A risk index prediction method and system based on multi-source data analysis

Abstract

The present invention provides a risk index prediction method and system based on multi-source data analysis. A risk index model is established based on historical data to derive the risk label value of the operating data. Then, a machine learning algorithm is used to predict the risk index based on real-time operating data. Finally, the display and application of the risk index are realized, and real-time risk index prediction based on multi-source data analysis is realized. This provides data support for traffic systems with complex and variable factors, and forms an intelligent early warning mechanism.

Description

Technical Field

[0001] The present invention relates to the field of data analysis, and in particular to a risk index prediction method and system based on multi-source data analysis. Background Art

[0002] With the development of society, people's demand for the security of various transportation systems is increasing. Especially for large airports, the continuous increase in airport passenger traffic has also led to a growing demand for airport transportation security. However, since subways, long-distance buses, and public buses all stop operating at night, the sharp decline in airport transportation capacity at night, combined with the peak of night flight arrivals, has placed tremendous pressure on airport landside transportation security operations. There is also a lack of risk awareness of this pressure, resulting in a lack of data and technical support for airport arrival evacuation traffic information management, making the service supply and security of airport landside transportation very passive. Therefore, focusing on airport landside arrival evacuation traffic information management, with the goal of achieving optimal perception of airport landside arrival traffic information, comprehensively collecting dynamic data related to air and land transportation, forming a multi-dimensional data collection and correlation model analysis, and providing more accurate airport transportation security risk prediction under the interweaving of complex and changing factors, is a very important and meaningful research topic.

[0003] In view of this, it is necessary to improve the airport traffic security risk prediction method and system in the existing technology to solve the above problems. Summary of the Invention

[0004] The purpose of this invention is to achieve real-time risk index prediction based on multi-source data analysis, provide data support for traffic systems with complex and variable factors, and form an intelligent early warning mechanism.

[0005] To achieve the above objectives, the present invention provides a risk index prediction method based on multi-source data analysis, comprising the following steps:

[0006] Step 1. Based on data analysis and feature construction methods, a risk index model is established using historical operating data to obtain historical data with risk label values;

[0007] Step 2. Establish a risk index prediction model using the historical data with risk label values ​​based on a machine learning algorithm;

[0008] Step 3. Based on the real-time operation data, the risk index prediction model is used to realize real-time risk index prediction.

[0009] Furthermore, the steps of establishing the risk index model include:

[0010] 1) Analyzing the historical data to obtain a preliminary risk indicator value;

[0011] 2) performing feature construction on the historical data to obtain feature control values;

[0012] 3) Combining the preliminary risk indicator value and the characteristic control value to obtain the historical data with the risk label value.

[0013] Furthermore, the data analysis method may include but is not limited to data cleaning, logical operations between data, and data normalization.

[0014] Furthermore, the characteristic control value is set with a characteristic control coefficient; the characteristic control coefficient can be an interval range or a fixed value.

[0015] Furthermore, the specific method steps for establishing the risk index prediction model are:

[0016] Step 1: Divide the historical operation data with risk label values ​​into a training data set and a test data set;

[0017] Step 2: The risk index prediction model is obtained by training the LightGBM algorithm on the training data set.

[0018] Step 3: Further optimize the model based on the validation loss to obtain a better model;

[0019] Step 4: Based on the better model, risk index prediction is implemented on the test data set to obtain the risk prediction result output.

[0020] Furthermore, the evaluation index parameters for training and verification are both RMSE; Step 3 and Step 4 are performed in a loop until relatively ideal model parameters are obtained; and the better model has the relatively ideal model parameters.

[0021] The present invention also provides a risk prediction system for the above-mentioned risk index prediction method, comprising a data module, a prediction model establishment module, and a real-time application module; the data module obtains data with risk label values ​​by collecting, analyzing, and processing data;

[0022] The prediction model establishment module establishes a risk index prediction model using the data with risk label values ​​based on a machine learning algorithm;

[0023] The real-time application module outputs a real-time risk index through the risk index prediction model based on real-time data and displays it through the platform.

[0024] Furthermore, the platform display can set different emergency warning plans according to different risk indices.

