TWI607324B - Data feature selection and data grouping system and method for vehicle travel time - Google Patents

Description

Data feature selection and data grouping system and method for vehicle travel time

The present invention relates to a data selection and grouping system and method, and particularly relates to a data feature selection and data grouping system and method for effectively classifying data into different clusters and improving intra-group dependencies.

At present, the arrival time estimation method still mainly collects the arrival information reported by the in-vehicle machine in the past to estimate the average speed and travel time between stations, and can be counted according to different weeks and time periods. Historical average speed and travel time can be given. Although this method can quickly provide the estimated arrival time, it is mainly based on the historical data average, and it is impossible to predict the arrival time due to the immediate road condition, so it may cause a large error in the arrival time prediction.

It can be seen that there are still many shortcomings in the above-mentioned methods of use, and there is still room for improvement, which needs to be improved.

In view of the shortcomings derived from the above-mentioned conventional methods, the inventors have improved and innovated, and after years of painstaking research, finally succeeded in research and development of a resource utilization analysis and prediction system and method for a cloud environment of the present invention.

The present invention provides a data feature selection and data grouping system and method, comprising: a data platform server, the data platform server is provided with a plurality of server communication modules, each of the server communication modules receiving a plurality of sensing And storing the sensing data in a database module, and receiving, by a processing module, data prediction and analysis query requirements for each of the sensing data; a plurality of sensing modules, each of the sensing The module receives the sensing data including the environmental information, and transmits the sensing data to the data platform server via a sensor communication module; a plurality of client modules, each of the customers The terminal module sends a data prediction and analysis query request to the data platform server periodically or irregularly via a set of client communication modules, and receives data prediction and analysis query request results; and a data analysis module, The data analysis module performs data prediction and analysis query on the sensing data stored in the database module in the data platform server, and predicts and analyzes the data. The results of seeking the information back to the server platform.

The data analysis module performs data feature selection and data grouping of the sensing data, and the result of generating the data prediction and analysis query request is transmitted back to the data platform server.

The invention provides a data feature selection and data grouping method, and the method comprises: setting a dependency threshold of a plurality of data clusters, and when the correlation between the data clusters is higher than the dependency threshold, each data is The cluster is merged; a plurality of sensing data are randomly selected, and each of the data clusters is grouped according to the set characteristic attribute; Generating each of the sensing data sets of the same feature attribute, generating each of the data clusters and separately calculating the group centers of the data clusters; calculating the dependencies between each of the data clusters and each of the data clusters; When the correlation between the data clusters is higher than the dependency threshold, the data clusters are merged in a high degree of dependency, and the group centers of the merged data clusters are calculated after the combination; the data clusters are calculated again Dependence between each data cluster, and sequentially merging each data cluster with a higher dependency than the dependency threshold, and the dependency between each data cluster and each of the data clusters is lower than the dependency. The threshold is stopped; and the grouping of the data clusters is performed again according to the set characteristic attribute, and stops when each of the sensing data has no feature attribute can be selected.

Among them, the dependence of each data cluster and its respective data clusters has been calculated by the chi-square distributed cumulative probability density function.

The data feature selection and data grouping system method provided by the present invention is compared with the prior art, and the invention is characterized in that the data feature selection and the data grouping system and method can be grouped according to the characteristic attribute values, and the inter-group calculation is performed. Dependency, then merge the highly dependent clusters and calculate the merged cluster center. It is possible to analyze whether the cluster generated by each feature attribute has a dependency, thereby selecting an appropriate feature attribute, and analyzing the influence of the feature attribute on the data. In addition, data feature selection and data grouping systems and methods use the squared ratio to calculate the difference between groups, and use the chi-squared probability density to judge the interdependence between groups, and then merge the clusters with high dependence to generate new ones. Cluster.

100‧‧‧Data Analysis Module

200‧‧‧Data Platform Server

300‧‧‧Sensor module

400‧‧‧Client Module

210‧‧‧Database Module

310‧‧‧Sensor communication module

S201~S207‧‧‧Step procedure

1 is a schematic diagram of a data feature selection and data grouping system of the present invention.

