Deep learning-driven fire fighting truck travel time prediction method based on small sample data multi-dimensional expansion

Abstract

The invention provides a deep learning driven small sample data multi-dimensional expansion fire fighting truck travel time prediction method, and relates to the technical field of emergency management, and the method comprises the steps: building a fire rescue travel time data set through combining fire fighting station information and historical rescue travel time information; expanding the fire rescue travel time data set by using similar historical data between fire stations and social vehicle travel time information to obtain a training sample set; constructing a fire fighting truck travel time prediction model based on multi-graph convolution, and training the model according to the training sample; and according to the trained fire fighting truck travel time prediction model, realizing fire fighting truck travel time prediction. The fire fighting station and the historical rescue information are combined, and the historical data of the fire fighting truck and the social vehicle data are utilized to expand the training set, so that the accurate prediction of the travel time of the fire fighting truck is realized, errors caused by insufficient data are avoided, and the prediction accuracy and the emergency response efficiency are improved.

Description

Technical Field

[0001] The present application relates to the fields of emergency management and traffic forecasting technology, and in particular to a fire truck travel time prediction method based on multi-dimensional expansion of small sample data driven by deep learning. Background Art

[0002] As a vehicle with the right of way, fire trucks can reach the scene at the fastest speed during emergency rescue missions without being restricted by ordinary traffic rules. The travel time of fire trucks can be used as an indicator to judge emergency rescue capabilities, and it is also the basis for fire stations to dispatch vehicles to carry out rescue activities, which is of great significance. If the travel time of fire trucks can be accurately predicted, it will be of great help both for the effectiveness of emergency rescue and for judging the timeliness of emergency rescue.

[0003] However, the travel time prediction of fire trucks has the following difficulties: (1) Although fire trucks have the right of way, they are still inevitably affected by special circumstances such as road congestion and intersection queues. The driving speed of fire trucks under complex traffic conditions is not the free flow speed, which makes the travel time of fire trucks difficult to predict. (2) Existing mature vehicle travel time prediction, such as taxis, private cars, buses, etc., can support the travel time prediction for these ordinary social vehicles due to their extremely high sampling rate and large historical sample data size. However, the number of fire trucks dispatching is limited, and each fire station only dispatches 1-2 times a day, resulting in an extremely small historical sample data size, which makes it difficult to support the research of fire truck travel time prediction methods. Summary of the invention

[0004] The present application aims to solve one of the technical problems in the related art at least to some extent.

[0005] To this end, the first objective of this application is to propose a fire truck travel time prediction method with multi-dimensional expansion of small sample data driven by deep learning.

[0006] The second object of the present application is to provide a device.

[0007] The third objective of the present application is to provide an electronic device.

[0008] A fourth objective of the present application is to provide a computer-readable storage medium.

[0009] A fifth object of the present application is to provide a computer program product.

[0010] To achieve the above-mentioned purpose, the first embodiment of the present application proposes a fire truck travel time prediction method based on multi-dimensional expansion of small sample data driven by deep learning, including:

[0011] Combine the fire station information and historical rescue travel time information to build a fire rescue travel time data set;

[0012] The fire rescue travel time data set is filled according to historical fire rescue data information, and the fire rescue travel time data set is expanded by using similar historical data between fire stations and social vehicle travel time information to obtain a training sample set;

[0013] Constructing a fire truck travel time prediction model based on multi-graph convolution, and training the fire truck travel time prediction model according to the training samples;

[0014] The fire truck travel time is predicted based on the trained fire truck travel time prediction model.

[0015] Optionally, the fire station information and historical rescue travel time information are combined to construct a fire rescue travel time data set, including:

[0016] In a specific city, the set S of all fire stations that can dispatch fire trucks is defined as:

[0017] S={S0,S1,...,S n}

[0018] In the formula, S represents the fire station, and the subscript is the number of each fire station;

[0019] For each fire station S, define its feature set in a specific period T for:

[0020]

[0021] in, represents the travel time from the fire station S to the rescue locations at different distances, and the corresponding rescue distances are: d0, d1, ..., d n ;

[0022] According to the feature set of fire station S in a specific period of time T The fire station S is divided into two groups in each historical period T0~T m Different distance ranges d0,d1,...,d n All features are represented as the following n×m matrix A n×m :

[0023]

[0024] Among them, T0 is limited to a time point in the period of 0:00 to 1:00 on any natural day;

[0025] For all fire station sets S, all time period sets T and all distance ranges d0,d1,...,d n A tensor feature library of s×n×m is established as the fire rescue travel time data set.

[0026] Optionally, filling the fire rescue travel time data set according to historical fire rescue data information includes:

[0027] Get historical fire rescue data information {I0,I1,...,I n}, I refers to a recorded rescue event;

[0028] According to the historical fire rescue data information {I0, I1, ..., I n}, get the latitude and longitude of the starting point of each rescue event I Rescue point latitude and longitude Police dispatch time Arrival time

[0029] According to the latitude and longitude of the starting point Rescue point latitude and longitude Police dispatch time Arrival time Calculate the travel time t for each rescue event I I and travel distance d I , the expression is:

[0030]

[0031] Among them, d eu and d ma They are the Euclidean distance and Manhattan distance calculated based on the latitude and longitude of the departure point and the longitude and longitude of the rescue point respectively;

[0032] According to the dispatch time The time period and the distance travelled I Corresponding distance range, the travel time t I Fill in the tensor feature library.