[0025] Compared with the existing technology, the beneficial effects of the present invention are: establishing a risk index model based on historical data, obtaining the risk label value of the operation data, and then realizing the risk index prediction based on real-time operation data through machine learning algorithms, and finally realizing the display and application of the risk index, thereby adjusting the various characteristics affecting the risk index, eliminating and reducing the operation guarantee risk, making the risk evenly distributed within a certain period of time, and reducing the fluctuation of the risk index. BRIEF DESCRIPTION OF THE DRAWINGS

[0026] Figure 1 This is a schematic diagram of the steps of a risk index prediction method based on multi-source data analysis of the present invention;

[0027] Figure 2 A schematic diagram of the specific method steps for establishing the risk index model;

[0028] Figure 3 A schematic diagram of the specific method steps for establishing a risk index prediction model;

[0029] Figure 4 A logical flow chart for the risk index prediction model;

[0030] Figure 5 This is a flow chart of a risk prediction system based on a risk index prediction method according to the present invention. DETAILED DESCRIPTION

[0031] The present invention is described in detail below with reference to the various embodiments shown in the accompanying drawings, but it should be noted that these embodiments are not limitations of the present invention, and any equivalent transformations or substitutions in functions, methods, or structures made by ordinary technicians in this field based on these embodiments are all within the scope of protection of the present invention.

[0032] Please refer to Figure 1 As shown, the present invention provides a risk index prediction method based on multi-source data analysis, which specifically includes the following steps:

[0033] Step 1. Based on the method of data analysis and feature construction, a risk index model is established using historical operation data to obtain each piece of historical operation data with a risk label value.

[0034] Specifically, such as Figure 2 As shown in Figure 2, the specific method for establishing the risk index model is:

[0035] 1. Analyze historical data to obtain preliminary risk indicators. Specifically, data analysis methods may include, but are not limited to, data cleaning, logical operations between data, and data normalization.

[0036] 2. Construct features from historical data to obtain feature control values. Specifically, feature construction requires determining specific control features based on the characteristics of the data itself and the degree to which the specificity of the actual scenario affects the final risk label value. For example, for flight delay risk prediction, in addition to direct data predictions such as flight speed and flight distance, it is also necessary to consider the air traffic control conditions at the arrival airport. Therefore, air traffic control conditions need to be identified as one of the control features for the flight delay risk prediction.

[0037] 3. Combine the preliminary risk index value and the characteristic control value to obtain historical operating data with a risk label value. Specifically, to limit the impact of the characteristic control value on the preliminary risk index value to a certain range and avoid incorporating too many extreme scenarios that would make the final risk label value lose its discriminatory power, in actual applications, an adjustable characteristic control coefficient is set for the characteristic control value. This characteristic control coefficient can be a range or a fixed value that is ultimately determined to be most favorable for the application scenario through a certain number of scenario simulations.

[0038] Step 2. Establish a risk index prediction model based on machine learning algorithms using historical operating data with risk label values.

[0039] Specifically, such as Figure 3-Figure 4 As shown in Figure 2, the specific method for establishing the risk index prediction model is as follows:

[0040] 1. Divide historical operational data with risk labels into training and testing datasets. Specifically, divide the older data into the training dataset, and the later data into the testing dataset. This will better align with the actual operational patterns of the model.

[0041] 2. The risk index prediction model is obtained by training the LightGBM algorithm on the training data set. Specifically, the preliminary model parameters of the risk index prediction model can be obtained through LightGBM algorithm training.

[0042] 3. Combined with the validation loss, the model is further optimized to obtain a better model. In particular, the evaluation indicator parameters for training and validation are both RMSE. Figure 4 As shown in the figure, after the data of the training dataset is trained by the LightGBM algorithm and the loss is verified, the model is optimized. Then the data can be cycled through the LightGBM algorithm again for training and verification of the loss, and the model is optimized again until the model parameters with ideal prediction effect are obtained, which is the final better model.

[0043] 4. Implement risk index prediction based on the optimal model on the test dataset and obtain a risk prediction output. Specifically, since the test dataset is still historical data, after obtaining the risk prediction results, the quality of the predictions is evaluated and whether further optimization is needed. Optimization methods include, but are not limited to, repartitioning the training and test data.

[0044] Step 3. Based on the real-time operation data, a risk index prediction model is used to achieve real-time risk index prediction. Specifically, the risk index prediction model is also a LightGBM regression prediction model, which is a regression prediction model based on the LightGBM algorithm applied to real-time operation data.

