CN104821082B - Short-time traffic flow prediction method based on integrated evaluation - Google Patents

Abstract

The invention discloses a short-term traffic flow prediction method based on comprehensive evaluation. Through research and analysis combined with actual traffic flow prediction scenarios, undetermined evaluation indicators are obtained; historical data of undetermined evaluation indicators is collected to obtain a historical database; through data preprocessing, construction Based on the historical data matrix of the historical database; by calculating the correlation coefficient between the undetermined evaluation indicators and traffic flow, the main evaluation indicators are screened out, and the respective contribution weights of the main evaluation indicators are calculated; combined with real-time traffic flow data, through the improved time series The similarity measurement method analyzes the historical data of traffic flow in the historical database to obtain the predicted value of traffic flow under each main evaluation index; according to the obtained predicted value of traffic flow under each main evaluation index, a linear weighted comprehensive evaluation is used The method predicts traffic flow, and the method of the present invention synthesizes multidimensional time series to obtain more accurate numerical prediction results.

Description

A Short-term Traffic Flow Forecasting Method Based on Comprehensive Evaluation

technical field

The invention relates to a traffic flow prediction method, in particular to a short-term traffic passenger flow prediction method based on comprehensive evaluation.

Background technique

In recent years, with the continuous development of the economy and the explosive growth of various types of motor vehicles, the traffic problems in cities have become increasingly serious. Under the increasing traffic pressure, the concept of intelligent transportation came into being. Intelligent transportation is the development direction of the future transportation system. It is a comprehensive transportation management system established by effectively integrating advanced information technology, data communication transmission technology, electronic sensing technology, control technology and computer technology into the entire ground traffic management system. system. Through the research and analysis of historical traffic flow data by data mining technology, relevant rules can be obtained. Applying the obtained rules to the current scene and predicting the short-term traffic flow in the future is the premise and basis of dynamic traffic guidance. Only with high-precision real-time traffic flow information can we further use modern communication technology, computer technology, etc. to serve travelers Provide the best driving route to achieve the goal of smooth and efficient network operation.

Traffic flow can be thought of as a time series of changing values over time. Most of the current research on time series is limited to one-dimensional time series. However, in complex practical application scenarios, research objectives are often affected by multiple factors, so if only one-dimensional situations are considered, it is likely to produce large deviations. Therefore, multidimensional time series must be introduced. Multidimensional time series forecasting refers to the forecasting results of the analysis target obtained after comprehensive analysis of a series of observations of various attributes obtained in time order. At present, the research direction of multidimensional time series mainly includes classification, clustering, correlation regularity, exploratory data analysis and other similarity-based pattern mining research. The commonly used time series analysis models include wavelet analysis, neural network, chaos theory, nonlinear prediction model of support vector machine method, etc. Moreover, the multi-dimensional time series prediction also focuses on the relatively macroscopic prediction analysis results and law summary based on the pattern mining of a large amount of data, but there is no accurate numerical prediction result. It cannot be applied to scenarios that have high requirements for numerical prediction accuracy. The combined forecasting model has strict data requirements and complex implementation, which is difficult to implement in practical application scenarios.

Contents of the invention

In order to solve the above technical problems, the present invention proposes a short-term traffic flow prediction method based on comprehensive evaluation.

The technical solution adopted in the present invention is: a short-term traffic flow prediction method based on comprehensive evaluation, comprising the following steps:

S1. Obtain undetermined evaluation indicators through research and analysis combined with actual traffic flow prediction scenarios;

S2. Collect the historical data of the undetermined evaluation indicators obtained in step S1 to obtain a historical database;

S3. Through data preprocessing, construct a historical data matrix based on the historical database;

S4. Calculate the correlation coefficient between each undetermined evaluation index and traffic flow, convert the correlation coefficient into a percentage, then sort the correlation coefficients in descending order, and accumulate the correlation coefficients in descending order until If the accumulation result is greater than or equal to the first threshold, the accumulation operation is stopped, thereby eliminating unaccumulated undetermined evaluation indicators with small correlation coefficients, thereby obtaining main evaluation indicators, and calculating the respective contribution weights of the main evaluation indicators;

S5. Combined with real-time traffic flow data, the traffic flow historical data in the historical database is analyzed through the improved Euclidean distance, and the traffic flow prediction value under each main evaluation index is obtained; according to the obtained traffic flow under each main evaluation index For the traffic flow forecast value, the linear weighted comprehensive evaluation method is used to predict the traffic flow.

Further, the historical database created in step S2 is: DB_TABLE=[DTTM,F,r ₁ ,r ₂ ,...,r _p ];

Among them, DTTM represents the record sampling time point, and F represents the historical traffic flow data corresponding to the sampling time point.

Further, the step S3 specifically includes: S31: data preprocessing; S32: constructing a historical data matrix;

The step S31 data preprocessing specifically includes the following sub-steps:

S311: Digitizing the symbolic information, transforming the symbolic information into numerical information;

S312: Numerical missing and numerical error preprocessing, for the discontinuous traffic flow numerical sequence, use the average interpolation method to complete the data; and for the statistical sum data of multiple time periods, use the method of equal division by time period to complete the data of each time period ; For obviously wrong data, the data point is deleted, and the average interpolation method is used to fill it up;

S313: Data preprocessing, using data segment processing to perform data processing on the value sets of each undetermined evaluation index;

The step S32: constructing a historical data matrix is specifically: constructing a historical data matrix based on the historical database in step S2.

