An emergency rescue management system and method based on a neural network

Abstract

The present invention discloses an emergency rescue management system and method based on a neural network, which relates to the technical field of emergency rescue management. The present invention obtains historical emergency rescue data, extracts key features from the cleaned historical emergency rescue data; uses the historical emergency rescue data to train a neural network model; extracts corresponding accident key features from real-time accident data; substitutes the accident key features into the trained neural network model; transports rescue resources according to the output result of the neural network model; analyzes the transportation influence coefficient according to the rescue resource supply situation at the real-time accident occurrence point; analyzes the threshold value of the transportation influence coefficient according to the historical emergency rescue data with secondary transportation of rescue resources; and performs marking and visual alarm display according to the transportation influence coefficient at the real-time accident occurrence point. The present invention automatically adjusts the rescue plan and improves the rescue efficiency.

Description

Technical Field

[0001] The present invention relates to the technical field of emergency rescue management, and specifically to an emergency rescue management system and method based on a neural network-like network. Background Art

[0002] Urban water supply is the lifeline of a city. The safety of urban water supply is directly related to people's lives and industrial production, and even affects life safety. In order to take effective measures in a timely manner when sudden abnormal situations occur in water supply and ensure the safety of urban water supply, it is necessary to formulate an emergency rescue management plan.

[0003] Currently, in an emergency state, manual allocation of rescue resources such as personnel and equipment often results in situations of insufficient resources or misallocation; the existing technology does not further analyze the supply situation of rescue resources at the real-time accident occurrence point and cannot automatically adjust the rescue plan, which is likely to cause further property losses.

[0004] Therefore, the present invention discloses an emergency rescue management system and method based on a neural network-like network to solve the above problems. Summary of the Invention

[0005] The purpose of the present invention is to provide an emergency rescue management system and method based on a neural network-like network to solve the problems raised in the existing technology.

[0006] To achieve the above purpose, the present invention provides the following technical solution: An emergency rescue management method based on a neural network-like network, the method comprising the following steps:

[0007] S1: Obtain historical emergency rescue data, perform data cleaning on the historical emergency rescue data, and extract key features from the cleaned historical emergency rescue data;

[0008] S2: Use the historical emergency rescue data to train a neural network-like network model;

[0009] S3: Obtain real-time accident data at the real-time accident occurrence point, extract corresponding accident key features from the real-time accident data; substitute the accident key features into the trained neural network-like network model; perform rescue resource transportation according to the output result of the neural network-like network model; analyze the transportation influence coefficient according to the supply situation of rescue resources at the real-time accident occurrence point;

[0010] S4: Analyze the threshold of the transportation influence coefficient according to the historical emergency rescue data with secondary transportation of rescue resources; perform marking and visual alarm display according to the transportation influence coefficient at the real-time accident occurrence point.

[0011] According to the above solution, in S1, the key features include accident data, resource demand data corresponding to the accident data, and resource supply data corresponding to the accident data; the accident data includes accident type, accident occurrence location, accident impact range, occurrence time, monitoring data at the accident occurrence point, and sensor data at the accident occurrence point; the resource demand data corresponding to the accident data includes rescue resource name, estimated arrival time of rescue resources, and required quantity of rescue resources; the resource supply data corresponding to the accident data includes resource supply point attributes and rescue resource transportation logs; the resource supply point attributes include resource supply point location and rescue resource inventory; the rescue resource transportation log refers to the transportation records of all rescue resource transportation tasks, each rescue resource transportation task corresponds to a transportation record, and the transportation record includes the supplied quantity of rescue resources, arrival time of rescue resources, and damage information; the damage information refers to the information of rescue resources damaged during transportation, including rescue resource name and damaged quantity.

[0012] The present invention extracts key features from historical emergency rescue data, where the key features include accident data, resource demand data corresponding to the accident data, and resource supply data corresponding to the accident data; and it can effectively provide effective data support for subsequent analysis.

[0013] According to the above solution, in S2, it includes the following content:

[0014] S201: Extract accident key features and rescue key features from historical emergency rescue data; the accident key features include accident type, accident occurrence location, accident impact range, occurrence time, monitoring data at the accident occurrence point, and sensor data at the accident occurrence point; the rescue key features include rescue resource name, estimated arrival time of rescue resources, required quantity of rescue resources, and supplied quantity of rescue resources at each resource supply point.

