that the danger level of the target corresponding to the sensing unit is greater than the warning threshold, the warning unit obtains a target gas with the maximum gas probability value from the gases based on the gas probability values, so as to obtain the strain method information corresponding to the target gas.

6. The hazard warning method as described in claim 2, wherein, Following step (D), the following steps are also included: (G) For each sensing unit, when the warning unit determines that the danger level of the target corresponding to the sensing unit is greater than the warning threshold, the warning unit designates the area corresponding to the sensing unit as a warning area; and (H) The warning unit uses the gradient descent backtracking algorithm to obtain a pollution source area from the at least one warning area based on the at least one warning area and the second model.

7. The hazard warning method as described in claim 2, wherein the warning unit further stores map information containing such areas and at least one escape exit, wherein, After step (D), the following steps are also included: (G) For each sensing unit, when the warning unit determines that the danger level of the target corresponding to the sensing unit is greater than the warning threshold, the warning unit designates the area corresponding to the sensing unit as the warning area; and (I) The warning unit uses a path planning algorithm to obtain an escape route to avoid the at least one warning area based on the at least one warning area and the map information.

Images