Further, the step S4 index screening and weight calculation specifically includes the following sub-steps:

S41: Data standardization operation, and obtain a standardized historical data matrix according to the standardized data;

S42: Calculate the correlation coefficients between each evaluation index and traffic flow in the standardized historical data matrix, sort the correlation coefficients, convert the correlation coefficients into percentages, and then sort the correlation coefficients in descending order, and sort the correlation coefficients by Accumulate sequentially from large to small until the accumulation result is greater than or equal to the first threshold, then stop the accumulation operation, and eliminate the undetermined evaluation indicators that have not been accumulated and have a small correlation coefficient, thereby obtaining K main evaluation indicators;

S43: Calculate the respective contribution rights of the main evaluation indicators to the traffic flow.

Further, the step S5 specifically includes the following sub-steps:

According to the length of the set time window T and the sampling frequency t of historical data, the number of historical data records in the time window is obtained;

S52: Construct the traffic flow matrix, and calculate the correlation coefficient matrix corresponding to the traffic flow matrix, and calculate the proportion of historical data in different time periods in the time window according to the obtained correlation coefficient matrix:

S53: Calculate the traffic flow prediction value under each main evaluation index according to the improved Euclidean distance;

S54: Predict the traffic flow using the prediction method of linear weighted comprehensive evaluation.

Further, the step S52 specifically includes the following sub-steps:

S521: Construct a traffic flow matrix, assuming that the time series of traffic flow F with a length of len is: f ₁ , f ₂ ,...f _i ,...,f _len , satisfying the condition len>>n, then construct the following traffic flow matrix TF:

Among them, the first column of the matrix represents the data of the forecast time, and the second to n+1 columns represent n traffic flow historical data within the length of the time window;

S522: Calculate the correlation coefficient matrix of the matrix TF:

S523: Calculate the proportion α _i′ of each historical data in the time window:

${\mathrm{<span\; class="notranslate">\; \&alpha;</span>}}_{{\mathrm{<span\; class="notranslate">\; i</span>}}^{<span\; class="notranslate">\; \&prime;</span>}}<span\; class="notranslate">\; =</span>\frac{{\mathrm{<span\; class="notranslate">\; r</span>}}_{\mathrm{<span\; class="notranslate">\; 1,1</span>}<span\; class="notranslate">\; +</span>{\mathrm{<span\; class="notranslate">\; i</span>}}^{<span\; class="notranslate">\; \&prime;</span>}}}{\underset{{\mathrm{<span\; class="notranslate">\; i</span>}}^{<span\; class="notranslate">\; \&prime;</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1</span>}}{\overset{\mathrm{<span\; class="notranslate">\; no</span>}}{\mathrm{<span\; class="notranslate">\; \&Sigma;</span>}}}{\mathrm{<span\; class="notranslate">\; r</span>}}_{\mathrm{<span\; class="notranslate">\; 1,1</span>}<span\; class="notranslate">\; +</span>{\mathrm{<span\; class="notranslate">\; i</span>}}^{<span\; class="notranslate">\; \&prime;</span>}}}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1,2</span>}<span\; class="notranslate">\; ,</span><span\; class="notranslate">\; \&Center\; Dot;</span><span\; class="notranslate">\; \&CenterDot;</span><span\; class="notranslate">\; \&CenterDot;</span><span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; no</span>}<span\; class="notranslate">\; .</span>$

Further, the step S53 specifically includes the following sub-steps:

S531: Obtain the current main evaluation index r _k , the value of the current main evaluation index r _k is Val, then extract the historical data with the value of r _k in the historical database to form a new data set DTS: <DTTM,F>, The number of records in the data set DTS is dts_count;

S532: If dts_count=0, it indicates that when the value of the main evaluation index r _k is Val, there is no corresponding data in the historical database, so set F _{predict_k} =0, and go to step S535 at the same time, otherwise go to step S533;

Among them, F _{predict_k} represents the traffic flow prediction value corresponding to the k-th main evaluation index;

S533: Record the time series of length m from the moment (DTTM _now -T) to the time period of DTTM _now as F_NOW, and for each DTTM _a in the data set DTS, record the time series from the moment (DTTM _a -T) to DTTM _a The time series with length m in the segment is recorded as F_HISTORY _a , where a=1,2,...,dts_count;

Among them, DTTM _now represents the prediction time point;

S534: Under the main evaluation index r _k , use the following formula to predict the specific value of traffic flow:

F _{predict_k} = F _{key_dttm}

Wherein, F _{key_dttm} represents the value of the traffic flow in the historical record corresponding to key_dttm in the DTTM value in the history database;

S535: Judging whether each main evaluation index has completed the traffic flow prediction, if yes, end the judgment; otherwise, go to step S531 to obtain the next main evaluation index.

Furthermore, the value of key_dttm in the step S534 should satisfy the following conditions:

dist(F_NOW,F_HISTORY _{key_dttm} )

=min{dist(F_NOW,F_HISTORY _a )}, a=1,2,...,dts_count,

Among them, dist(i) means to calculate the similarity based on the improved Euclidean distance, and min{·} means to take the minimum value in the set.

Further, the step S54 is specifically:

S541: record the cumulative contribution weight of each evaluation index as w _total , then when the evaluation index r _k produces an effective prediction result, accumulate its contribution weight to the traffic flow;

S542 : According to w _total obtained in step S541 , predict the traffic flow F _predict using a linear weighted comprehensive evaluation prediction method.