[0015] S202: Manually score the accidents according to the accident key features of each historical emergency rescue data.

[0016] S203: Divide the accident key features and rescue key features of each historical emergency rescue data after manual scoring into a training set and a validation set according to a ratio, select the ZF-Net network structure, use the cross-entropy function as the loss function, and input the scored training set data into a neural network model for training.

[0017] S204: Analyze the accuracy of the output results of the neural network model according to the data in the validation set; if the accuracy is greater than the preset accuracy threshold, stop training; the output results include score value, rescue resource name, estimated arrival time of rescue resources, required quantity of rescue resources, and supplied quantity of rescue resources at each resource supply point.

[0018] The neural network can quickly process a large amount of data and achieve a rapid response to emergencies; through training with historical data, it provides a reliable basis for command and decision-making.

[0019] According to the above solution, in S3, it includes the following content:

[0020] S301: Obtain the real-time accident data at the real-time accident occurrence point, extract the corresponding accident key features from the real-time accident data; substitute the accident key features into the trained neural network model;

[0021] S302: In the output result of the neural network model, denote the set composed of the rescue resource names corresponding to the i-th resource supply point as: RES i ={RES （i，1） , RES （i，2） , …, RES （i，j） , …, RES （i，J）}; where RES （i，j） represents the j-th rescue resource in the set of rescue resource names corresponding to the i-th resource supply point; i ∈ [1, I], I represents the total number of resource supply points providing rescue resources; j ∈ [1, J], J represents the total number of rescue resources in the set of rescue resource names corresponding to the i-th resource supply point; conduct rescue resource transportation according to the output result of the neural network model;

[0022] S303: Analyze the transportation impact coefficient TIC according to the rescue resource supply situation at the real-time accident occurrence point:

[0023] TIC = log a [∑I i = 1∑J j = 1DR （i，j） + a] × |S(PDT i - DT)|;

[0024] Among them, DR （i，j） represents the damage rate corresponding to the rescue resource RES （i，j） , PDT i represents the delivery time of the rescue resource corresponding to the i-th resource supply point, DT represents the pre-delivery time of the rescue resource at the accident occurrence point; S() represents the sample standard deviation function; a is a constant preset by the system;

[0025] The damage rate DR （i，j） corresponding to the rescue resource RES （i，j） is equal to the damaged quantity corresponding to the rescue resource RES （i，j） divided by the supply quantity of the rescue resource corresponding to the rescue resource RES （i，j） .

[0026] According to the supply situation of rescue resources at the real-time accident occurrence point, the present invention analyzes the transportation influence coefficient, can intuitively judge the transportation situation of rescue resources, improve the rescue efficiency, and provide more effective historical emergency rescue data for subsequent accidents.

[0027] According to the above solution, in S4, it includes the following content:

[0028] S401: Mark the historical emergency rescue data with secondary transportation of rescue resources, analyze the transportation influence coefficient for the marked historical emergency rescue data, and form a set, denoted as CTIC;

[0029] S402; Analyze the transportation influence coefficient threshold CTIC according to the set of transportation influence coefficients of the marked historical emergency rescue data T :

[0030] CTIC T =min{(CTIC min +CTIC max )÷2, CTIC mean};

[0031] Among them, CTIC min represents the minimum value of the transportation influence coefficient in the set CTIC; CTIC max represents the maximum value of the transportation influence coefficient in the set CTIC; CTIC mean represents the average value of the transportation influence coefficient in the set CTIC; min{} represents the minimum value calculation;

[0032] Analyzing the transportation influence coefficient threshold according to the corresponding data of the accidents with secondary transportation of rescue resources in the historical emergency rescue data can more accurately judge the accident situation and improve the accuracy of the system.

[0033] S403: If the transportation influence coefficient TIC at the real-time accident occurrence point is less than the transportation influence coefficient threshold CTIC T , no processing is performed; if the transportation influence coefficient TIC at the real-time accident occurrence point is greater than or equal to the transportation influence coefficient threshold CTIC T , then mark the real-time accident occurrence point and perform visual alarm display for the administrator;

[0034] S404: Substitute the latest real-time accident data with the marked real-time accident occurrence point into the trained neural network model, and perform rescue resource transportation again according to the output result of the neural network model; after the real-time accident ends, substitute the latest real-time accident data and the corresponding rescue key features into S203 to retrain the neural network model.