[0033] Optionally, the process of expanding the fire rescue travel time data set by using similar historical data between fire stations includes:

[0034] For a given fire station S i and S j , according to their respective historical feature datasets and , calculate the similarity of historical data between the two fire stations;

[0035] According to the similarity of historical data between two fire stations, the fire rescue travel time data set is expanded.

[0036] Optionally, for a given fire station S i and S j , according to their respective historical feature datasets and Calculate the similarity of historical data between two fire stations, including:

[0037] Historical feature dataset and Respectively reduced to and That is, the data of m time periods in the historical feature data is reduced to 24 time periods, and the formula for the travel time value of time period 0 to 1 is as follows:

[0038]

[0039] That is, the travel time value in the period 0 to 1 is equal to the average value of the period 0:00 to 1:00 of each natural day in the historical period, and n day for The number of non-zero values ​​in the set, the remaining 23 periods The calculation method is similar to this, and by analogy, we can get the fire station S i and S j Feature matrix after dimensionality reduction and

[0040] According to the feature matrix after dimensionality reduction and Calculate the similarity of historical data between two fire stations The expression is:

[0041]

[0042] Among them, L(S i ,S j )、L′(S i ,S j ) and L″(S i ,S j ) are calculated as follows:

[0043]

[0044] in, is a matrix The average of all non-zero values ​​in is a matrix The average of all non-zero values ​​in ;

[0045]

[0046] Among them, σ Si is a matrix The standard deviation of all non-zero values ​​in , σ Sj is a matrix The standard deviation of all non-zero values ​​in ;

[0047]

[0048] where ρ(·) is the Pearson correlation coefficient between two vectors, is a matrix Vector data flattened by rows and columns, is a matrix Vector data flattened by rows and columns.

[0049] Optionally, the expanding the fire rescue travel time data set according to the similarity of historical data between the two fire stations includes:

[0050] S1. Traverse the fire station set S and take out any fire station S i and its characteristic matrix Mark as visited, if there is no fire station that has not been visited, go to step S4;

[0051] S2. If there are other fire stations S j With Fire Station S i The similarity satisfies If κ is the similarity threshold, go to step S3, otherwise go to step S1;

[0052] S3. Reserved with Fire Station S i The fire station with the highest similarity j , and to the fire station S i The characteristic matrix To supplement, the expression is:

[0053]

[0054] S4. The loop ends and the characteristic matrix of all fire stations is output

[0055] Optionally, the process of expanding the fire rescue travel time data set using social vehicle travel time information includes:

[0056] Collect the GPS positioning point set D of all social vehicles in the same city;

[0057] In the GPS positioning point set D, select all GPS positioning point sets D that belong to the same administrative division as the fire station S. S ;

[0058] Meet at GPS location D S , filter all GPS positioning points in a specific period T

[0059] According to GPS positioning point collection Calculate the travel time corresponding to different distance ranges of fire station S during this period T The expression is:

[0060]

[0061] in, is a range of different distances d0,d1,...,d n The median value of Is a collection The average instantaneous velocity of all the positioning points in;

[0062] If there is no historical data of fire trucks that meets the calculation requirements in the tensor feature library Will Fill in the corresponding position in the tensor feature library, where is the corresponding attenuation factor;

[0063] The above screening step, calculation step, and filling step are repeated until all elements in the fire station set S and all elements in the time period set T are traversed.

[0064] Optionally, the constructing of a fire truck travel time prediction model based on multi-graph convolution includes:

[0065] The structure of the model is determined as follows: a graph structure model is constructed with fire stations as graph nodes and the relationships between stations as edges;

[0066] The output of the confirmation model is: All fire stations S = {S0, S1, ..., S n In the forecast period T m+1 Travel time τ at a specific rescue distance d (out) , a total of s nodes;

[0067] Confirm that the input to the model is:

[0068] τ (0)

[0069] Among them, the τ (0) is a s×(k+h)-dimensional matrix, τ (0) Includes: the first k periods T m-k+1 ,...,T m-1 ,T m , all fire stations S={S0,S1,...,Sn} The travel time at a specific rescue distance d, i.e., an s×k dimensional matrix; and the prediction period T m+1 Same first h natural day period All fire stations S={S0,S1,...,S n} The travel time at a specific rescue distance d, i.e., an s×h dimensional matrix;

[0070] Confirm that the first graph convolution output of the model is:

[0071]

[0072] Among them, the τ (1) It is still an s×(k+h)-dimensional matrix. A1 is an s×s-dimensional matrix, representing the adjacency relationship between fire stations. i and j The spatial distance of is less than or equal to the preset spatial distance δ, then in the adjacency matrix A1 a ij =1, otherwise it is 0; I is the identity matrix, is the degree matrix, σ() is the activation function, θ (0) are all the parameters involved in the first layer of graph convolution;

[0073] Confirm that the second graph convolution output of the model is:

[0074]

[0075] Among them, the τ (2) It is still an s×(k+h)-dimensional matrix, A2 is an s×s-dimensional matrix, representing the correlation between fire stations, θ (1) are all the parameters involved in the second layer of graph convolution, at any position a in the matrix A2 ij The values ​​are:

[0076]

[0077] Where cor() is the Pearson correlation coefficient of two vectors;

[0078] Determine the fully connected layer of the model, which is used to establish τ (2) and τ (out) The fully connected mapping is established.

