A feature extraction method for driving data based on topological analysis

Abstract

The present invention provides a method for extracting features of driving data based on topological analysis, comprising the following steps: S1: obtaining original operation and maintenance data samples; S2: preprocessing the original operation and maintenance data samples to obtain multidimensional data samples; S3: forming local sub-features of data in the multidimensional data samples; S4: obtaining optimal local features and optimal global features; S5: feature extraction is completed; The method for extracting features of driving data based on topological analysis proposed by the present invention extracts vehicle network data features by combining local and overall topological analysis means, obtains local optimal solutions and global optimal solutions in data features, eliminates a large number of invalid analyses, and improves the efficiency of algorithm analysis; reduces analysis costs, and avoids the problem of increasingly low benefit-to-cost ratio of offline analysis.

Description

Technical Field

[0001] The present invention relates to the field of vehicle networking data analysis, and in particular to a feature extraction method for driving data based on topological analysis. Background Art

[0002] In recent years, with the widespread use and penetration of mobile networks, automobiles have also rapidly entered the era of networking. Different from traditional Internet and mobile Internet based on mobile phones, the Internet of Vehicles contains a large amount of time-series collected data, as well as a lot of interactive data for enterprises, vehicles, people, life, entertainment, driving, supervision, etc.; these same representation data analysis and modeling dimensions are more, with both intersections and overlaps; with the rise of the concept of intelligent data platforms and data middle platforms, these multi-dimensional data need to be analyzed and integrated to support differentiated business needs such as fault analysis, driving assistance, and event warnings. In order to improve the computing and analysis speed and platform response efficiency, and reduce the time and space complexity, storage and computing costs of the analysis algorithm, it is necessary to first extract the eigenvalues ​​of these data and then substitute them into the algorithm for use.

[0003] The existing methods for extracting vehicle-mounted data features have the following drawbacks:

[0004] 1. Due to the complexity of multidimensional data, the existing algorithms themselves will have the problem of dimensionality explosion, causing the time and space complexity of the algorithm to increase exponentially with the number of data dimensions, and the performance and efficiency of the algorithm will drop sharply;

[0005] 2. The current methods for extracting global and local features of IoV data require different algorithms, and lack a relatively complete and unified means to characterize the internal connection between global and local features. Using traditional methods to find implicit patterns between IoV data has not made much progress. In addition, the cost of collecting data is very high, and the cost of ineffective analysis is even higher. With the continuous accumulation of data, the benefit-cost ratio of offline analysis is getting lower and lower; Summary of the invention

[0006] The purpose of the present invention is to solve the shortcomings of the prior art and to propose a feature extraction method for driving data based on topological analysis.

[0007] To achieve the above object, the present invention adopts the following technical solutions:

[0008] A method for extracting features from driving data based on topological analysis comprises the following steps:

[0009] S1: Obtain original operation and maintenance data samples;

[0010] Engineers obtain raw operation and maintenance data samples through the data platform API or database raw tables;

[0011] The original operation and maintenance data sample includes driving speed, fuel consumption, etc.;

[0012] S2: pre-process the original operation and maintenance data samples to obtain multi-dimensional data samples;

[0013] The original operation and maintenance data sample includes multiple dimensions;

[0014] Engineers select the target filtering function to pre-process the original operation and maintenance data samples to obtain filtered multi-dimensional data samples;

[0015] The target filter function includes a Mahalanobis distance function, etc.;

[0016] S3: forming local sub-features of data in multidimensional data samples;

[0017] The following sub-steps are included:

[0018] S31: Allocating data in the multidimensional data sample to different filtering spaces;

[0019] Engineers classify and cut multidimensional data samples by dimensional cutting, and classify data with consistent features in multidimensional data samples into the same filtering space;

[0020] Furthermore, the characteristics of the data in each filter space are consistent;

[0021] The features include numerical features, category features, etc.