[0035] The present invention can compare the transportation impact coefficient with the transportation impact coefficient threshold, automatically adjust the rescue plan, and improve the rescue efficiency.

[0036] In another aspect of the present application, there is provided an emergency rescue management system based on a neural network-like model. The system is implemented by applying the above-mentioned emergency rescue management method based on a neural network-like model. The system includes a rescue data feature analysis module, a neural network-like model construction module, a real-time accident analysis module, and a threshold calculation and marking module.

[0037] The rescue data feature analysis module is used to obtain historical emergency rescue data, perform data cleaning on the historical emergency rescue data, and extract key features from the cleaned historical emergency rescue data.

[0038] The neural network-like model construction module is used to train the neural network-like model using historical emergency rescue data.

[0039] The real-time accident analysis module is used to obtain real-time accident data at the real-time accident occurrence point, extract corresponding accident key features from the real-time accident data; substitute the accident key features into the trained neural network-like model; perform rescue resource transportation according to the output result of the neural network-like model; analyze the transportation impact coefficient according to the rescue resource supply situation at the real-time accident occurrence point.

[0040] The threshold calculation and marking module is used to analyze the transportation impact coefficient threshold according to the historical emergency rescue data with secondary transportation of rescue resources; perform marking and visual alarm display according to the transportation impact coefficient at the real-time accident occurrence point.

[0041] According to the above solution, the rescue data feature analysis module includes a rescue data preprocessing unit and a feature analysis unit.

[0042] The rescue data preprocessing unit is used to obtain historical emergency rescue data and perform data cleaning on the historical emergency rescue data; the data cleaning includes removing noise and outliers in the data.

[0043] The feature analysis unit is used to extract key features from the historical emergency rescue data, and the key features include accident data, resource demand data corresponding to the accident data, and resource supply data corresponding to the accident data.

[0044] According to the above solution, the neural network-like model construction module includes a feature scoring unit and a neural network-like model training unit.

[0045] The feature scoring unit is used to extract accident key features and rescue key features from the historical emergency rescue data; perform manual scoring on the accidents according to the accident key features of each historical emergency rescue data.

[0046] The neural network model training unit is used to divide the accident key features and rescue key features of each historical emergency rescue data after manual scoring into a training set and a validation set according to a ratio, select the ZF-Net network structure, use the cross-entropy function as the loss function, and input the scored training set data into the neural network model for training; analyze the accuracy of the output results of the neural network model according to the validation set data; if the accuracy is greater than the preset accuracy threshold, stop training.

[0047] According to the above solution, the real-time accident analysis module includes a rescue resource calculation unit and a transportation impact coefficient analysis unit;

[0048] The rescue resource calculation unit is used to obtain the real-time accident data of the real-time accident occurrence point, extract the corresponding accident key features from the real-time accident data; substitute the accident key features into the trained neural network model; perform rescue resource transportation according to the output results of the neural network model;

[0049] The transportation impact coefficient analysis unit is used to analyze the transportation impact coefficient according to the rescue resource supply situation at the real-time accident occurrence point.

[0050] According to the above solution, the threshold calculation and marking module includes a transportation impact coefficient threshold calculation unit and a marking and warning unit;

[0051] The transportation impact coefficient threshold calculation unit is used to mark the historical emergency rescue data with secondary transportation of rescue resources, analyze the transportation impact coefficient of the marked historical emergency rescue data, and form a set; analyze the transportation impact coefficient threshold according to the set of transportation impact coefficients of the marked historical emergency rescue data;

[0052] The marking and warning unit is used to do nothing if the transportation impact coefficient at the real-time accident occurrence point is less than the transportation impact coefficient threshold; if the transportation impact coefficient at the real-time accident occurrence point is greater than or equal to the transportation impact coefficient threshold, mark the real-time accident occurrence point and perform visual warning display for the administrator; substitute the latest real-time accident data of the marked real-time accident occurrence point into the trained neural network model, and perform rescue resource transportation again according to the output results of the neural network model; after the real-time accident ends, retrain the neural network model using the latest real-time accident data and the corresponding rescue key features.