[0079] Optionally, the training of the fire truck travel time prediction model according to the training samples includes:

[0080] The training sample set is calculated continuously. When entering the period T m After that, use T mThe fire truck travel time prediction model is trained using the training sample set of all previous time periods until the model converges, thereby obtaining a trained fire truck travel time prediction model, wherein each distance range corresponds to a model, and a total of n trained fire truck travel time prediction models are obtained.

[0081] Optionally, the prediction of the fire truck travel time is realized according to the trained fire truck travel time prediction model, including:

[0082] According to the n trained fire truck travel time prediction models, the travel time data τ of the next period is calculated respectively. m+1 , get each fire station in T m+1 During the time period, different distance ranges d0, d1, ..., d n The travel time, the prediction matrix of travel time is as follows:

[0083]

[0084] When in T m+1 The actual demand point is generated in the time period, the distance between the fire station and the demand point is automatically calculated, and the corresponding travel time is selected from the travel time prediction matrix diagram to realize the prediction of the travel time from the fire truck to the demand point.

[0085] To achieve the above-mentioned purpose, the second embodiment of the present application proposes a fire truck travel time prediction device with multi-dimensional expansion of small sample data driven by deep learning, comprising:

[0086] A travel time data set building module, used to build a fire rescue travel time data set by combining fire station information and historical rescue travel time information;

[0087] An expansion module is used to fill the fire rescue travel time data set according to historical fire rescue data information, and to expand the fire rescue travel time data set by using similar historical data between fire stations and social vehicle travel time information to obtain a training sample set;

[0088] A model building and training module, used to build a fire truck travel time prediction model based on multi-graph convolution, and train the fire truck travel time prediction model according to the training samples;

[0089] The prediction module is used to predict the travel time of a fire truck based on the trained fire truck travel time prediction model.

[0090] To achieve the above-mentioned purpose, the third aspect of the present application provides an electronic device, including: a processor, and a memory communicatively connected to the processor;

[0091] The memory stores computer-executable instructions;

[0092] The processor executes the computer-executable instructions stored in the memory to implement the method as described in any one of the first aspects.

[0093] To achieve the above-mentioned purpose, the fourth aspect embodiment of the present application proposes a computer-readable storage medium, in which computer-readable storage medium is stored computer execution instructions, and when the computer execution instructions are executed by a processor, they are used to implement the method as described in any one of the first aspects.

[0094] To achieve the above-mentioned purpose, the fifth aspect of the present application proposes a computer program product, which implements any method in the first aspect when executed by a processor.

[0095] The fire truck travel time prediction method, device, electronic device and storage medium for multi-dimensional expansion of small sample data driven by deep learning provided in the present application, by combining fire station information and historical rescue travel time information, and using the data of similar fire stations and social vehicle travel time information to expand the historical fire rescue small sample, realizes the enrichment and diversification of fire rescue travel time data, expands the travel time data of specific time periods, specific distances, and specific fire stations, so that the data set can meet the requirements of deep learning calculations, avoids the problem of poor model generalization ability caused by insufficient data volume, and improves the accuracy and reliability of the model; by constructing a fire truck travel time prediction model based on multi-graph convolution, two graph structures, namely adjacency graph and feature similarity measurement, are designed in the model for the travel time features of fire trucks. The two graph structures are used to describe the connection between different fire stations respectively, and the spatial features of the fire truck travel time information are mined through multi-graph convolution, and the model is trained using the expanded training samples, so as to realize the accurate prediction of the fire truck travel time, avoid the inaccuracy of traditional prediction methods under complex traffic conditions, and improve the efficiency and effectiveness of emergency response. Additional aspects and advantages of the present application will be given in part in the description below, and in part will become apparent from the description below, or will be learned through the practice of the present application. BRIEF DESCRIPTION OF THE DRAWINGS

[0096] The above and / or additional aspects and advantages of the present application will become apparent and easily understood from the following description of the embodiments in conjunction with the accompanying drawings, in which:

[0097] Figure 1 A flowchart of a fire truck travel time prediction method based on multi-dimensional expansion of small sample data driven by deep learning provided in an embodiment of the present application;

[0098] Figure 2A schematic diagram of the structure of a fire truck travel time prediction device with multi-dimensional expansion of small sample data driven by deep learning provided in an embodiment of the present application. DETAILED DESCRIPTION

[0099] Embodiments of the present application are described in detail below, and examples of the embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals throughout represent the same or similar elements or elements having the same or similar functions. The embodiments described below with reference to the accompanying drawings are exemplary and are intended to be used to explain the present application, and should not be construed as limiting the present application.

[0100] The travel time prediction of fire trucks has the following difficulties:

[0101] (1) Although fire trucks have the right of way, they are still inevitably affected in special circumstances such as road congestion and intersection queues. The speed of fire trucks under complex traffic conditions is not the free flow speed, which makes the travel time of fire trucks difficult to predict.

[0102] (2) Existing mature vehicle travel time prediction, such as taxis, private cars, and buses, can support the travel time prediction for these common social vehicles due to their extremely high sampling rate and large historical sample data size. However, the number of fire truck dispatches is limited, and each fire station only dispatches 1-2 times a day, resulting in an extremely small historical sample data size, which makes it difficult to support the research of fire truck travel time prediction methods.