[0022] After the classification is completed, the global features in the multidimensional data sample are extracted through the Mahalanobis distance function and recorded as global feature 1;

[0023] Furthermore, the filter space forms different feature spaces by selecting different calculation factors;

[0024] Aggregate the data of each filter space, adjust the calculation factor value, and extract the local sub-features of each filter space;

[0025] The local sub-features include respective data points;

[0026] S32: performing secondary aggregation on multiple local sub-features according to the distance dimension of the filtering space;

[0027] The following sub-steps are included:

[0028] S321: Building parallel relationship;

[0029] Compare the data points of each local sub-feature, and if there are local sub-features with the same data points, build a parallel relationship between the local sub-features with the same data points;

[0030] S322: performing secondary aggregation on local sub-features;

[0031] The local sub-features with parallel relationship are secondary aggregated through weighted average, convolution operation and other functions, and the local sub-features with the same data points are aggregated into the same feature space;

[0032] S33: Fitting multiple feature spaces into a new global feature;

[0033] Multiple feature spaces are fitted into a new global feature through a weighted average function, which is recorded as global feature 2;

[0034] S4: Get the optimal local features and the optimal global features;

[0035] The global features 1 and 2 are fitted by a fitting algorithm to form an optimal global feature;

[0036] Perform algorithm fitting on each local sub-feature obtained in step S31 to form an optimal local feature;

[0037] S5: Feature extraction completed;

[0038] After the optimal global features and the optimal local features are extracted, the extracted optimal global features and the optimal local features are output to the downstream algorithm for subsequent calculations;

[0039] Compared with the prior art, the present invention has the following beneficial effects:

[0040] The present invention proposes a method for extracting driving data features based on topological analysis, which obtains original operation and maintenance data samples of the Internet of Vehicles, selects a target filter function to preprocess the original operation and maintenance data samples, and obtains filtered multidimensional data samples; extracts local sub-features and global features of data in the multidimensional data samples, and performs secondary aggregation on the extracted multiple local sub-features, aggregates local sub-features with the same data points into the same feature space, fits the obtained multiple feature spaces into a new global feature, fits multiple local sub-features to obtain the optimal local feature, and fits two global features to obtain the optimal global feature;

[0041] The data features of the Internet of Vehicles are extracted by combining local and overall topological analysis methods, and the local optimal solution and global optimal solution in the data features are obtained, which eliminates a large number of invalid analyses and improves the efficiency of algorithm analysis.

[0042] The method provided by the present invention reduces the analysis cost and avoids the problem of the offline analysis benefit-cost ratio becoming increasingly low; BRIEF DESCRIPTION OF THE DRAWINGS

[0043] Figure 1A flowchart of the steps of a method for extracting features of driving data based on topological analysis according to the present invention; Implementation

[0044] In order to provide a further understanding of the purpose, structure, features, and functions of the present invention, the following detailed description is given in conjunction with the embodiments.

[0045] A method for extracting features from driving data based on topological analysis comprises the following steps:

[0046] S1: Obtain original operation and maintenance data samples;

[0047] Engineers obtain raw operation and maintenance data samples through the data platform API or database raw tables;

[0048] The original operation and maintenance data sample includes driving speed, fuel consumption, etc.;

[0049] S2: pre-process the original operation and maintenance data samples to obtain multi-dimensional data samples;

[0050] The original operation and maintenance data sample <driving speed, fuel consumption> includes multiple dimensions, such as time dimension, vehicle model dimension, location coordinate dimension, driver ID number dimension, etc.;

[0051] Select each dimension and target filter function in turn to pre-process the original operation and maintenance data samples to obtain filtered multi-dimensional data samples;

[0052] The target filter function includes a Mahalanobis distance function, etc.;

[0053] Furthermore, the way of selecting each dimension and objective function is random selection by the program or manual selection by the engineer;

[0054] S3: forming local sub-features of data in multidimensional data samples;

[0055] The following sub-steps are included:

[0056] S31: Allocating data in the multidimensional data sample to different filtering spaces;

[0057] Engineers classify and cut multidimensional data samples by dimensional cutting, and classify data with consistent features in multidimensional data samples into the same filtering space;

[0058] Furthermore, the characteristics of the data in each filter space are consistent;

[0059] The features include numerical features, category features, etc.