[0053] Compared with the prior art, the beneficial effects of the present invention are as follows: The present invention extracts key features from historical emergency rescue data, where the key features include accident data, resource demand data corresponding to the accident data, and resource supply data corresponding to the accident data; it can effectively provide effective data support for subsequent analysis; the neural network can quickly process a large amount of data and achieve a rapid response to emergencies; through the training of historical data, it provides a reliable basis for command and decision-making; the present invention can analyze the transportation impact coefficient according to the rescue resource supply situation at the real-time accident occurrence point, intuitively judge the transportation situation of rescue resources, improve the rescue efficiency, and provide more effective historical emergency rescue data for subsequent accidents; according to the corresponding data of accidents with secondary transportation of rescue resources in the historical emergency rescue data, analyze the threshold of the transportation impact coefficient, and can more accurately judge the accident situation, improving the accuracy of the system; the present invention can compare the transportation impact coefficient with the threshold of the transportation impact coefficient, automatically adjust the rescue plan, and improve the rescue efficiency. BRIEF DESCRIPTION OF THE DRAWINGS

[0054] The accompanying drawings are used to provide a further understanding of the present invention, and constitute a part of the specification. Together with the embodiments of the present invention, they are used to explain the present invention, and do not constitute a limitation to the present invention. In the accompanying drawings:

[0055] Figure 1 is a schematic flow chart of an emergency rescue management method based on a neural network according to the present invention;

[0056] Figure 2 is a schematic structural diagram of an emergency rescue management system based on a neural network according to the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0057] The following will clearly and completely describe the technical solutions in the embodiments of the present invention with reference to the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present invention.

[0058] Please refer to Figure 1 , the present invention provides a technical solution: an emergency rescue management method based on a neural network, the method includes the following steps:

[0059] S1: Obtain historical emergency rescue data, perform data cleaning on the historical emergency rescue data, and extract key features from the cleaned historical emergency rescue data;

[0060] In S1, the key features include accident data, resource demand data corresponding to the accident data, and resource supply data corresponding to the accident data; the accident data includes accident type, accident occurrence location, accident impact range, occurrence time, monitoring data at the accident occurrence point, and sensor data at the accident occurrence point; the resource demand data corresponding to the accident data includes rescue resource name, estimated arrival time of the rescue resource, and required quantity of the rescue resource; the resource supply data corresponding to the accident data includes resource supply point attributes and rescue resource transportation logs; the resource supply point attributes include resource supply point location and rescue resource inventory; the rescue resource transportation log refers to the transportation records of all rescue resource transportation tasks, each rescue resource transportation task corresponds to a transportation record, and the transportation record includes the supplied quantity of the rescue resource, arrival time of the rescue resource, and damage information; the damage information refers to the information of the damaged rescue resources during transportation, including the rescue resource name and the damaged quantity.

[0061] S2: Train the neural network model using historical emergency rescue data;

[0062] In S2, the following steps are included:

[0063] S201: Extract the accident key features and rescue key features from the historical emergency rescue data; the accident key features include accident type, accident occurrence location, accident impact range, occurrence time, monitoring data at the accident occurrence point, and sensor data at the accident occurrence point; the rescue key features include rescue resource name, estimated arrival time of the rescue resource, required quantity of the rescue resource, and the supplied quantity of the rescue resource at each resource supply point;

[0064] S202: Manually score the accidents according to the accident key features of each historical emergency rescue data;

[0065] S203: Divide the accident key features and rescue key features of each historical emergency rescue data after manual scoring into a training set and a validation set according to a ratio, select the ZF-Net network structure, use the cross-entropy function as the loss function, and input the scored training set data into the neural network model for training;

[0066] S204: Analyze the accuracy of the output results of the neural network model according to the data in the validation set; if the accuracy is greater than the preset accuracy threshold, stop training; the output results include the scored value, rescue resource name, estimated arrival time of the rescue resource, required quantity of the rescue resource, and the supplied quantity of the rescue resource at each resource supply point.