[0103] To address this problem, the present application provides a fire truck travel time prediction method based on multi-dimensional expansion of small sample data driven by deep learning. Figure 1 A flowchart of a fire truck travel time prediction method based on multi-dimensional expansion of small sample data driven by deep learning provided in an embodiment of the present application. Figure 1 As shown, the method comprises the following steps:

[0104] Step 101, combining fire station information and historical rescue travel time information to construct a fire rescue travel time data set.

[0105] This application is based on a deep learning model, so a complete training dataset needs to be established.

[0106] In the embodiment of the present application, within a specific city, the set S of all fire stations to which fire trucks can be dispatched is defined as:

[0107] S={S0,S1,...,S n}

[0108] In the formula, S represents the fire station, and the subscript is the number of each fire station;

[0109] For each fire station S, define its characteristics (i.e. travel time in different distance ranges) in a specific period T for:

[0110]

[0111] in, It represents the time taken to reach the rescue location at different distances from the fire station S. The corresponding rescue distances are: d0, d1, ..., d n .

[0112] It should be noted that d is a range value, for example, 100 meters to 500 meters.

[0113] In a possible embodiment, the specific time period T is 8:00-9:00 on December 23, 2023.

[0114] Therefore, according to the above definition, according to the feature set of fire station S in a specific period T The fire station S is divided into two groups in each historical period T0~T m Different distance ranges d0,d1,...,d n All features are represented as the following n×m matrix A n×m :

[0115]

[0116] Here, T0 is defined as a time point in the period from 0:00 to 1:00 on any natural day.

[0117] Then, for all fire station sets S, all time period sets T, and all distance ranges d0, d1, ..., d n The travel time can establish a s×n×m tensor feature library as the fire rescue travel time data set, which is the data basis of the deep learning model of this application.

[0118] Step 102, fill the fire rescue travel time data set according to historical fire rescue data information, and use similar historical data between fire stations and social vehicle travel time information to expand the fire rescue travel time data set to obtain a training sample set.

[0119] Furthermore, in an embodiment of the present application, the tensor feature library obtained in step 102 is filled with historical fire rescue data information.

[0120] In the embodiment of the present application, firstly, historical fire rescue data information {I0, I1, ..., I n}, I refers to a recorded rescue event; then according to the historical fire rescue data information {I0,I1,...,I n}, get the latitude and longitude of the starting point of each rescue event I Rescue point latitude and longitude Police dispatch time (e.g. October 23, 2023 12:10), arrival time (such as 12:20 on October 23, 2023), and based on the latitude and longitude of the departure point Rescue point latitude and longitude Police dispatch time Arrival time Calculate the travel time t for each rescue event I I and the travel time t I Corresponding travel distance d I ; Finally, according to the police dispatch time The time period and the distance travelled I Corresponding distance range, the travel time t I Fill in the tensor feature library constructed in step 101.

[0121] Among them, the travel time t of each rescue event I I and the travel time t I Corresponding travel distance d I The calculation expression is:

[0122] The expression is:

[0123]

[0124] Where, d eu and d ma The Euclidean distance and the Manhattan distance are calculated based on the longitude and latitude of the departure point and the longitude and latitude of the rescue point, respectively. The calculation formulas of these two distances have been made public and are conventional technical contents in this field, and this application does not describe or limit them.

[0125] In a possible embodiment, historical fire rescue data information {I0, I1, ..., I n}.

[0126] However, due to the limited number of fire truck dispatches, each fire station only dispatches 1-2 times a day, resulting in a very small scale of historical sample data. Based on the historical dispatch data, it is impossible to effectively fill the tensor feature library, so the tensor feature library needs to be expanded.

[0127] In an embodiment of the present application, the historical data of the fire truck itself and the GPS data of social vehicles (such as taxis, online car-hailing vehicles, and private cars, and the data must include the instantaneous speed data of each sample) are further used to fill in the missing features in the tensor feature library, as explained below.

[0128] (1) Using the fire truck historical data to expand the small sample

[0129] In some embodiments of the present application, the process of using the fire truck historical data to expand the fire rescue travel time data set is based on the basic idea of ​​first measuring the similarity of historical data between different fire stations, and then using the data of fire stations with close similarity to supplement the vacancies of fire stations that lack rescue data. Specifically, the following steps are included: for a given fire station S i and S j , according to their respective historical feature datasets and Calculate the historical data similarity between the two fire stations; and expand the fire rescue travel time data set according to the historical data similarity between the two fire stations.

[0130] In the embodiment of the present application, for a given fire station S i and S j , according to their respective historical feature datasets and The steps to calculate the similarity of historical data between two fire stations are:

[0131] Historical feature dataset and Respectively reduced to and That is, the data of m time periods in the historical feature data are reduced to 24 time periods. These 24 time periods are 0:00~1:00, 1:00~2:00, 2:00~3:00,…, 22:00~23:00, 23:00~24:00. The formula for the travel time value of time period 0~1 is as follows:

[0132]

[0133] That is, the travel time value in the period 0 to 1 is equal to the average value of the period 0:00 to 1:00 of each natural day in the historical period, and n day for The number of non-zero values ​​in the set, the remaining 23 periods The calculation method is similar to this, and by analogy, we can get the fire station S i and S j Feature matrix after dimensionality reduction and

[0134] Then, according to the feature matrix after dimensionality reduction and Calculate the similarity of historical data between two fire stations The expression is:

[0135]

[0136] Among them, L(S i ,S j )、L′(S i ,S j ) and L″(S i ,S j ) are calculated as follows:

[0137]

[0138] in, is a matrix The average of all non-zero values ​​in is a matrix The average of all non-zero values ​​in ;

[0139]

[0140] Among them, σ Si is a matrix The standard deviation of all non-zero values ​​in , σ Sj is a matrix The standard deviation of all non-zero values ​​in ;

[0141]

[0142] where ρ(·) is the Pearson correlation coefficient between two vectors, is a matrix Vector data flattened by rows and columns, is a matrix Vector data flattened by rows and columns.