[0060] After the classification is completed, the global features in the multidimensional data sample are extracted through the Mahalanobis distance function and recorded as global feature 1;

[0061] Furthermore, the filter space forms different feature spaces by selecting different calculation factors;

[0062] Aggregate the data of each filter space, adjust the calculation factor value, and extract the local sub-features of each filter space;

[0063] By selecting the calculation factor value, the smoothness of each local sub-feature is eliminated, making each local sub-feature more obvious;

[0064] The local sub-features include respective data points;

[0065] S32: secondary aggregation of multiple local sub-features;

[0066] The following sub-steps are included:

[0067] S321: Building parallel relationship;

[0068] Compare the data points of each local sub-feature, and if there are local sub-features with the same data points, build a parallel relationship between the local sub-features with the same data points;

[0069] S322: performing secondary aggregation on local sub-features;

[0070] The local sub-features with parallel relationship are secondary aggregated through weighted average, convolution operation and other functions, and the local sub-features with the same data points are aggregated into the same feature space;

[0071] S33: Fitting multiple feature spaces into a new global feature;

[0072] Multiple feature spaces are fitted into a new global feature through a weighted average function, which is recorded as global feature 2;

[0073] By performing secondary aggregation on local sub-features, local sub-features are converted into feature data of composite dimensions based on a single dimension.

[0074] S4: Get the optimal local features and the optimal global features;

[0075] The global features 1 and 2 are fitted by a fitting algorithm to form an optimal global feature;

[0076] Perform algorithm fitting on each local sub-feature obtained in step S31 to form an optimal local feature;

[0077] S5: Feature extraction completed;

[0078] After the optimal global features and the optimal local features are extracted, the extracted optimal global features and the optimal local features are output to the downstream algorithm for subsequent calculations;

[0079] The present invention has been described by the above-mentioned relevant embodiments, however, the above-mentioned embodiments are only examples for implementing the present invention. It must be pointed out that the disclosed embodiments do not limit the scope of the present invention. On the contrary, changes and modifications made without departing from the spirit and scope of the present invention are all within the scope of patent protection of the present invention.

Claims

1. A feature extraction method for driving data based on topological analysis, characterized in that: The following steps are involved: S1: Obtain original operation and maintenance data samples; Engineers obtain raw operation and maintenance data samples through the data platform API or database raw tables; The original operation and maintenance data sample includes driving speed and fuel consumption; S2: pre-process the original operation and maintenance data samples to obtain multi-dimensional data samples; The original operation and maintenance data sample includes multiple dimensions; Engineers select the target filtering function to pre-process the original operation and maintenance data samples to obtain filtered multi-dimensional data samples; The target filter function includes a Mahalanobis distance function; S3: forming local sub-features of data in multidimensional data samples; Assign the data in the multidimensional data sample to different filter spaces, extract the local sub-features of each filter space, perform secondary aggregation on multiple local sub-features, aggregate them into multiple feature spaces, and fit multiple feature spaces into a new global feature; The following sub-steps are included: S31: Allocating data in the multidimensional data sample to different filtering spaces; Engineers classify and cut multidimensional data samples by dimensional cutting, and classify data with consistent features in multidimensional data samples into the same filtering space; The characteristics of the data in each filtering space are consistent; After the classification is completed, the global features in the multidimensional data sample are extracted through the Mahalanobis distance function; And extract local sub-features of each filter space; S32: secondary aggregation of multiple local sub-features; The following sub-steps are included: S321: Building parallel relationship; Compare the data points of each local sub-feature, and if there are local sub-features with the same data points, build a parallel relationship between the local sub-features with the same data points; S322: performing secondary aggregation on local sub-features; The local sub-features with parallel relationship are secondary aggregated by weighted average and convolution operation functions, and the local sub-features with the same data points are aggregated into the same feature space; S33: multiple feature spaces are fitted into a new global feature; Fit multiple feature spaces into a new global feature through a weighted average function; S4: Get the optimal local features and the optimal global features; The global features and the new global features are fitted by a fitting algorithm to form the optimal global features; Perform algorithm fitting on the multiple local sub-features obtained in step S3 to form an optimal local feature; S5: Feature extraction completed; After the optimal global features and the optimal local features are extracted, the extracted optimal global features and the optimal local features are output to the downstream algorithm for subsequent calculations.

2. The method for extracting driving data features based on topological analysis according to claim 1, characterized in that: In step S31, the method for extracting local sub-features of each filtering space is as follows: The filter space forms different feature spaces by selecting different calculation factors; Aggregate the data of each filter space, adjust the calculation factor value, and extract the local sub-features of each filter space.

3. The method for extracting driving data features based on topological analysis according to claim 1, characterized in that: The features include numerical features and category features.

4. The method for extracting driving data features based on topological analysis according to claim 1, characterized in that: The local sub-features include respective data points.

Images