[0045] like Figure 5 As shown, the present invention also proposes a risk prediction system based on the above-mentioned risk index prediction method: comprising a data module, a prediction model building module, and a real-time application module. The details are as follows:

[0046] 1. The data module collects, analyzes, and processes data to obtain data with risk-labeled values. Specifically, the data module includes a feature extraction module, a data acquisition module, and a risk index modeling module. The feature extraction module determines the features required to establish the risk index prediction model. The data acquisition module collects data based on the extracted features, including historical data and real-time data. The risk index modeling module analyzes and constructs features on the historical data obtained by the data acquisition module to obtain historical data with risk-labeled values. The specific methods for establishing the risk index model have been described above and will not be repeated here.

[0047] The second prediction model building module uses machine learning algorithms to build a risk index prediction model using data with risk labels. The specific method for building the risk index prediction model has been described above and will not be repeated here.

[0048] The third module, the real-time application module, uses real-time data to generate a real-time risk index using a risk index prediction model and displays it on the platform. Specifically, the module applies the real-time data to the LightGBM regression prediction model to achieve real-time risk index prediction, and then displays it on the platform in a digital and graphical form using media. Furthermore, the platform can also set different emergency warning plans based on different risk indices to achieve more effective risk management.

[0049] The following uses the risk perception application of landside transportation at a large airport as an example to describe how this system analyzes and studies data related to landside transportation at a large airport, and how it predicts and displays the risks caused by various related activities and influencing factors in the supply and maintenance of landside transportation services, especially in extreme weather conditions and nighttime rush hours.

[0050] 1. The data processing module includes the following steps:

[0051] Step 1. Feature extraction module. Determine the features required to build the risk index prediction model:

[0052] 1) Station passenger flow statistics: Station entrance flow (t-1)

[0053] 2) Station passenger flow statistics: Station exit flow (t-1)

[0054] 3) Public parking lot traffic statistics: inbound traffic (t-1) - outbound traffic (t-1)

[0055] 4) Taxi parking lot traffic statistics: Exit traffic (t-1)

[0056] 5) Flight statistics: actual number (t)

[0057] 6) Weather data: precipitation (t)

[0058] In practical applications, the risk index prediction model uses the past hour's station entrance and exit traffic, bus entrance and exit traffic, taxi depot exit traffic, as well as the expected number of flight arrivals and weather conditions to predict the next hour's vehicle supply risk index. The past hour's data features are required for two reasons:

[0059] The first point is that the risk level of vehicle supply in the next hour is related to the number of passengers entering and leaving in the past hour, the number of stranded passengers, and the level of security of taxis, buses, and social vehicles.

[0060] Second point: When actually using the model for prediction, we only have historical data on characteristic quantities such as station entrance flow, exit flow, bus entrance flow-exit flow, and taxi parking lot exit flow. We can only use historical data to predict the intensity of passenger delays / waiting for the next hour.

[0061] Step 2. Data collection module, including the collection of historical data and real-time data: station passenger flow statistics, station passenger flow statistics, public parking building flow statistics, taxi parking lot flow statistics, flight statistics, weather data, etc.

[0062] Step 3. Risk index model establishment module, establish risk index models for the above categories of historical data respectively.

[0063] Specifically, the following uses station passenger flow data as an example to describe the method and steps for establishing a risk index model in this application:

[0064] 1. Obtain the exit flow (t) and entry flow (t) of the passenger flow statistical data of the station through the data acquisition device;

[0065] 2. Clean the statistical data: Abnormal data processing: 1) Delete the data with an entry flow of 0; 2) Delete the data with a negative exit flow or entry flow.

[0066] 3. Set the data analysis object D(t): D(t) = exit flow (t) / entry flow (t)

[0067] 4. Conduct data analysis on the data analysis object D(t), determine that the data D(t) is mainly distributed in [a, A], obtain that the maximum value of D(t) is A and the minimum value is a.

[0068] 5. Optimize the outlier data of the data analysis object D(t) to reduce the impact of outliers on the model:

[0069] 1) When D(t) > A, then assign the value of D(t) as A;

[0070] 2) When D(t) < a, then assign the value of D(t) as a.