[0143] It should be noted that according to the above steps, the similarity between any two fire stations can be obtained Determine if It is considered that the two fire stations are very similar (κ is the similarity threshold, usually 0.8, the larger the value, the greater the similarity between the two), and the data are shared and supplemented with each other in the following way:

[0144] S1. Traverse the fire station set S and take out any fire station S i and its characteristic matrix Mark as visited, if there is no fire station that has not been visited, go to step S4;

[0145] S2. If there are other fire stations S j With Fire Station S i The similarity satisfies If κ is the similarity threshold, go to step S3, otherwise go to step S1;

[0146] S3. Reserved with Fire Station S i The fire station with the highest similarity j , and to the fire station S i The characteristic matrix To supplement, the expression is:

[0147]

[0148] S4. The loop ends and the characteristic matrix of all fire stations is output

[0149] Finally, according to the characteristic matrix of all fire stations Expand the tensor database.

[0150] (2) Using social vehicle travel time information to expand small samples

[0151] 1) Collect the GPS positioning point set D of all social vehicles in the same city.

[0152] 2) In the GPS positioning point set D, select all GPS positioning point sets D that belong to the same administrative division as the fire station S. S .

[0153] 3) Gather at GPS location point D S , filter all GPS positioning points in a specific period T

[0154] 4) According to the GPS positioning point collection Calculate the travel time corresponding to different distance ranges of fire station S during this period T The expression is:

[0155]

[0156] in, is a range of different distances d0,d1,...,d n The median value of (e.g., in the range of 100 to 500 meters, the median value is 300 meters), Is a collection The average instantaneous velocity of all the positioning points in .

[0157] 5) If there is no historical data of fire trucks that meets the calculation requirements in the tensor feature library Will Fill in the corresponding position in the tensor feature library. Among them, is the corresponding attenuation factor.

[0158] 6) Repeat the above steps 2)-5) until all elements in the fire station set S and all elements in the time period set T are traversed.

[0159] Step 103: construct a fire truck travel time prediction model based on multi-graph convolution, and train the fire truck travel time prediction model according to the training samples.

[0160] In the embodiment of the present application, a graph structure model is constructed with fire stations as graph nodes and the relationships between stations as edges, and feature mining is performed based on a graph convolutional neural network model to establish a mapping between historical travel time and future travel time.

[0161] Specifically, the output of the confirmation model is: all fire stations S = {S0, S1, ..., S n In the forecast period T m+1 Travel time τ at a specific rescue distance d (out) , a total of s nodes.

[0162] Confirm that the input to the model is:

[0163] τ (0)

[0164] Among them, τ (0) is a s×(k+h)-dimensional matrix, τ (0) Includes: the first k periods T m-k+1 ,...,T m-1 ,T m , all fire stations S={S0,S1,...,S n} The travel time at a specific rescue distance d, i.e., an s×k dimensional matrix; and the prediction period T m+1 Same first h natural day period All fire stations S={S0,S1,...,S n}The travel time at a specific rescue distance d, that is, an s×h dimensional matrix.

[0165] Confirm that the first graph convolution output of the model is:

[0166]

[0167] Among them, τ (1) It is still an s×(k+h)-dimensional matrix. A1 is an s×s-dimensional matrix, representing the adjacency relationship between fire stations. i and jThe spatial distance of is less than or equal to the preset spatial distance δ, then in the adjacency matrix A1 a ij =1, otherwise it is 0; I is the identity matrix, is the degree matrix, σ() is the activation function, θ (0) These are all the parameters involved in the first layer of graph convolution.

[0168] After the first graph convolution, the output τ is obtained (1) , which is still a s×(k+h)-dimensional matrix.

[0169] Confirm that the second graph convolution output of the model is:

[0170]

[0171] Among them, τ (2) It is still an s×(k+h)-dimensional matrix, A2 is an s×s-dimensional matrix, representing the correlation between fire stations, θ (1) are all the parameters involved in the second layer of graph convolution, at any position a in the matrix A2 ij The values ​​are:

[0172]

[0173] Wherein, cor() is the Pearson correlation coefficient of two vectors. The correlation coefficient calculation formula has been made public and is a conventional technical content in this field, and this application does not describe or limit it.

[0174] After the second graph convolution, the output τ is obtained (2) , which is still a s×(k+h)-dimensional matrix.

[0175] Determine the fully connected layer of the model, which is used to establish τ (2) and τ (out) The fully connected mapping is established.

[0176] It should be noted that in order to respond to fire needs in a timely manner, model training should be carried out on a rolling basis during actual application.