Beneficial effects of the present invention: a short-term traffic flow prediction method based on comprehensive evaluation of the present invention obtains undetermined evaluation indicators through research and analysis combined with actual traffic flow prediction scenarios; collects historical data of undetermined evaluation indicators to obtain a historical database; Through data preprocessing, construct a historical data matrix based on the historical database; by calculating the correlation coefficient between each undetermined evaluation index and traffic flow, select the main evaluation index, and calculate the respective contribution weight of the main evaluation index; combined with real-time traffic flow data , through the improved time series similarity measurement method to analyze the historical traffic flow data in the historical database, and obtain the traffic flow forecast value under each main evaluation index; according to the obtained traffic flow forecast value under each main evaluation index , using a linear weighted comprehensive evaluation method to predict traffic flow, and the method of the present invention synthesizes multidimensional time series to obtain more accurate numerical prediction results.

Description of drawings

Fig. 1 is a flow chart of the method of the present invention.

Fig. 2 is a flow chart of data preprocessing in the present invention.

Fig. 3 is a flow chart of the screening and weight calculation of the evaluation index in the present invention.

Fig. 4 is a schematic diagram of the time window of the present invention.

Fig. 5 is the time series similarity measurement method based on the improved Euclidean distance of the present invention.

FIG. 6 is a schematic diagram of the number of records of historical data in a time window provided by an embodiment of the present invention.

FIG. 7 is a detailed step-by-step exploded diagram of step S52.

FIG. 8 is a detailed step-by-step exploded diagram of step S53.

detailed description

In order to facilitate those skilled in the art to understand the technical content of the present invention, the content of the present invention will be further explained below in conjunction with the accompanying drawings.

As shown in Figure 1, it is the method flowchart of the present invention, and the short-term traffic flow prediction algorithm based on comprehensive evaluation proposed by this patent is divided into the following five steps:

S1. Select the factors r ₁ , r ₂ ,…r _p that have an impact on traffic flow through research and analysis combined with the actual traffic flow forecasting scene, and use them as undetermined evaluation indicators;

S2. Collect the historical data of the undetermined evaluation indicators obtained in step S1 to obtain a historical database;

S3. Through data preprocessing, construct a historical data matrix based on the historical database;

S4. Index screening and weight calculation, the indicators r ₁ , r ₂ ,...r _K that have a greater impact on traffic flow are obtained, and the corresponding contribution weights are w ₁ , w ₂ ,...,w _K ;

S5. Consider each evaluation index r _k , combined with real-time traffic flow data, analyze the traffic flow historical data in the historical database through the improved time series similarity measurement method, and obtain the traffic flow prediction value F under the condition of a single evaluation index _{predict_k} ; The future short-term traffic flow F _predict is obtained by using the prediction method of linear weighted comprehensive evaluation.

The embodiment of the present invention takes an airport as an example.

The selection of evaluation indicators in step S1 can be based on research and analysis combined with actual traffic flow forecasting scenarios, and a total of p factors that have an impact on traffic flow F can be selected as undetermined evaluation indicators for traffic flow forecasting, denoted as r ₁ , r ₂ ,… , r _p . Airport taxi capacity is affected by many factors, including flight information r ₁ , weather information r ₂ , different time periods r ₃ , the number of people taking taxis r ₄ , the current capacity of taxis r ₅ and so on.

Step S2 collects the historical data of the undetermined evaluation indicators obtained in step S1, and obtains the historical database as follows: DB_TABLE=[DTTM,F,r ₁ ,r ₂ ,...,r _p ];

Among them, DTTM represents the record sampling time point, and F represents the historical traffic flow data corresponding to the sampling time point.

The step S3 includes: S31: data preprocessing; S32: constructing a historical data matrix;

As shown in Figure 2, the data preprocessing in step S31 is specifically: the data preprocessing part considers that the indicators r ₁ , r ₂ ,...r _p have various data forms, including symbols, values and other data types, which cannot be directly performed At the same time, irregular or wrong data will cause large deviations in the calculation results. Therefore, it is necessary to preprocess the corresponding data, so that not only better prediction results can be obtained, but also the calculation speed can be effectively improved. As shown in Figure 2, it is the data preprocessing flowchart of the present invention, and the main steps of concrete data preprocessing are as follows:

S311: Digitize the symbol information. Since the evaluation indicators r ₁ , r ₂ ,...r _p may contain text and other symbolic information, such as weather information that has a great impact on traffic flow. Therefore, it is necessary to digitize the data of text type to facilitate calculation.

Taking the weather information r ₂ as an example, the degree of its impact on the flight arrival divides the weather information r ₂ into four categories, as shown in Table 1.

Table 1 Classification of weather information

weather type describe Category I weather (no impact) sunny, cloudy Category II weather (weak impact) Showers, light to moderate rain, light rain, light to moderate snow, light snow Category III weather (moderate impact) Moderate snow, sleet, moderate rain, moderate to heavy rain, heavy rain, light fog, haze Category IV weather (severe impact) Heavy snow, floating dust, torrential rain, thunderstorm, rainstorm, fog

Taking the four types of weather as the basic types, the weather change trend and its symbolic representation as shown in Table 2 can be obtained. In this paper, the time window length for investigating the weather change trend is set to 2 hours.