2 is a flow chart of the data feature selection and data grouping method of the present invention.

The embodiments of the multi-lingual speech recognition apparatus and the method thereof according to the present invention will be described below with reference to the related drawings. For the sake of understanding, the same components in the following embodiments are denoted by the same reference numerals.

In the present embodiment, the estimation of the arrival time of the cleaning vehicle will be taken as an example, and the traveling time between the stations and the stations in each section and time period is collected in conjunction with the present invention, and the clustering method is used to analyze the data characteristics and the judgment data. Dependency, cutting data with high data independence into different groups, can avoid the impact of extreme values and improve the accuracy of prediction, based on which to predict the arrival time, as detailed below.

Please refer to FIG. 1, which is a schematic diagram of a data feature selection and data grouping system of the present invention. The data feature selection and data grouping system includes a data analysis module 100, a data platform server 200, a plurality of sensing modules 300, and a plurality of client modules 400. The sensing module 300 obtains periodic and non-periodic sensing environment information and reports the sensing result to the data platform server 200. The sensing module 300 is a sensing device that uses a position on the vehicle-mounted device (eg, The GPS system senses the position of the vehicle-mounted device, and uses the sensing module 300 to detect environmental information (including temperature and humidity, etc.), and determines whether it has reached the location of the clearing station, and when it arrives at the clearing station, it must be used. The sensor communication module 310 reports its arrival information to the data platform server 200. example For example, the vehicle-mounted machine number 1 travels on route No. 1 on 2014/04/01, and arrives at station number 1 at 14:08:02, and the relative humidity of the day is 85.8%. When the data platform server 200 receives the sensing data reported by the sensing module 300, the stored continuous data may be converted into a name data or a discrete data, and then stored in the data platform server 200. Database module 210. Taking the relative humidity as an example, it is set to be an interval of 10%, and 0% to 10% is 1, 10% to 20%, as shown in Table 1.

In addition, the data platform server 200 also has to perform related data processing operations, and the data platform server 200 will calculate the travel time between each station and the station, for example: 2014/04/01 1st station to 2nd The travel time of the station is 2.77 minutes, as shown in Table 2 below. The data analysis module 100 will be able to obtain data from the data platform server 200 and use data feature selection. And the data grouping method performs data feature analysis and data grouping on the obtained data, and then stores the analysis result to the data platform server 200. The client module 400 may periodically or irregularly send a query request for data prediction and analysis to the data platform server 200, and then the data platform server 200 queries the analysis result, and returns the data prediction and analysis result to the client module 400. .

Please refer to FIG. 2, which is a schematic diagram of the data feature selection and data grouping method of the present invention. The main steps are as follows: step S201: setting parameter settings; step S202: selecting feature attributes; step S203: grouping according to the feature attribute values, and setting an initial cluster; step S204: calculating inter-group dependencies; step S205: high dependency The clusters are merged and the merged cluster center is calculated; step S206: confirming whether there are clusters not calculated, and repeating the calculation to no clusters to merge; and step S207: confirming whether the feature attributes are not calculated.

The setting parameter setting of step S201 is to set the dependency threshold of the cluster merge, and the cluster will be merged when the inter-group dependency is higher than the threshold. Such as this In the embodiment, the dependency threshold is set to 90%, and the inter-group dependency needs to be higher than 90% for cluster consolidation.

The selected feature attribute of step S202 is that the feature set is randomly selected according to the data set to perform grouping, and the inter-group dependency calculation is performed, and then the feature attribute of the cluster with high dependency is prioritized for group calculation.

Taking the station-to-station travel time record from 2014/04/01 to 2014/10/23 as an example, the main name data or discrete data has data characteristics such as week, week, relative humidity interval, and have to be selected. A plurality of data feature attributes are set as initial grouping. Take the data characteristics of the week and week and analyze the travel time from the 5th stop to the 6th stop. For example, the travel time from the 5th stop to the 6th stop on the first week of Tuesday is 4.90 points, and so on. 3 results. And you have to choose different data feature attributes for cluster analysis.