[0067] S3: Obtain the real-time accident data at the real-time accident occurrence point, extract the corresponding accident key features from the real-time accident data; substitute the accident key features into the trained neural network-like model; perform rescue resource transportation according to the output result of the neural network-like model; analyze the transportation influence coefficient according to the rescue resource supply situation at the real-time accident occurrence point;

[0068] In S3, it includes the following:

[0069] S301: Obtain the real-time accident data at the real-time accident occurrence point, extract the corresponding accident key features from the real-time accident data; substitute the accident key features into the trained neural network-like model;

[0070] S302: In the output result of the neural network-like model, denote the set composed of the rescue resource names corresponding to the i-th resource supply point as: RES i ={RES （i，1） , RES （i，2） , …, RES （i，j） , …, RES （i，J）}; where RES （i，j） represents the j-th rescue resource in the rescue resource name set corresponding to the i-th resource supply point; i ∈ [1, I], I represents the total number of resource supply points providing rescue resources; j ∈ [1, J], J represents the total number of rescue resources in the rescue resource name set corresponding to the i-th resource supply point; perform rescue resource transportation according to the output result of the neural network-like model;

[0071] S303: Analyze the transportation influence coefficient TIC according to the rescue resource supply situation at the real-time accident occurrence point:

[0072] TIC = log a [∑I i = 1∑J j = 1DR （i，j） + a] × |S(PDT i - DT)|;

[0073] Among them, DR （i，j） represents the damage rate corresponding to the rescue resource RES （i，j） , PDT i represents the delivery time of the rescue resource corresponding to the i-th resource supply point, DT represents the pre-delivery time of the rescue resource at the accident occurrence point; S() represents the sample standard deviation function; a is a constant preset by the system;

[0074] The damage rate DR （i，j） corresponding to the rescue resource RES （i，j） is equal to the damaged quantity corresponding to the rescue resource RES （i，j） divided by the supply quantity of the rescue resource corresponding to the rescue resource RES （i，j） .

[0075] S4: Analyze the threshold of the transportation impact coefficient based on the historical emergency rescue data with secondary transportation rescue resources; mark and visually alarm and display according to the transportation impact coefficient at the real-time accident occurrence point.

[0076] In S4, the following contents are included:

[0077] S401: Mark the historical emergency rescue data with secondary transportation rescue resources, analyze the transportation impact coefficient for the marked historical emergency rescue data, and form a set, denoted as CTIC;

[0078] S402; Analyze the threshold of the transportation impact coefficient CTIC according to the set of transportation impact coefficients of the marked historical emergency rescue data T :

[0079] CTIC T = min{(CTIC min + CTIC max )÷2, CTIC mean};

[0080] Among them, CTIC min represents the minimum value of the transportation impact coefficient in the set CTIC; CTIC max represents the maximum value of the transportation impact coefficient in the set CTIC; CTIC mean represents the average value of the transportation impact coefficient in the set CTIC; min{} represents the minimum value calculation;

[0081] Example 1: In this example, CTIC = {5, 8, 8, 7, 9, 6, 10, 12};

[0082] Therefore, in this example, CTIC min = 5; CTIC max = 12, then (CTIC min + CTIC max )÷2 = 8.5;

[0083] CTIC mean = (5 + 8 + 8 + 7 + 9 + 6 + 10 + 12)÷8 = 8.125;

[0084] Then CTIC T = min{8.5, 8.125} = 8.125;

[0085] S403: If the transportation impact coefficient TIC at the real-time accident occurrence point is less than the threshold of the transportation impact coefficient CTIC T , no processing is performed; if the transportation impact coefficient TIC at the real-time accident occurrence point is greater than or equal to the threshold of the transportation impact coefficient CTIC T, mark the real-time accident occurrence point and display visual alarms to the administrator;

[0086] S404: Substitute the latest real-time accident data with the marked real-time accident occurrence point into the trained neural network model, and perform rescue resource transportation again according to the output result of the neural network model; after the real-time accident ends, substitute the latest real-time accident data and the corresponding rescue key features into S203 to retrain the neural network model.

[0087] Please refer to Figure 2 , the present invention provides a technical solution: an emergency rescue management system based on a neural network, which includes a rescue data feature analysis module, a neural network model construction module, a real-time accident analysis module, and a threshold calculation and marking module;

[0088] The rescue data feature analysis module is used to obtain historical emergency rescue data, clean the historical emergency rescue data, and extract key features from the cleaned historical emergency rescue data;

[0089] The neural network model construction module is used to train the neural network model with historical emergency rescue data;

[0090] The real-time accident analysis module is used to obtain the real-time accident data at the real-time accident occurrence point, extract the corresponding accident key features from the real-time accident data; substitute the accident key features into the trained neural network model; perform rescue resource transportation according to the output result of the neural network model; analyze the transportation influence coefficient according to the rescue resource supply situation at the real-time accident occurrence point;

[0091] The threshold calculation and marking module is used to analyze the threshold of the transportation influence coefficient according to the historical emergency rescue data with secondary transportation of rescue resources; perform marking and visual alarm display according to the transportation influence coefficient at the real-time accident occurrence point.