Table 2 Weather change trend and its types

weather change trend Types of Ⅰ → Ⅰ 1 I → II 2 I → III 3 I → IV 4 II→I 5 II → II 6 II → III 7 II → IV 8 Ⅲ→Ⅰ 9 Ⅲ→Ⅱ 10 Ⅲ→Ⅲ 11 Ⅲ→Ⅳ 12 IV → I 13 IV → II 14 IV → III 15 IV → IV 16

S312: Numerical missing and numerical error preprocessing. For the discontinuous numerical sequence of traffic flow, the average interpolation method is used to complete the data. For the statistics and data of multiple time periods, the method of evenly dividing by time period is used to complete the data of each time period. For obviously wrong data, the data point is deleted, and the average interpolation method is used to fill it up.

S313: data preprocessing. Since any evaluation index r _j ,j=1,2,...,p has a relatively large range of values, if each value is regarded as a category, it will lead to too many categories and affect the prediction efficiency. For the evaluation index r _j , values of the same order of magnitude often have the same impact on the traffic flow F, so the method of data segment processing can be used to evaluate the value set of the evaluation index r _j {v _je |j=1, 2,...,p; e=1,2,...s} for processing, where s represents the value number of the evaluation index r _j . The value set of evaluation index r _i after data preprocessing is expressed as:

r _j → {v ^* _je |j＝1,2,…,p; e＝1,2,…s ^* }

Among them, v ^* _je represents the value of the evaluation index r _j after data preprocessing, and s ^* represents the number of values of the evaluation index r _j after data preprocessing.

Step S32: Constructing the historical data matrix is specifically: according to the data table of the historical database obtained in step S2, as follows:

DB_TABLE=[DTTM,F,r ₁ ,r ₂ ,...,r _p ];

Among them, DTTM represents the record sampling time point, and F represents the historical traffic flow data corresponding to the sampling time point. The historical data matrix is constructed based on the above historical data table as follows:

MTX＝[F,r ₁ ,r ₂ ,...,r _p ]

Step S4 evaluation index screening and weight calculation mainly judges and selects the evaluation index proposed in step S1. As shown in Figure 3, by calculating the correlation between each index and traffic flow, the evaluation index with greater correlation is selected, and the index with less correlation or no correlation is removed. Specific steps are as follows:

S41: data standardization operation, standardize the data according to the following formula:

${{\mathrm{<span\; class="notranslate">\; x</span>}}_{\mathrm{<span\; class="notranslate">\; ig</span>}}}^{<span\; class="notranslate">\; *</span>}<span\; class="notranslate">\; =</span>\frac{{\mathrm{<span\; class="notranslate">\; x</span>}}_{\mathrm{<span\; class="notranslate">\; ig</span>}}<span\; class="notranslate">\; -</span>\stackrel{<span\; class="notranslate">\; \&OverBar;</span>}{{\mathrm{<span\; class="notranslate">\; x</span>}}_{\mathrm{<span\; class="notranslate">\; g</span>}}}}{\sqrt{\mathrm{<span\; class="notranslate">\; var</span>}<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; x</span>}}_{\mathrm{<span\; class="notranslate">\; g</span>}}<span\; class="notranslate">\; )</span>}}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1,2</span>}<span\; class="notranslate">\; ,</span><span\; class="notranslate">\; \&CenterDot;</span><span\; class="notranslate">\; \&Center\; Dot;</span><span\; class="notranslate">\; \&Center\; Dot;</span><span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; no</span>}<span\; class="notranslate">\; ;</span>\mathrm{<span\; class="notranslate">\; g</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1,2</span>}<span\; class="notranslate">\; ,</span><span\; class="notranslate">\; \&Center\; Dot;</span><span\; class="notranslate">\; \&Center\; Dot;</span><span\; class="notranslate">\; \&Center\; Dot;</span><span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; q</span>}$

in, and Represent the mean value and variance of the gth item in the historical data matrix respectively, and q=p+1, obtain the matrix MTX'=(x _ig ^* ) _n×q after standardization;

S42: Calculate the correlation coefficient u _Fg between each evaluation index of the matrix MTX′ and the traffic flow, as follows:

${\mathrm{<span\; class="notranslate">\; u</span>}}_{\mathrm{<span\; class="notranslate">\; Fg</span>}}<span\; class="notranslate">\; =</span>\frac{\underset{\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1</span>}}{\overset{\mathrm{<span\; class="notranslate">\; no</span>}}{\mathrm{<span\; class="notranslate">\; \&Sigma;</span>}}}<span\; class="notranslate">\; (</span>{{\mathrm{<span\; class="notranslate">\; x</span>}}^{<span\; class="notranslate">\; *</span>}}_{\mathrm{<span\; class="notranslate">\; iF</span>}}<span\; class="notranslate">\; -</span>\stackrel{<span\; class="notranslate">\; \&OverBar;</span>}{{{\mathrm{<span\; class="notranslate">\; x</span>}}_{\mathrm{<span\; class="notranslate">\; f</span>}}}^{<span\; class="notranslate">\; *</span>}}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; (</span>{{\mathrm{<span\; class="notranslate">\; x</span>}}^{<span\; class="notranslate">\; *</span>}}_{\mathrm{<span\; class="notranslate">\; ig</span>}}<span\; class="notranslate">\; -</span>\stackrel{<span\; class="notranslate">\; \&OverBar;</span>}{{{\mathrm{<span\; class="notranslate">\; x</span>}}_{\mathrm{<span\; class="notranslate">\; g</span>}}}^{<span\; class="notranslate">\; *</span>}}<span\; class="notranslate">\; )</span>}{\sqrt{\underset{\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1</span>}}{\overset{\mathrm{<span\; class="notranslate">\; no</span>}}{\mathrm{<span\; class="notranslate">\; \&Sigma;</span>}}}{<span\; class="notranslate">\; (</span>{{\mathrm{<span\; class="notranslate">\; x</span>}}^{<span\; class="notranslate">\; *</span>}}_{\mathrm{<span\; class="notranslate">\; iF</span>}}<span\; class="notranslate">\; -</span>\stackrel{<span\; class="notranslate">\; \&OverBar;</span>}{{{\mathrm{<span\; class="notranslate">\; x</span>}}_{\mathrm{<span\; class="notranslate">\; f</span>}}}^{<span\; class="notranslate">\; *</span>}}<span\; class="notranslate">\; )</span>}^{\mathrm{<span\; class="notranslate">\; 2</span>}}}\sqrt{\underset{\mathrm{<span\; class="notranslate">\; k</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1</span>}}{\overset{\mathrm{<span\; class="notranslate">\; no</span>}}{\mathrm{<span\; class="notranslate">\; \&Sigma;</span>}}}{<span\; class="notranslate">\; (</span>{{\mathrm{<span\; class="notranslate">\; x</span>}}^{<span\; class="notranslate">\; *</span>}}_{\mathrm{<span\; class="notranslate">\; ig</span>}}<span\; class="notranslate">\; -</span>\stackrel{<span\; class="notranslate">\; \&OverBar;</span>}{{{\mathrm{<span\; class="notranslate">\; x</span>}}_{\mathrm{<span\; class="notranslate">\; g</span>}}}^{<span\; class="notranslate">\; *</span>}}<span\; class="notranslate">\; )</span>}^{\mathrm{<span\; class="notranslate">\; 2</span>}}}}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1,2</span>}<span\; class="notranslate">\; ,</span><span\; class="notranslate">\; \&Center\; Dot;</span><span\; class="notranslate">\; \&Center\; Dot;</span><span\; class="notranslate">\; \&Center\; Dot;</span><span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; no</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; g</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 2,3</span>}<span\; class="notranslate">\; ,</span><span\; class="notranslate">\; \&Center\; Dot;</span><span\; class="notranslate">\; \&Center\; Dot;</span><span\; class="notranslate">\; \&CenterDot;</span><span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; q</span>}$

Sort the correlation coefficients, convert the correlation coefficients into percentages, then sort the correlation coefficients in descending order, and accumulate the correlation coefficients in descending order until the cumulative result is greater than or equal to the first threshold, In this embodiment, the value of the first threshold is 80%, then the accumulation operation is stopped, and the undetermined evaluation indicators with small correlation coefficients that have not been accumulated are eliminated, so as to obtain K main evaluation indicators r ₁ , r ₂ ,...r _k ,... , r _K ;

S43: Calculate the contribution weight w _k of the main evaluation indicators to the traffic flow, the formula is as follows:

${\mathrm{<span\; class="notranslate">\; w</span>}}_{\mathrm{<span\; class="notranslate">\; k</span>}}<span\; class="notranslate">\; =</span>\frac{{\mathrm{<span\; class="notranslate">\; u</span>}}_{\mathrm{<span\; class="notranslate">\; Fk</span>}}}{\underset{\mathrm{<span\; class="notranslate">\; k</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1</span>}}{\overset{\mathrm{<span\; class="notranslate">\; K</span>}}{\mathrm{<span\; class="notranslate">\; \&Sigma;</span>}}}{\mathrm{<span\; class="notranslate">\; u</span>}}_{\mathrm{<span\; class="notranslate">\; Fk</span>}}}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; k</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1,2</span>}<span\; class="notranslate">\; ,</span><span\; class="notranslate">\; \&CenterDot;</span><span\; class="notranslate">\; \&CenterDot;</span><span\; class="notranslate">\; \&CenterDot;</span><span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; K</span>}<span\; class="notranslate">\; ;</span>$

Among them, u _Fk represents the correlation coefficient between the kth main and traffic flow.

In step S5, in the process of traffic flow prediction considering only a single evaluation index, the time series similarity measurement method is used to find the most similar traffic flow data from the historical database as the predicted value. When examining the law of traffic flow over time, it can be considered that traffic flow is a time series whose values change over time. Due to the continuity of traffic flow, the traffic flow data in the past period will affect the traffic flow in the current period, as shown in Figure 4, which is a schematic diagram of the time window. Considering that different historical time periods have different influences on the current traffic flow, the design flow of the method using the improved Euclidean distance as time series similarity measure in the present invention is shown in Fig. 5 . The detailed design measurement method includes the following steps:

S51: Set the length of the time window, assuming that the length of the time window is T (min), then when the sampling frequency of historical data is t (min), the number of historical data records in the time window is:

Among them, record the current prediction time point as t _predict , record the current latest sampling time point as t _sample , Δt=t _predict -t _sample , Indicates rounding.

S52: Construct the traffic flow matrix, and calculate the correlation coefficient matrix corresponding to the traffic flow matrix, and calculate the proportion α _i′ of historical data in different time periods in the time window according to the obtained correlation coefficient matrix:

S53: Calculate the traffic flow F _{predict_k} under each main evaluation index;

S54: Predict the traffic flow F _predict using a linear weighted comprehensive evaluation prediction method.

Described step S51 is specifically: the length of setting time window is T (min), then assumes that the sampling frequency of historical data is t (min), the record number of historical data in available time window is:

Record the current prediction time point as t _predict and the latest sampling time point as t _sample , then the number of historical data records in the time window can be expressed as:

Among them, Δt=t _predict -t _sample , Indicates rounding.