[0177] In the embodiment of the present application, after the fire truck travel time prediction model is established, the training sample set in step 102 is continuously calculated, and a new round of training can be carried out when the time period is switched, such as entering the time period T m After that, use T m The fire truck travel time prediction model is trained by using the training sample set of all previous time periods until the model converges, and a trained fire truck travel time prediction model is obtained, wherein each distance range corresponds to a model, and a total of n trained fire truck travel time prediction models are obtained.

[0178] Step 104, predicting the travel time of a fire truck based on the trained fire truck travel time prediction model.

[0179] In the embodiment of the present application, the travel time data τ of the next period is calculated respectively according to the n trained fire truck travel time prediction models. m+1 , that is, get each fire station in T m+1 During the time period, different distance ranges d0, d1, ..., d n The travel time prediction matrix is ​​as follows:

[0180]

[0181] When in T m+1 When actual demand points (such as fire points) are generated during a period of time, the distance between the fire station and the demand point can be automatically calculated, and then the travel time corresponding to the corresponding distance is selected from the travel time prediction matrix diagram as the prediction result, thereby realizing the prediction of the travel time from the fire truck to the demand point.

[0182] In order to implement the above-mentioned embodiments, the present application also proposes a fire truck travel time prediction device with multi-dimensional expansion of small sample data driven by deep learning. Figure 2 A schematic diagram of the structure of a fire truck travel time prediction device 10 for multi-dimensional expansion of small sample data driven by deep learning provided in an embodiment of the present application. Figure 2 As shown, the device comprises:

[0183] The travel time data set building module 100 is used to build a fire rescue travel time data set by combining the fire station information and the historical rescue travel time information;

[0184] An expansion module 200 is used to fill the fire rescue travel time data set according to historical fire rescue data information, and expand the fire rescue travel time data set by using similar historical data between fire stations and social vehicle travel time information to obtain a training sample set;

[0185] A model building and training module 300 is used to build a fire truck travel time prediction model based on multi-graph convolution, and train the fire truck travel time prediction model according to training samples;

[0186] The prediction module 400 is used to predict the travel time of a fire truck based on the trained fire truck travel time prediction model.

[0187] Regarding the device in the above embodiment, the specific manner in which each module performs operations has been described in detail in the embodiment of the method, and will not be elaborated here.

[0188] In order to implement the above embodiments, the present application also proposes an electronic device, comprising: a processor, and a memory communicatively connected to the processor; the memory stores computer-executable instructions; the processor executes the computer-executable instructions stored in the memory to implement the method provided by the above embodiments.

[0189] In order to implement the above embodiments, the present application also proposes a computer-readable storage medium, in which computer-executable instructions are stored. When the computer-executable instructions are executed by a processor, they are used to implement the methods provided by the above embodiments.

[0190] In order to implement the above embodiments, the present application also proposes a computer program product, including a computer program, which implements the methods provided by the above embodiments when executed by a processor.

[0191] The collection, storage, use, processing, transmission, provision and disclosure of user personal information involved in this application are in compliance with relevant laws and regulations and do not violate public order and good morals.

[0192] It should be noted that personal information from users should be collected for legitimate and reasonable purposes and should not be shared or sold outside of these legitimate uses. In addition, such collection / sharing should be carried out after receiving the user's informed consent, including but not limited to notifying the user to read the user agreement / user notice and sign the agreement / authorization including authorization of relevant user information before the user uses the function. In addition, any necessary steps should be taken to protect and safeguard access to such personal information data and ensure that others who have access to personal information data comply with its privacy policy and procedures.

[0193] The present application is expected to provide an implementation scheme for users to selectively block the use or access of personal information data. That is, the present disclosure is expected to provide hardware and / or software to prevent or block access to such personal information data. Once the personal information data is no longer needed, the risk can be minimized by limiting data collection and deleting the data. In addition, when applicable, such personal information is de-identified to protect the privacy of the user.

[0194] In the description of the aforementioned embodiments, the description with reference to the terms "one embodiment", "some embodiments", "example", "specific example", or "some examples" etc. means that the specific features, structures, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present application. In this specification, the schematic representations of the above terms do not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described may be combined in any one or more embodiments or examples in a suitable manner. In addition, those skilled in the art may combine and combine the different embodiments or examples described in this specification and the features of the different embodiments or examples, without contradiction.

[0195] In addition, the terms "first" and "second" are used for descriptive purposes only and should not be understood as indicating or implying relative importance or implicitly indicating the number of the indicated technical features. Therefore, the features defined as "first" and "second" may explicitly or implicitly include at least one of the features. In the description of this application, the meaning of "plurality" is at least two, such as two, three, etc., unless otherwise clearly and specifically defined.

[0196] Any process or method description in a flowchart or otherwise described herein may be understood to represent a module, fragment or portion of code comprising one or more executable instructions for implementing the steps of a custom logical function or process, and the scope of the preferred embodiments of the present application includes alternative implementations in which functions may not be performed in the order shown or discussed, including performing functions in a substantially simultaneous manner or in the reverse order depending on the functions involved, which should be understood by technicians in the technical field to which the embodiments of the present application belong.