[0092] The rescue data feature analysis module includes a rescue data preprocessing unit and a feature analysis unit;

[0093] The rescue data preprocessing unit is used to obtain historical emergency rescue data and clean the historical emergency rescue data; data cleaning includes removing noise and outliers from the data;

[0094] The feature analysis unit is used to extract key features from historical emergency rescue data, and the key features include accident data, resource demand data corresponding to the accident data, and resource supply data corresponding to the accident data.

[0095] The neural network model construction module includes a feature scoring unit and a neural network model training unit;

[0096] The feature scoring unit is used to extract the accident key features and rescue key features from the historical emergency rescue data; and conduct manual scoring on the accidents according to the accident key features of each historical emergency rescue data.

[0097] The neural network model training unit is used to divide the accident key features and rescue key features of each historical emergency rescue data after manual scoring into a training set and a validation set according to a ratio, select the ZF-Net network structure, use the cross-entropy function as the loss function, and input the scored training set data into the neural network model for training; analyze the accuracy of the output results of the neural network model according to the validation set data; if the accuracy is greater than the preset accuracy threshold, stop training.

[0098] The real-time accident analysis module includes a rescue resource calculation unit and a transportation impact coefficient analysis unit.

[0099] The rescue resource calculation unit is used to obtain the real-time accident data at the real-time accident occurrence point, extract the corresponding accident key features from the real-time accident data; substitute the accident key features into the trained neural network model; and conduct rescue resource transportation according to the output results of the neural network model.

[0100] The transportation impact coefficient analysis unit is used to analyze the transportation impact coefficient according to the rescue resource supply situation at the real-time accident occurrence point.

[0101] The threshold calculation and marking module includes a transportation impact coefficient threshold calculation unit and a marking and warning unit.

[0102] The transportation impact coefficient threshold calculation unit is used to mark the historical emergency rescue data with secondary transportation of rescue resources, analyze the transportation impact coefficients of the marked historical emergency rescue data, and form a set; analyze the transportation impact coefficient threshold according to the set of transportation impact coefficients of the marked historical emergency rescue data.

[0103] The marking and warning unit is used to not perform any processing if the transportation impact coefficient at the real-time accident occurrence point is less than the transportation impact coefficient threshold; if the transportation impact coefficient at the real-time accident occurrence point is greater than or equal to the transportation impact coefficient threshold, mark the real-time accident occurrence point and conduct visual warning display for the administrator; substitute the latest real-time accident data of the marked real-time accident occurrence point into the trained neural network model, and conduct rescue resource transportation again according to the output results of the neural network model; after the real-time accident ends, retrain the neural network model using the latest real-time accident data and the corresponding rescue key features.

[0104] It should be noted that, in this document, relational terms such as first and second are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any actual relationship or order between these entities or operations. Moreover, the terms "comprising", "including" or any other variant thereof are intended to cover non-exclusive inclusion, such that a process, method, article or apparatus comprising a series of elements not only includes those elements but also includes other elements not expressly listed, or elements inherent to such process, method, article or apparatus.

[0105] For those skilled in the art, it is obvious that the present invention is not limited to the details of the above-described exemplary embodiments, and that the present invention can be implemented in other specific forms without departing from the spirit or essential characteristics of the present invention. Therefore, in any respect, the embodiments should be regarded as exemplary and non-limiting. The scope of the present invention is defined by the appended claims rather than the above description, and thus is intended to embrace all changes that fall within the meaning and scope of the equivalent elements of the claims. Any reference signs in the claims should not be construed as limiting the claims involved.