As shown in Figure 6, 1) The prediction point is half an hour, namely 8:00, 8:30, 9:00, 9:30, 10:00, and the sampling time is half an hour, namely 8:00, 8:30 , 9:00, 9:30, 10:00, then according to the current prediction time point is t _predict = 9:30, and the current latest sampling time point is t _sample = 9:00, then Δt = t _predict -t _sample = 30min<T=130min, then

2) The prediction point is half an hour, that is, 8:00, 8:30, 9:00, 9:30, 10:00, and the sampling time point is 5 minutes from the half point, that is, 7:55, 8:25, 8:00 55, 9:25, 9:55, then according to the current prediction time point is t _predict = 9:30, and the current latest sampling time point is t _sample = 9:25, then Δt = t _predict -t _sample = 5min< T=130min, then

3) The prediction point is half an hour, that is, 8:00, 8:30, 9:00, 9:30, 10:00, and the sampling time is 15 minutes from the half point, that is, 7:45, 8:15, 8:00 45, 9:15, 9:45, then according to the current prediction time point is t _predict = 9:30, and the current latest sampling time point is t _sample = 9:15, then Δt = t _predict -t _sample = 15min< T=130min, then

As shown in FIG. 7 , the step S52 is specifically: when examining the law of traffic flow changing over time, it can be considered that the traffic flow is a time series whose values change over time. This paper uses the method of using the improved Euclidean distance as the time series similarity measure, and finds the record most similar to the current scene from the historical database, and uses it as the predicted value. The expression of the improved Euclidean distance is as follows:

$\mathrm{<span\; class="notranslate">\; dist</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; P</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; Q</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; =</span>\sqrt{\underset{\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1</span>}}{\overset{\mathrm{<span\; class="notranslate">\; m</span>}}{\mathrm{<span\; class="notranslate">\; \&Sigma;</span>}}}{\mathrm{<span\; class="notranslate">\; \&alpha;</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}{<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; p</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; -</span>{\mathrm{<span\; class="notranslate">\; q</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; )</span>}^{\mathrm{<span\; class="notranslate">\; 2</span>}}}$

Among them, P and Q represent the time series of length _m respectively, p _i′ and q _i′ represent the sequence values at the corresponding time points, α _i′ represents the proportion of the time series, and satisfy As shown in Figure 7, it specifically includes the following sub-steps:

S521: Assume that the time series of traffic flow F with a length of len is: f ₁ , f ₂ ,...f _i ,...,f _len , and the condition len>>n is satisfied. Then construct the following traffic flow matrix TF:

Among them, the first column of the matrix represents the data of the forecast time ^, and the last n columns represent the n ^· traffic flow historical data in the time window.

S522: Calculate a correlation coefficient matrix.

Calculate the correlation coefficient matrix for the matrix TF:

S523: Calculate the proportion α _i′ of each historical data in the time window:

${\mathrm{<span\; class="notranslate">\; \&alpha;</span>}}_{{\mathrm{<span\; class="notranslate">\; i</span>}}^{<span\; class="notranslate">\; \&prime;</span>}}<span\; class="notranslate">\; =</span>\frac{{\mathrm{<span\; class="notranslate">\; r</span>}}_{\mathrm{<span\; class="notranslate">\; 1,1</span>}<span\; class="notranslate">\; +</span>{\mathrm{<span\; class="notranslate">\; i</span>}}^{<span\; class="notranslate">\; \&prime;</span>}}}{\underset{{\mathrm{<span\; class="notranslate">\; i</span>}}^{<span\; class="notranslate">\; \&prime;</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1</span>}}{\overset{\mathrm{<span\; class="notranslate">\; no</span>}}{\mathrm{<span\; class="notranslate">\; \&Sigma;</span>}}}{\mathrm{<span\; class="notranslate">\; r</span>}}_{\mathrm{<span\; class="notranslate">\; 1,1</span>}<span\; class="notranslate">\; +</span>{\mathrm{<span\; class="notranslate">\; i</span>}}^{<span\; class="notranslate">\; \&prime;</span>}}}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1,2</span>}<span\; class="notranslate">\; ,</span><span\; class="notranslate">\; \&CenterDot;</span><span\; class="notranslate">\; \&CenterDot;</span><span\; class="notranslate">\; \&CenterDot;</span><span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; no</span>}$

Calculate the similarity according to the improved Euclidean distance formula

Among them, P represents the time series of length m, Q represents the time series of length m, p _i′ and q _i′ represent the sequence values of corresponding time points, and α _i′ represents the proportion of each historical data in the time series specific gravity, and satisfy

$\underset{{\mathrm{<span\; class="notranslate">\; i</span>}}^{<span\; class="notranslate">\; \&prime;</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1</span>}}{\overset{\mathrm{<span\; class="notranslate">\; m</span>}}{\mathrm{<span\; class="notranslate">\; \&Sigma;</span>}}}{\mathrm{<span\; class="notranslate">\; \&alpha;</span>}}_{{\mathrm{<span\; class="notranslate">\; i</span>}}^{<span\; class="notranslate">\; \&prime;</span>}}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1</span>}$

The step S53 specifically includes: assuming that the predicted time point is DTTM _now , the value of the corresponding evaluation index r _k is Val. As shown in Figure 8, the specific steps are as follows:

S531: Obtain the current main evaluation index r _k , and the current value of the main evaluation index r _k is Val, then extract the historical data whose value of r _k is Val in the historical database to form a new data set DTS: <DTTM,F> , the number of records in the dataset DTS is dts_count.