[0197] The logic and / or steps represented in the flowchart or otherwise described herein, for example, can be considered as an ordered list of executable instructions for implementing logical functions, and can be embodied in any computer-readable medium for use by an instruction execution system, device or apparatus (such as a computer-based system, a system including a processor, or other system that can fetch instructions from an instruction execution system, device or apparatus and execute instructions), or in combination with these instruction execution systems, devices or apparatuses. For the purpose of this specification, "computer-readable medium" can be any device that can contain, store, communicate, propagate or transmit a program for use by an instruction execution system, device or apparatus, or in combination with these instruction execution systems, devices or apparatuses. More specific examples of computer-readable media (a non-exhaustive list) include the following: an electrical connection with one or more wires (electronic device), a portable computer disk box (magnetic device), a random access memory (RAM), a read-only memory (ROM), an erasable and programmable read-only memory (EPROM or flash memory), a fiber optic device, and a portable compact disk read-only memory (CDROM). In addition, the computer-readable medium may even be paper or other suitable medium on which the program is printed, since the program may be obtained electronically, for example, by optically scanning the paper or other medium and then editing, interpreting or processing in other suitable ways if necessary, and then stored in a computer memory.

[0198] It should be understood that the various parts of the present application can be implemented by hardware, software, firmware or a combination thereof. In the above-mentioned embodiments, multiple steps or methods can be implemented by software or firmware stored in a memory and executed by a suitable instruction execution system. For example, if implemented by hardware, as in another embodiment, it can be implemented by any one of the following technologies known in the art or their combination: a discrete logic circuit having a logic gate circuit for implementing a logic function for a data signal, a dedicated integrated circuit having a suitable combination of logic gate circuits, a programmable gate array (PGA), a field programmable gate array (FPGA), etc.

[0199] A person skilled in the art may understand that all or part of the steps in the method for implementing the above-mentioned embodiment may be completed by instructing related hardware through a program, and the program may be stored in a computer-readable storage medium, which, when executed, includes one or a combination of the steps of the method embodiment.

[0200] In addition, each functional unit in each embodiment of the present application may be integrated into a processing module, or each unit may exist physically separately, or two or more units may be integrated into one module. The above-mentioned integrated module may be implemented in the form of hardware or in the form of a software functional module. If the integrated module is implemented in the form of a software functional module and sold or used as an independent product, it may also be stored in a computer-readable storage medium.

[0201] The storage medium mentioned above may be a read-only memory, a magnetic disk or an optical disk, etc. Although the embodiments of the present application have been shown and described above, it can be understood that the above embodiments are exemplary and cannot be understood as limiting the present application. A person of ordinary skill in the art may change, modify, replace and modify the above embodiments within the scope of the present application.

Claims

1. A fire truck travel time prediction method based on multi-dimensional expansion of small sample data driven by deep learning, characterized in that: The following steps are involved: Combine the fire station information and historical rescue travel time information to build a fire rescue travel time data set; The fire rescue travel time data set is filled according to historical fire rescue data information, and the fire rescue travel time data set is expanded by using similar historical data between fire stations and social vehicle travel time information to obtain a training sample set; Constructing a fire truck travel time prediction model based on multi-graph convolution, and training the fire truck travel time prediction model according to the training samples; The fire truck travel time is predicted based on the trained fire truck travel time prediction model.

2. The method according to claim 1, characterized in that The fire station information and historical rescue travel time information are combined to construct a fire rescue travel time data set, including: In a specific city, the set S of all fire stations that can dispatch fire trucks is defined as: S={S0,S1,...,S n } In the formula, S represents the fire station, and the subscript is the number of each fire station; For each fire station S, define its feature set in a specific period T for: in, represents the travel time from the fire station S to the rescue locations at different distances, and the corresponding rescue distances are: d0, d1, ..., d n ; According to the feature set of fire station S in a specific period of time T The fire station S is divided into two groups in each historical period T0~T m Different distance ranges d0, d1, ..., d n All features are represented as the following n×m matrix A n×m : Among them, T0 is limited to a time point in the period of 0:00 to 1:00 on any natural day; For all fire station sets S, all time period sets T and all distance ranges d0,d1,...,d n A tensor feature library of s×n×m is established as the fire rescue travel time data set.

3. The method according to claim 2, characterized in that The filling of the fire rescue travel time data set according to the historical fire rescue data information includes: Get historical fire rescue data information {I0,I1,...,I n }, I refers to a recorded rescue event; According to the historical fire rescue data information {I0, I1, ..., I n }, get the latitude and longitude of the starting point of each rescue event I Rescue point latitude and longitude Police dispatch time Arrival time According to the latitude and longitude of the starting point Rescue point latitude and longitude Police dispatch time Arrival time Calculate the travel time t for each rescue event I I and travel distance d I , the expression is: Among them, d eu and d ma They are the Euclidean distance and Manhattan distance calculated based on the longitude and latitude of the departure point and the longitude and latitude of the rescue point respectively; According to the dispatch time The time period and the distance travelled I Corresponding distance range, the travel time t I Fill in the tensor feature library.

4. The method according to claim 3, characterized in that The process of expanding the fire rescue travel time data set by using similar historical data between fire stations includes: For a given fire station S i and S j , according to their respective historical feature datasets and Calculate the similarity of historical data between two fire stations; According to the similarity of historical data between two fire stations, the fire rescue travel time data set is expanded.