[0071] 6. Conduct data normalization on the data analysis object D(t) to obtain the preliminary risk index value R0(t), such that R0(t) ∈ (0, 1). The specific normalization method is as follows:

[0072] At this point, normalize the value of D(t) to a value between 0 and 1, and obtain such a value R0(t), which is the preliminary risk index value. We can vividly call it the passenger departure rate, and use it to measure the degree to which passengers do not need to wait or do not need to stay for a long time during this hour period. However, the formula described is not yet precise enough. The reasons are as follows:

[0073] Case 1: Exit flow = 80, entry flow = 800

[0074] Case 2: Exit flow = 10, entry flow = 100

[0075] In the above two cases, D(t) is equal to 0.1 in both cases, but the passenger retention situation in Case 1 is more serious than that in Case 2. The reason for the severity is that the difference between the entry and exit flows is larger. Therefore, the exit flow (t) - entry flow (t) can also measure the degree to which passengers do not need to wait or do not need to stay for a long time, that is, measure the risk / shortage degree of vehicle supply guarantee during this hour period.

[0076] In view of the above situation, it is necessary to further deepen the processing of the preliminary risk index value R0(t) to improve the accuracy of measuring the risk degree.

[0077] 7. Perform risk superposition control on the preliminary risk index value R0(t):

[0078] 1) Set the data analysis object d(t): d(t) = outlet flow (t) - inlet flow (t)

[0079] 2) Perform data analysis on the data analysis object d(t) and determine that the data d(t) is mainly distributed in [b, B]. The maximum value of d(t) is B and the minimum value is b.

[0080] It is particularly important to note that the main purpose of determining the distribution range of the data d(t) is to eliminate data that is too large or too small (outliers). The data distribution range is set to +max and -max, so that the exit-entry value can be quantified to a value between 0 and 1 to measure the risk / scarcity of vehicle supply during that hour. If outliers are taken into account, the measurement indicators will be overly affected by outliers / abnormal situations, which will reduce the differentiation of vehicle supply risks in common situations and reduce the ability to characterize and measure the risk level of vehicle supply in common situations. Here, we can directly eliminate outliers, but we set values ​​greater than B to B and values ​​less than b to b in order to retain negative samples.

[0081] 3) Construct characteristic control value Δ(t):

[0082]

[0083] 4) The risk label value R(t) is obtained by superimposing the preliminary risk index value R0(t) and the characteristic control value Δ(t). The specific method is as follows:

[0084]

[0085] Where α is the feature control coefficient. In practical applications, R0(t)-αΔ(t) may inevitably overflow [0, 1]. In the case of overflow, we set it to 0 or 1. In addition, the feature control coefficient α controls the impact of Δ(t) on R(t). The feature control coefficient α cannot be too large or too small, because it will cause Δ(t) to overflow [0, 1] too much, resulting in too many positive and negative samples being set to 0 or 1, causing the risk label value R(t) to lose a certain degree of discrimination.

[0086] In particular, in this embodiment, we use a step size of 0.1 to traverse the interval from 0.2 to 0.8 for the feature control coefficient α, and combine the obtained risk label value distribution and the quality of the final prediction result of the model to determine the optimal α value of 0.4.

[0087] 2. Risk index prediction model establishment module: Based on the machine learning algorithm, the risk index prediction model is established using the historical operating data with risk label values ​​of the above features. Specifically, in this embodiment, a data output is predicted by the 6 input features extracted in step 1. This data output is the risk index obtained by the risk index prediction model based on the LightGBM regression prediction algorithm. The establishment of the risk index prediction model here follows the risk index prediction model establishment method described above, and the details will not be repeated one by one.

[0088] 3. Real-time Application Module: This module includes real-time risk index prediction output and platform display. Based on real-time data, the risk index prediction model generates a real-time risk index forecast for the airport's landside transportation service supply and support. Based on the real-time risk index, corresponding warning signals or emergency response plans are displayed on the platform. Specifically, staff can make timely dispatches based on the emergency response plans corresponding to different risk indices, adjusting the impact of different risk characteristics on the risk index to quickly restore the airport's landside transportation service supply and support.

[0089] In particular, in practical applications, after staff have implemented the corresponding emergency plan, they can verify the effectiveness of the emergency plan based on the real-time changes in the risk index, thereby achieving the purpose of verifying the disposal process and evaluating the implementation effect. Therefore, we can say that this risk index prediction method is not only a prediction model, but also a process verification and implementation evaluation model.