S532: If dts_count indicates that when the value of the main evaluation index r _k is Val, there is no corresponding data in the historical database, so set F _{predict_k} =0, and go to step S535, otherwise go to step S533.

S533: Record the time series of length m from the moment (DTTM _now -T) to the time period of DTTM _now as F_NOW, and for each DTTM _a in the data set DTS, record the time series from the moment (DTTM _a -T) to DTTM _a The time series with length m in the segment is recorded as F_HISTORY _a , where a=1,2,...,dts_count;

S534: Under the main evaluation index r _k , use the following formula to predict the specific value of traffic flow:

F _{predict_k} = F _{key_dttm}

Wherein, F _{key_dttm} represents the value of the traffic flow in the historical record corresponding to the DTTM value in the historical database key_dttm. The value of key_dttm should meet the following conditions:

The value of DTTM in the data set DTS as key_dttm should meet the following conditions:

dist(F_NOW,F_HISTORY _{key_dttm} )

=min{dist(F_NOW,F_HISTORY _a )}, a=1,2,...,dts_count

That is, when only the evaluation index r _k is considered, among all historical records, the historical data corresponding to the sampling time key_dttm is most similar to the current scene. Therefore, taking the value of F _{key_dttm} as the traffic flow when examining the main evaluation index r _k is F _{predict_k} .

S535: Judging whether each main evaluation index has completed the traffic flow prediction, if yes, then end the judgment, otherwise go to step S531, execute k=k+1;

Wherein, k represents the kth main evaluation index, and k=1,2,...,K.

The step S54 specifically includes the following sub-steps:

S541: Record the cumulative contribution weight of each evaluation index as w _total , then when the evaluation index r _k produces an effective prediction result, add its contribution weight to the traffic flow to w _total , as shown in the following formula:

w _total = w _total + w _k ;

S542: According to w _total obtained in step S541, the traffic flow F _predict is predicted by using the prediction method of linear weighted comprehensive evaluation, and the calculation formula is as follows:

${\mathrm{<span\; class="notranslate">\; f</span>}}_{\mathrm{<span\; class="notranslate">\; predict</span>}}<span\; class="notranslate">\; =</span>\underset{\mathrm{<span\; class="notranslate">\; k</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1</span>}}{\overset{\mathrm{<span\; class="notranslate">\; K</span>}}{\mathrm{<span\; class="notranslate">\; \&Sigma;</span>}}}<span\; class="notranslate">\; (</span>\frac{{\mathrm{<span\; class="notranslate">\; w</span>}}_{\mathrm{<span\; class="notranslate">\; k</span>}}}{{\mathrm{<span\; class="notranslate">\; w</span>}}_{\mathrm{<span\; class="notranslate">\; total</span>}}}<span\; class="notranslate">\; \&times;</span>{\mathrm{<span\; class="notranslate">\; f</span>}}_{\mathrm{<span\; class="notranslate">\; predict</span>}<span\; class="notranslate">\; \_</span>\mathrm{<span\; class="notranslate">\; k</span>}}<span\; class="notranslate">\; )</span>$

Among them, w _k is the contribution degree of the main evaluation index r _k to the traffic flow F _predict , w _total is the cumulative contribution degree of each main evaluation index, and F _{predict_k} is the traffic flow obtained under the premise of only examining the main evaluation index r _k forecast result.

A short-term traffic flow prediction method based on comprehensive evaluation of the present invention obtains undetermined evaluation indicators through research and analysis combined with actual traffic flow prediction scenarios; collects historical data of undetermined evaluation indicators to obtain a historical database; through data preprocessing, constructs Based on the historical data matrix of the historical database; by calculating the correlation coefficient between the undetermined evaluation indicators and traffic flow, the main evaluation indicators are screened out, and the respective contribution weights of the main evaluation indicators are calculated; combined with real-time traffic flow data, through the improved time series The similarity measurement method analyzes the historical data of traffic flow in the historical database to obtain the predicted value of traffic flow under each main evaluation index; according to the obtained predicted value of traffic flow under each main evaluation index, a linear weighted comprehensive evaluation is used The method predicts traffic flow, and the method of the present invention synthesizes multidimensional time series to obtain more accurate numerical prediction results.

Those skilled in the art will appreciate that the embodiments described here are to help readers understand the principles of the present invention, and it should be understood that the protection scope of the present invention is not limited to such specific statements and embodiments. Various modifications and variations of the present invention will occur to those skilled in the art. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principles of the present invention shall be included within the scope of the claims of the present invention.

Claims (8)

1. A short-term traffic flow prediction method based on comprehensive evaluation is characterized by comprising the following steps:

s1, obtaining an evaluation index to be determined by combining research and analysis with an actual traffic flow prediction scene;

s2, acquiring historical data of the to-be-determined evaluation index obtained in the step S1 to obtain a historical database;

s3, constructing a historical data matrix based on a historical database through data preprocessing;

s4, calculating correlation coefficients of each undetermined evaluation index and traffic flow, converting the correlation coefficients into percentages, sequencing the correlation coefficients in descending order, accumulating the correlation coefficients in descending order until the accumulation result is larger than or equal to a first threshold value, stopping accumulation operation, eliminating the undetermined evaluation indexes with smaller correlation coefficients which are not accumulated, obtaining main evaluation indexes, and calculating the contribution weight of each main evaluation index;

s5, analyzing the traffic flow historical data in the historical database through the improved Euclidean distance by combining with the real-time traffic flow data to obtain a traffic flow predicted value under each main evaluation index; and predicting the traffic flow by adopting a linear weighted comprehensive evaluation method according to the obtained traffic flow predicted value under each main evaluation index.