Claims

1. An emergency rescue management method based on a neural network, characterized in that, The method includes the following steps: S1: Obtain historical emergency rescue data, clean the historical emergency rescue data, and extract key features from the cleaned historical emergency rescue data; S2: Use the historical emergency rescue data to train a neural network model; S3: Obtain real-time accident data at the real-time accident occurrence point, and extract corresponding accident key features from the real-time accident data; Substitute the accident key features into the trained neural network model; conduct rescue resource transportation according to the output result of the neural network model; analyze the transportation influence coefficient according to the rescue resource supply situation at the real-time accident occurrence point; S4: Analyze the threshold of the transportation influence coefficient according to the historical emergency rescue data with secondary transportation of rescue resources; Conduct marking and visual alarm display according to the transportation influence coefficient at the real-time accident occurrence point; In S2, it includes the following content: S201: Extract accident key features and rescue key features from the historical emergency rescue data; The accident key features include accident type, accident occurrence point location, accident influence range, occurrence time, accident occurrence point monitoring data, and accident occurrence point sensor data; the rescue key features include rescue resource name, rescue resource pre-delivery time, rescue resource demand, and rescue resource supply at each resource supply point; S202: Manually score the accidents according to the accident key features of each historical emergency rescue data; S203: Divide the accident key features and rescue key features of each historical emergency rescue data after manual scoring into a training set and a validation set according to a ratio, select the ZF-Net network structure, use the cross-entropy function as the loss function, and input the scored training set data into the neural network model for training; S204: Analyze the accuracy of the output result of the neural network model according to the validation set data; if the accuracy is greater than the preset accuracy threshold, stop training; the output result includes the score value, rescue resource name, rescue resource pre-delivery time, rescue resource demand, and rescue resource supply at each resource supply point; In S3, it includes the following content: S301: Obtain real-time accident data at the real-time accident occurrence point, and extract corresponding accident key features from the real-time accident data; Substitute the accident key features into the trained neural network model; S302: In the output result of the neural network model, denote the set composed of the names of the rescue resources corresponding to the \(i\)-th resource supply point as: RES i ={RES （i，1） , RES （i，2） , …, RES （i，j） , …, RES （i，J）}; where RES （i，j） represents the \(j\)-th rescue resource in the set of the names of the rescue resources corresponding to the \(i\)-th resource supply point; i ∈ [1, I], where I represents the total number of resource supply points providing rescue resources; j ∈ [1, J], where J represents the total number of rescue resources in the set of rescue resource names corresponding to the i-th resource supply point; conduct rescue resource transportation according to the output result of the neural network model; S303: Analyze the transportation influence coefficient TIC according to the rescue resource supply situation at the real-time accident occurrence point; ； Among them, DR （i，j） represents the damage rate of the rescue resource RES （i，j） corresponding to PDT i represents the delivery time of the rescue resource corresponding to the i-th resource supply point, DT represents the pre-delivery time of the rescue resource at the accident site; S() represents the sample standard deviation function; a is a constant preset by the system; Rescue resource RES （i，j） Corresponding damage rate DR （i，j） Equal to the rescue resource RES （i，j） Corresponding damaged quantity divided by the rescue resource RES （i，j） Corresponding rescue resource supply quantity; In S4, it includes the following content: S401: Mark the historical emergency rescue data with secondary transportation of rescue resources, analyze the transportation influence coefficient of the marked historical emergency rescue data, and form a set, denoted as CTIC; S402: Analyze the transport impact coefficient threshold CTIC based on the set of transport impact coefficients of the marked historical emergency rescue data T : ； Among them, CTIC min represents the minimum value of the transportation influence coefficient in the set CTIC; CTIC max represents the maximum value of the transportation influence coefficient in the set CTIC; CTIC mean represents the average value of the transportation influence coefficient in the set CTIC; min{} represents the minimum value calculation; S403: If the transportation impact coefficient TIC of the real-time accident occurrence point is less than the transportation impact coefficient threshold CTIC T , no processing is performed; if the transportation impact coefficient TIC of the real-time accident occurrence point is greater than or equal to the transportation impact coefficient threshold CTIC T , then mark the real-time accident occurrence point and perform visual alarm display for the administrator; S404: Substitute the latest real-time accident data marked with the real-time accident occurrence point into the trained neural network model, and transport rescue resources again according to the output result of the neural network model; after the real-time accident ends, substitute the latest real-time accident data and the corresponding rescue key features into S203 to retrain the neural network model.

2. The emergency rescue management method based on a neural network according to claim 1, characterized in that: In S1, the key features include accident data, resource demand data corresponding to the accident data, and resource supply data corresponding to the accident data; the accident data includes accident type, accident occurrence point location, accident impact range, occurrence time, accident occurrence point monitoring data, and accident occurrence point sensor data; The resource demand data corresponding to the accident data includes rescue resource name, estimated time of arrival of rescue resources, and required quantity of rescue resources; the resource supply data corresponding to the accident data includes resource supply point attributes and rescue resource transportation logs; the resource supply point attributes include resource supply point location and rescue resource inventory; the rescue resource transportation log refers to the transportation records of all rescue resource transportation tasks, each rescue resource transportation task corresponds to a transportation record, and the transportation record includes the quantity of rescue resources supplied, the arrival time of rescue resources, and damage information; the damage information refers to the information of damaged rescue resources during transportation, including rescue resource name and quantity of damage.