5. The method according to claim 3, characterized in that: For a given fire station S i and S j , according to their respective historical feature datasets and Calculate the similarity of historical data between two fire stations, including: Historical feature dataset and Respectively reduced to and That is, the data of m time periods in the historical feature data is reduced to 24 time periods, and the formula for the travel time value of time period 0 to 1 is as follows: That is, the travel time value in the period 0 to 1 is equal to the average value of the period 0:00 to 1:00 of each natural day in the historical period, and n day for The number of non-zero values ​​in the set, the remaining 23 periods The calculation method is similar to this, and by analogy, we can get the fire station S i and S j Feature matrix after dimensionality reduction and According to the feature matrix after dimensionality reduction and Calculate the similarity of historical data between two fire stations The expression is: Among them, L(S i ,S j )、L′(S i ,S j ) and L″(S i ,S j ) are calculated as follows: in, is a matrix The average of all non-zero values ​​in is a matrix The average of all non-zero values ​​in ; Among them, σ Si is a matrix The standard deviation of all non-zero values ​​in , σ Sj is a matrix The standard deviation of all non-zero values ​​in ; where ρ(·) is the Pearson correlation coefficient between two vectors, is a matrix Vector data flattened by rows and columns, is a matrix Vector data flattened by rows and columns.

6. The method according to claim 5, characterized in that The method of expanding the fire rescue travel time data set according to the similarity of historical data between the two fire stations includes: S1. Traverse the fire station set S and take out any fire station S i and its characteristic matrix Mark as visited, if there is no fire station that has not been visited, go to step S4; S2. If there are other fire stations S j With Fire Station S i The similarity satisfies If κ is the similarity threshold, go to step S3, otherwise go to step S1; S3. Reserved with Fire Station S i The fire station with the highest similarity j , and to the fire station S i The characteristic matrix To supplement, the expression is: S4. The loop ends and the characteristic matrix of all fire stations is output 7. The method according to claim 6, characterized in that The process of expanding the fire rescue travel time data set by using social vehicle travel time information includes: Collect the GPS positioning point set D of all social vehicles in the same city; In the GPS positioning point set D, select all GPS positioning point sets D that belong to the same administrative division as the fire station S. S ; Meet at GPS location D S , filter all GPS positioning points in a specific period T According to GPS positioning point collection Calculate the travel time corresponding to different distance ranges of fire station S during this period T The expression is: Where d is the distance range of different distances d0,d1,...,d n The median value of Is a collection The average instantaneous velocity of all the positioning points in; If there is no historical data of fire trucks that meets the calculation requirements in the tensor feature library Will Fill in the corresponding position in the tensor feature library, where is the corresponding attenuation factor; The above screening step, calculation step, and filling step are repeated until all elements in the fire station set S and all elements in the time period set T are traversed.

8. The method according to claim 7, characterized in that The method of constructing a fire truck travel time prediction model based on multi-graph convolution includes: The structure of the model is determined as follows: a graph structure model is constructed with fire stations as graph nodes and the relationships between stations as edges; The output of the confirmation model is: All fire stations S = {S0, S1, ..., S n In the forecast period Tm +1 Travel time τ at a specific rescue distance d (out) , a total of s nodes; Confirm that the input to the model is: t (0) Among them, the τ (0) is a s×(k+h)-dimensional matrix, τ (0) Includes: the first k periods T m-k+1 ,...,T m-1 ,T m , all fire stations S={S0,S1,...,S n } The travel time at a specific rescue distance d, i.e., an s×k dimensional matrix; and the prediction period T m+1 Same first h natural day period All fire stations S={S0,S1,...,S n } The travel time at a specific rescue distance d, i.e., an s×h dimensional matrix; Confirm that the first graph convolution output of the model is: Among them, the τ (1) It is still an s×(k+h)-dimensional matrix. A1 is an s×s-dimensional matrix, representing the adjacency relationship between fire stations. i and j The spatial distance of is less than or equal to the preset spatial distance δ, then in the adjacency matrix A1 a ij =1, otherwise it is 0; I is the identity matrix, is the degree matrix, σ() is the activation function, θ (0) are all the parameters involved in the first layer of graph convolution; Confirm that the second graph convolution output of the model is: Among them, the τ( 2 ) is still an s×(k+h)-dimensional matrix, A2 is an s×s-dimensional matrix, representing the correlation between the fire stations, θ (1) are all the parameters involved in the second layer of graph convolution, at any position a in the matrix A2 ij The values ​​are: Where cor() is the Pearson correlation coefficient of two vectors; Determine the fully connected layer of the model, which is used to establish τ (2) and τ (out) The fully connected mapping is established.

9. The method according to claim 8, characterized in that The training of the fire truck travel time prediction model according to the training samples includes: The training sample set is calculated continuously. When entering the period T m After that, use T m The fire truck travel time prediction model is trained using the training sample set of all previous time periods until the model converges, thereby obtaining a trained fire truck travel time prediction model, wherein each distance range corresponds to a model, and a total of n trained fire truck travel time prediction models are obtained.

10. The method according to claim 9, characterized in that The method of realizing the prediction of the fire truck travel time according to the trained fire truck travel time prediction model includes: According to the n trained fire truck travel time prediction models, the travel time data τ of the next period is calculated respectively. m+1 , get each fire station in T m+1 During the time period, different distance ranges d0, d1, ..., d n The travel time, the prediction matrix of travel time is as follows: When in T m+1 The actual demand point is generated in the time period, the distance between the fire station and the demand point is automatically calculated, and the corresponding travel time is selected from the travel time prediction matrix diagram to realize the prediction of the travel time from the fire truck to the demand point.

Images