2. The short-term traffic flow prediction method based on comprehensive evaluation according to claim 1, wherein the historical database created in step S2 is: DB _ TABLE [ DTTM, F, r ]₁,r₂,…,r_p]；

Wherein, DTTM represents the recording sampling time point, F represents the traffic flow historical data corresponding to the sampling time point; r is₁,r₂,…,r_pIndicates an evaluation index to be determined.

3. The short-term traffic flow prediction method based on comprehensive evaluation according to claim 1, wherein the step S3 specifically includes: s31: preprocessing data; s32: constructing a historical data matrix;

the step S31 of preprocessing data specifically includes the following sub-steps:

s311: digitizing the symbol information, and converting the symbol information into digitized information;

s312: numerical value missing and numerical value error preprocessing, namely, for a discontinuous traffic flow numerical value sequence, filling data by adopting an average interpolation method; the data of each time interval is supplemented by adopting a method of equally dividing the data of the statistical sums of a plurality of time intervals according to the time intervals; deleting the data point for the obviously wrong data, and simultaneously adopting an average interpolation method for completing;

s313: data preprocessing, namely performing data processing on the value set of each to-be-evaluated index by adopting data segmentation processing;

the step S32: the specific steps for constructing the historical data matrix are as follows: a history data matrix is constructed based on the history database of step S2.

4. The short-term traffic flow prediction method based on comprehensive evaluation according to claim 1, wherein the step S4 of index screening and weight calculation specifically comprises the following substeps:

s41: carrying out data standardization operation, and obtaining a standardized historical data matrix according to the standardized data;

s42: calculating correlation coefficients of all evaluation indexes and traffic flow in a standardized historical data matrix, sequencing the correlation coefficients, converting the correlation coefficients into percentages, sequencing the correlation coefficients from large to small, sequentially accumulating the correlation coefficients from large to small until an accumulation result is larger than or equal to a first threshold value, stopping accumulation operation, and eliminating undetermined evaluation indexes with smaller correlation coefficients which are not accumulated, thereby obtaining K main evaluation indexes;

s43: and calculating the contribution of each main evaluation index to the traffic flow.

5. The short-term traffic flow prediction method based on comprehensive evaluation according to claim 1, wherein the step S5 specifically includes the following substeps:

s51: obtaining the record number of the historical data in the time window according to the set length of the time window as T and the sampling frequency T of the historical data;

s52: constructing a traffic flow matrix, calculating a correlation coefficient matrix corresponding to the traffic flow matrix, and calculating the ratio of historical data in different time periods in a time window according to the obtained correlation coefficient matrix;

s53: calculating a traffic flow predicted value under each main evaluation index according to the improved Euclidean distance;

s54: and predicting the traffic flow by adopting a linear weighted comprehensive evaluation prediction method.

6. The short-term traffic flow prediction method based on comprehensive evaluation according to claim 5, wherein the step S53 specifically comprises the following substeps:

s531: obtaining the current main evaluation index r_kCurrent main evaluation index r_kWhen the value of (a) is Val, r in the history database is set_kThe historical data with the value Val is extracted to form a new data set DTS:<DTTM,F>the number of records in the data set DTS is DTS _ count;

s532: if dts _ count is 0, this indicates the main evaluation index r_kWhen the value of (A) is Val, there is no corresponding data in the history database, so let F_{predict_k}Go to step S535 if it is 0, otherwise go to step S533;

wherein, F_{predict_k}Representing a traffic flow predicted value corresponding to the kth main evaluation index;

s533: time of Day (DTTM)_now-T) to DTTM_nowThe time series of length m within the time period is denoted as F _ NOW, DTTM for each of the data sets DTS_aTime of Day (DTTM)_a-T) to DTTM_aThe time series with length m in the time period is recorded as F _ HISTORY_aWhere a is 1,2, …, dts _ count;

wherein, DTTM_nowRepresenting a predicted time point;

s534: in the main evaluation index r_kNext, specific values for traffic flow are predicted using the following formula:

F_{predict_k}＝F_{key_dttm}

wherein, F_{key_dttm}The DTTM value in the historical database is a traffic flow value in the historical record corresponding to the key _ DTTM;

s535: and judging whether each main evaluation index completes the prediction of the traffic flow, if so, ending the judgment, otherwise, turning to the step S531 to obtain the next main evaluation index.

7. The short-term traffic flow prediction method based on comprehensive evaluation according to claim 6, wherein the value of key _ dttm in the step S534 should satisfy the following condition:

$\begin{array}{c}dist(F\_NOW,F\_{\mathrm{HISTORY}}_{key\_dttm})\\ =\min \left\{dist(F\_NOW,F\_{\mathrm{HISTORY}}_a)\right\},a=1,2,...,dts\_count\end{array};$

wherein, F _ HISTORY_{key_dttm}Representing the historical data corresponding to the key _ dttm of sampling time, dist (·) representing the similarity obtained according to the improved Euclidean distance, and min {. represents the minimum value in the set.

8. The short-term traffic flow prediction method based on comprehensive evaluation according to claim 5, wherein the step S54 is specifically:

s541: the accumulated contribution weight of each evaluation index is recorded as w_totalWhen the evaluation index r is determined_kWhen an effective prediction result is generated, accumulating the contribution weight of the effective prediction result to the traffic flow;

s542: w obtained from step S541_totalPredicting the traffic flow F by adopting a prediction method of linear weighted comprehensive evaluation_predict。