3. An emergency rescue management system based on a neural network, the system is implemented by applying the emergency rescue management method based on a neural network described in any one of claims 1-2, and is characterized in that, The system includes a rescue data feature analysis module, a neural network model construction module, a real-time accident analysis module, and a threshold calculation and marking module; The rescue data feature analysis module is used to obtain historical emergency rescue data, clean the historical emergency rescue data, and extract key features from the cleaned historical emergency rescue data; The neural network model construction module is used to train the neural network model using historical emergency rescue data; The real-time accident analysis module is used to obtain the real-time accident data of the real-time accident occurrence point and extract the corresponding accident key features from the real-time accident data; Substitute the accident key features into the trained neural network model; transport rescue resources according to the output result of the neural network model; analyze the transportation impact coefficient according to the rescue resource supply situation at the real-time accident occurrence point; The threshold calculation and marking module is used to analyze the threshold of the transportation impact coefficient according to the historical emergency rescue data with secondary transportation of rescue resources; Mark and visually alarm and display according to the transportation impact coefficient at the real-time accident occurrence point.

4. The emergency rescue management system based on a neural network according to claim 3, characterized in that: The rescue data feature analysis module includes a rescue data preprocessing unit and a feature analysis unit; The rescue data preprocessing unit is used to obtain historical emergency rescue data and clean the historical emergency rescue data; the data cleaning includes removing noise and outliers from the data; The feature analysis unit is used to extract the key features from the historical emergency rescue data, and the key features include accident data, resource demand data corresponding to the accident data, and resource supply data corresponding to the accident data.

5. The emergency rescue management system based on a neural network according to claim 3, characterized in that: The neural network model construction module includes a feature scoring unit and a neural network model training unit; The feature scoring unit is used to extract the accident key features and rescue key features from the historical emergency rescue data; and conduct manual scoring of the accidents based on the accident key features of each historical emergency rescue data. The neural network model training unit is used to divide the accident key features and rescue key features of each historical emergency rescue data after manual scoring into a training set and a validation set according to a ratio, select the ZF-Net network structure, use the cross-entropy function as the loss function, and input the scored training set data into the neural network model for training; analyze the accuracy of the output results of the neural network model according to the validation set data; if the accuracy is greater than the preset accuracy threshold, stop training.

6. The emergency rescue management system based on a neural network according to claim 3, characterized in that: The real-time accident analysis module includes a rescue resource calculation unit and a transportation impact coefficient analysis unit. The rescue resource calculation unit is used to obtain the real-time accident data at the real-time accident occurrence point, and extract the corresponding accident key features from the real-time accident data. Substitute the accident key features into the trained neural network model; conduct rescue resource transportation according to the output results of the neural network model. The transportation impact coefficient analysis unit is used to analyze the transportation impact coefficient according to the rescue resource supply situation at the real-time accident occurrence point.

7. The emergency rescue management system based on a neural network according to claim 3, wherein: The threshold calculation and marking module includes a transportation impact coefficient threshold calculation unit and a marking and warning unit. The transportation impact coefficient threshold calculation unit is used to mark the historical emergency rescue data with secondary transportation of rescue resources, analyze the transportation impact coefficient of the marked historical emergency rescue data, and form a set; analyze the transportation impact coefficient threshold according to the set of transportation impact coefficients of the marked historical emergency rescue data. The marking and warning unit is used to do nothing if the transportation impact coefficient at the real-time accident occurrence point is less than the transportation impact coefficient threshold. If the transportation impact coefficient at the real-time accident occurrence point is greater than or equal to the transportation impact coefficient threshold, mark the real-time accident occurrence point, and conduct visual warning display for the administrator; substitute the latest real-time accident data of the marked real-time accident occurrence point into the trained neural network model, and conduct rescue resource transportation again according to the output results of the neural network model. After the real-time accident ends, retrain the neural network model using the latest real-time accident data and the corresponding rescue key features.

Images