The feature attribute values in step S203 are grouped and the initial cluster is set. The initial cluster setting is performed according to the value of the selected feature attribute, so that the data of each attribute value is collected into a cluster, and the data is divided into a plurality of clusters, and the group center of each cluster is calculated separately. Taking the 5th to 6th travel time records of the week and week data feature attributes as an example, each data feature attribute value will be used as the initial cluster. For example, Monday is the first group C ₁ = {null ,null,13.33,14.20,...,16.70}, Tuesday is the second group C ₂ ={4.90,null,null,..,4.77},..., Saturday is the 5th group C ₅ ={null,null, 3.92,...,null}, each cluster is represented as formula (1), where there are n clusters (in this case, n is 5), and the number of data per cluster is m (in this case, m is 30).

C _i ={ c _{i ,1} , c _{i ,2} , c _{i ,3} ,..., c _{i , m} } formula (1)

In addition, the data characteristics of the week and relative humidity intervals are selected and the travel time of the 5th station to the 6th station is analyzed as an example, and each data feature attribute value is used as an initial cluster, that is, a relative humidity interval is a group. And since the data with the same relative humidity interval in the same week is a plurality of pens, the travel time record of the plurality of pens is averaged to obtain the group center, as shown in Table 4.

For example, in the first week, the relative humidity interval is 8 for 2014/04/01 and 2014/04/03, and the travel time from the 5th station to the 6th station is 4.90 minutes and 15.74 minutes, respectively. The travel time of the weekly relative humidity interval 8 is 10.32 minutes, and the cluster center of each cluster can be obtained accordingly. In this embodiment, the data feature attributes of the weekly and relative humidity intervals are selected. The total grouping is divided into four groups, wherein the relative humidity section 6 is the first group, the relative humidity section 7 is the second group, the relative humidity section 8 is the third group, and the relative humidity section 9 is the fourth group.

Step S204 calculates the dependencies between groups, and calculates the dependence of each cluster on the data set between other clusters. In this embodiment, the inter-group chi-square value x _{i , j of} the i-th group C _i and the j-th group C _j is calculated using the formula (2), and the number of data is calculated by the formula (3), and the calculation is dependent on the calculation. It is necessary to use the formula (4) to analyze the cumulative probability density function of the chi-square distribution, and then use the formula (5) to calculate the inter-group dependence s _{i , j} . For example, the 5th to 6th travel time records after selecting the week and week data feature attributes from 2014/04/01 to 2014/10/23, for example, the first group C ₁ (Monday cluster) and the second Group C ₂ (Tuesday cluster) has a chi-squared value x _1,2 of 637.77, the number of data k _{i , j} is 18, and the inter-group dependence s _{1,2 is} calculated using equation (4) and formula (5). For 0%, and so on, the dependence between the two groups can be obtained, as shown in Table 5.

s _{i , j} =1- F ( x _{i , j} ; k _{i , j} -1) Equation (5)

In step S205, the clusters with high dependency are merged, and the merged cluster center is calculated: when the inter-cluster dependency is determined to be higher than the set dependency threshold, the clusters with high dependency are preferentially merged, and the cluster is clustered. The group center of the merged cluster is calculated after the merge. Taking the inter-group dependence of the data feature attributes of the selected week and relative humidity intervals as an example, since the dependence of the second group and the fourth group is up to 99.47% and is higher than the dependency threshold of 90%, the second group The group 4 is preferentially merged, and the group center is calculated by the formula (6) at the time of the merger. The combined results are shown in Table 6.