[0090] The series of detailed descriptions listed above are only specific descriptions of feasible implementation methods of the present invention. They are not intended to limit the scope of protection of the present invention. Any equivalent implementation methods or changes that do not deviate from the technical spirit of the present invention should be included in the scope of protection of the present invention.

[0091] It will be apparent to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above and that the invention can be embodied in other specific forms without departing from the spirit or essential characteristics of the invention. Therefore, the embodiments should be considered in all respects as illustrative and non-restrictive, and the scope of the invention is defined by the appended claims, not the foregoing description, and all variations within the meaning and range of equivalents of the claims are intended to be included therein. Any reference sign in a claim should not be construed as limiting the claim to which it relates.

[0092] In addition, it should be understood that although this specification is described in terms of implementation methods, not every implementation method contains only one independent technical solution. This narrative method of the specification is only for the sake of clarity. Those skilled in the art should regard the specification as a whole. The technical solutions in each embodiment can also be appropriately combined to form other implementation methods that can be understood by those skilled in the art.

Claims

1. A risk index prediction method based on multi-source data analysis, characterized by: The vehicle supply risk index R(t) for the next hour is predicted based on the station entrance flow (t-1), exit flow (t-1), bus entrance flow (t-1) minus exit flow (t-1), taxi parking lot exit flow (t-1), and the expected number of flight arrivals (t) and weather (t) for the next hour. Step 1. Based on data analysis and feature construction methods, a risk index model is established using historical operating data to obtain historical data with risk label values; Step 2. Establish a risk index prediction model using the historical data with risk label values ​​based on a machine learning algorithm; Step 3. Based on the real-time operation data, the risk index prediction model is used to realize real-time risk index prediction; The steps of establishing the risk index model include: 1) Analyzing the historical data to obtain a preliminary risk indicator value; 2) performing feature construction on the historical data to obtain feature control values; 3) Combining the preliminary risk indicator value and the characteristic control value to obtain the historical data with the risk label value; The preliminary risk index value R0(t) is obtained by performing data normalization processing on the data analysis object D(t) so that R0(t)∈(0,1); the specific normalization method is: The data analysis object D(t) = outlet flow(t) / inlet flow(t); D(t) is mainly distributed in [a, A], and the maximum value of D(t) is A and the minimum value is a; Construct characteristic control value Δ(t): The risk label value R(t) is obtained by superimposing the preliminary risk index value R0(t) and the characteristic control value Δ(t). The specific method is as follows: Where α is the characteristic control coefficient; the data analysis object d(t) = outlet flow(t) - inlet flow(t), the data d(t) is distributed in [b, B], and the maximum value of d(t) is B, and the minimum value is b.

2. The risk index prediction method according to claim 1, wherein: The data analysis method includes data cleaning, logical operations between data, and data normalization.

3. The risk index prediction method according to claim 1, wherein: The characteristic control value is set with a characteristic control coefficient; the characteristic control coefficient is an interval range or a fixed value.

4. The risk index prediction method according to claim 1, wherein: The specific method steps for establishing the risk index prediction model are: Step 1: Divide the historical operation data with risk label values ​​into a training data set and a test data set; Step 2: The risk index prediction model is obtained by training the LightGBM algorithm on the training data set. Step 3: Further optimize the model based on the validation loss to obtain a better model; Step 4: Based on the better model, risk index prediction is implemented on the test data set to obtain the risk prediction result output.

5. The risk index prediction method according to claim 4, wherein: The evaluation index parameters for training and verification are both RMSE; Step 3 and Step 4 are repeated until relatively ideal model parameters are obtained; and the preferred model has the relatively ideal model parameters.

6. A risk prediction system based on the risk index prediction method according to any one of claims 1 to 5, characterized in that: It includes a data module, a prediction model building module, and a real-time application module; the data module obtains data with risk label values ​​by collecting, analyzing, and processing data; The prediction model establishment module establishes a risk index prediction model using the data with risk label values ​​based on a machine learning algorithm; The real-time application module outputs a real-time risk index through the risk index prediction model based on real-time data and displays it through the platform.

7. The risk index prediction system according to claim 6, wherein: The platform displays that different emergency warning plans can be set according to different risk indices.

Images