Step S206, confirming whether there is a cluster uncalculated, and repeating the calculation to no cluster can be merged: calculating (1) the dependencies between groups in each cluster, (2) merging the clusters with high dependency, and calculating the merge After the cluster center; stop until no clusters can be merged. Taking the inter-group dependence of the data feature attributes of the selected week and relative humidity intervals as an example, after the combination of the second group (Tuesday) and the fourth group (Friday), the inter-group dependence is recalculated. The results are shown in Table 7. . And since the dependence of the first group (Monday) and the third group (Thursday) is 92.84%, and is higher than the dependency threshold of 90%, the first group and the third group are merged.

After the completion of the first group (Monday) and the third group (Thursday), it will be heavy again. The new inter-group dependence is calculated, and since there is no inter-group dependence higher than the dependency threshold of 90%, the sub-group under the feature attribute of this data is selected.

In step S207, it is confirmed whether the feature attribute is not calculated, and each feature attribute is calculated: the data set merged according to the cluster is calculated by other unselected feature attributes (1) the selected feature attribute is selected; 2) grouping according to the characteristic attribute values and setting the initial cluster; (3) calculating the dependencies between the groups; (4) merging the clusters with high dependency and calculating the merged cluster center; (5) confirming whether there is a cluster Not calculated, and repeated calculations to no cluster can be merged; it is stopped until no feature attributes can be selected. Taking this embodiment as an example, after completing the grouping of the feature feature attributes of the selected week and the week, it is determined whether other feature attributes can be grouped. The data characteristics of the selected week and relative humidity intervals can be observed in the data (as shown in Table 4), and the inter-group dependence of the grouping under the characteristic attribute of the data is calculated. Table 8 can be obtained. And because the correlation between groups is not higher than the dependency threshold of 90% under the data characteristic attribute of the selected week and relative humidity interval, cluster consolidation is not performed.

Data feature selection and data grouping systems and methods can be used to group data and obtain high-dependency data feature attributes. Taking this example as an example, select data features. The attribute 'week' will be more suitable as the data grouping reference attribute than the selection material feature attribute 'relative humidity interval'.

In summary, the present invention has achieved the desired effect under the prior art, and is not familiar with the skill of the art, and its progress and practicality have been consistent with patents. Apply for the requirements, and file a patent application in accordance with the law, and ask your bureau to approve the application for the invention patent, in order to encourage creation, to the sense of virtue.

The above is intended to be illustrative only and not limiting. Any other equivalent modifications or alterations of the present invention are intended to be included in the scope of the appended claims.

S201~S207‧‧‧Step procedure

Claims (8)

A data feature selection and data grouping method for vehicle travel time includes the steps of: obtaining, by a data platform server, a time interval between a vehicle and a station at each time interval; setting a plurality of data clusters to merge one another a threshold value, when each of the data clusters is higher than the dependency threshold, the data cluster is merged into the data platform server; and the environment associated with the vehicle is collected through the sensing module. Each of the plurality of sensing materials of the information randomly selects each of the sensing data and performs grouping of the data clusters according to the set characteristic attribute and the time when the vehicle arrives at the station and the other station at different time periods and returns the data platform. a server; the data platform server collects each of the sensing data of the same feature attribute, generates each data cluster, and separately calculates a group center of each data cluster and calculates a dependency between each data cluster and each of the data clusters; Sexuality; when the dependence between each data cluster and its respective data clusters is higher than the dependency threshold, the dependency is sequential Each of the data clusters is merged until the dependency between each of the data clusters and each of the data clusters is lower than the dependency threshold, and the merged data cluster is calculated according to the data platform server. The group center recalculates the dependencies between groups; and predicts the time of arrival of the vehicle based on the degree of dependence of different groups. For example, the data feature selection and data grouping method of the vehicle travel time described in claim 1 wherein the probability distribution function is calculated by the chi-square distribution to calculate the dependency between each data cluster and each of the data clusters. Data feature selection and data points for vehicle travel time as described in item 1 of the patent application scope The group method, wherein each data attribute calculation step of the cluster center of the merged data cluster of the two clusters comprises: if the data attribute has a value in the cluster i, and the data attribute has no value in the cluster j, the merged data cluster The data attribute adopts the value of the data attribute of the cluster i; if the data attribute has no value in the cluster i, and the data attribute has a value in the cluster j, the data attribute of the merged data cluster uses the data attribute cluster j to the value of the data attribute If the data attribute has a value in cluster i, and the data attribute has a value in cluster j, then the data attribute of the merged data cluster uses the data attribute to average the data attribute of the cluster i and the data of the cluster j; If the data attribute has no value in cluster i and the data attribute has no value in cluster j, then the data attribute of the merged data cluster is set to a null value. A data grouping method for vehicle travel time includes the steps of: obtaining, by a data platform server, a time interval between a vehicle arriving at a station and another station at different time periods, and collecting, by the sensing module, a travel associated with the vehicle. a plurality of sensing data of the environmental information; collecting the sensing data of the same characteristic attribute, generating each of the data clusters and separately calculating the group centers of the data clusters; calculating the data clusters by using the data platform server Dependency between each data cluster; when the dependency between each data cluster and its respective data clusters is higher than the dependency threshold, each data cluster with high dependency is sequentially merged, and the combined calculation is performed. The cluster center of the merged data cluster; again, the data platform server calculates the dependency between each data cluster and each of the data clusters, and sequentially performs the data clusters whose dependencies are higher than the dependency threshold. Combined And stopping when each of the data clusters and their respective data clusters are below the dependency threshold; and predicting the time of arrival of the vehicle according to different group dependency degrees. A data feature selection and data grouping method for vehicle travel time, the steps include: obtaining a plurality of sensing materials and vehicle travel time from a data platform server or a database module; setting a dependency threshold of a plurality of data clusters When the correlation between the data clusters is higher than the dependency threshold, each of the data clusters is merged; a plurality of sensing data are randomly selected, and each of the data clusters is grouped according to the set characteristic attribute; Each of the sensing data sets of the feature attributes generates each of the data clusters and separately calculates a group center of each of the data clusters; calculates a dependency between each of the data clusters and each of the data clusters; and each of the data clusters and the respective data When the inter-cluster dependence is higher than the dependency threshold, each of the data clusters with high dependency is sequentially merged, and the group center of the merged data cluster is calculated after the combination; each of the data clusters is calculated again. Dependency between data clusters, and sequentially clustering each of the data clusters with a higher dependency than the dependency threshold, When the dependency between the data cluster and each of the data clusters is lower than the dependency threshold, the clustering of the data clusters is performed again according to the set characteristic attribute, and stops when each of the sensing data has no characteristic attribute can be selected. And according to the results of data feature selection and data grouping, the data of different clusters must be built separately to predict the travel time, and travel time prediction in each cluster. Information feature selection and data points of vehicle travel time as described in item 5 of the patent application scope A group method in which a feature attribute includes at least a week, a week, a relative humidity, and a travel time. For example, the data feature selection and data grouping method of the vehicle travel time described in claim 5, wherein the dependence of the data cluster and each of the data clusters has been calculated by the chi-square distributed cumulative probability density function. For example, the data feature selection and data grouping method of the vehicle travel time described in claim 5, wherein each data attribute calculation step of the cluster center of the merged data cluster of the two clusters comprises: if the data attribute is in the cluster i If there is a value, and the data attribute has no value in the cluster j, then the data attribute of the merged data cluster uses the value of the data attribute of the cluster i; if the data attribute has no value in the cluster i, and the data attribute has a value in the cluster j, Then, the data attribute of the merged data cluster uses the value of the data attribute of the data attribute cluster; if the data attribute has a value in the cluster i, and the data attribute has a value in the cluster j, the data attribute of the merged data cluster adopts the attribute The data attribute is the average of the data attribute of the cluster i and the data of the cluster j; and if the data attribute has no value in the cluster i, and the data attribute has no value in the cluster j, then the data attribute of the merged data cluster is set to be empty. value.