Train Traction Energy Consumption Abnormality Detection Method Based on Boosting and Multi-Predictor Fusion

Abstract

The present invention provides a method for detecting abnormal train traction energy consumption based on Boosting and multi-predictor fusion. The method includes: processing the second-level cumulative train traction energy consumption data to form a time series data set on daily, weekly, monthly, and annual time scales; constructing a variety of time series predictors, verifying each time series predictor, and using the Boosting algorithm to adjust the prediction result weights of each time series predictor to form a multi-predictor fusion model; inputting the second-level cumulative train traction energy consumption data of the train to be analyzed into the multi-predictor fusion model, and judging the traction energy consumption status of the train to be analyzed according to the output value of the multi-predictor fusion model. The abnormal judgment proposed by the present invention is based on a variety of time series prediction models and the Boosting algorithm to optimize the weights. The obtained prediction values are weighted and verified, which conforms to the actual energy consumption level of the engineering site and is convenient for the staff to understand and use.

Description

Technical Field

[0001] The present invention relates to the technical field of train energy consumption control, and particularly to a method for detecting abnormal train traction energy consumption based on the fusion of Boosting and multiple predictors. Background Art

[0002] In recent years, the scale of the urban rail transit network and the passenger volume have been continuously increasing, and the overall energy consumption has been rising. Analyzing abnormal traction energy consumption to avoid unnecessary energy loss is a new way to achieve energy conservation and emission reduction. Therefore, how to timely and efficiently detect and locate abnormal energy consumption of trains to improve the management level of train energy consumption, achieve energy conservation, and reduce costs has become the focus of attention of relevant urban rail transit operation units and many scholars.

[0003] Currently, the methods for detecting abnormal train traction energy consumption in the prior art include: regularly manually copying the power meter data of the substation and the on-vehicle TMS (Train Management System) data, and combining the specific energy consumption threshold to judge abnormalities. Using feature label data combined with machine learning algorithms to obtain the energy consumption prediction value as the basis for abnormal judgment.

[0004] The disadvantages of the above-mentioned methods for detecting abnormal train traction energy consumption in the prior art are as follows: The specific energy consumption index and the threshold method do not fully utilize the data recorded during the train operation process, and a single threshold cannot specifically determine abnormalities, resulting in many missed reports and false alarms. Applying machine learning algorithms to the analysis of abnormal urban rail traction energy consumption has poor interpretability.

[0005] Therefore, whether it is possible to establish a framework and method for analyzing abnormal train traction energy consumption values that fully apply train operation data and have a more comprehensive evaluation angle is a problem that needs to be solved currently. Summary of the Invention

[0006] An embodiment of the present invention provides a method for detecting abnormal train traction energy consumption based on the fusion of Boosting and multiple predictors to effectively detect abnormal train traction energy consumption.

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

[0008] A method for detecting abnormal train traction energy consumption based on the fusion of Boosting and multiple predictors includes:

[0009] Processing the second-level cumulative train traction energy consumption data to form a time series data set at daily, weekly, monthly, and annual time scales;

[0010] Constructing multiple time series predictors, verifying each time series predictor, and using the Boosting algorithm to adjust the prediction result weights of each time series predictor to form a multi-predictor fusion model;

[0011] Input the train traction energy consumption data accumulated in seconds of the train to be analyzed into the multi-predictor fusion model, and judge the traction energy consumption status of the train to be analyzed according to the output value of the multi-predictor fusion model.

[0012] Preferably, the processing of the train traction energy consumption data accumulated in seconds to form a time series data set on daily, weekly, monthly, and annual time scales includes:

[0013] Sort out the power consumption objects included in the urban rail transit system from macro to micro, and the power consumption objects include three-level research objects: lines, trains, and traction auxiliary power units;

[0014] For the three-level research objects, perform a difference processing on the original train traction energy consumption data accumulated in seconds to obtain daily time series data v d , weekly time series data v w , monthly time series data v m and annual time series data v y .

[0015] Preferably, the construction of multiple time series predictors includes:

[0016] Construct an ARIMA model, verify the stability of the time series, select the model order according to the truncation and trailing of the partial autocorrelation function, verify the fitting residuals of the ARIMA model, and output the prediction results;

[0017] Select the polynomial order, construct a polynomial fitting model, verify the fitting results of the polynomial fitting model, and optimize the polynomial fitting model;

[0018] Decompose the time series into seasonal, trend, and holiday terms, construct a Prophet prediction model, select breakpoints to fit the trend term with a piecewise linear function, fit the seasonal term with a Fourier series, add holiday terms according to the time series scale, integrate each sub-model and verify to obtain the Prophet time series prediction model.

[0019] Preferably, the calibration and verification of each time series predictor, and the use of the Boosting algorithm to adjust the prediction result weights of each time series predictor to form a multi-predictor fusion model, includes:

[0020] Perform calibration and verification on a single time series predictor, and compare and analyze the prediction output results of each single time series predictor;

[0021] Taking the daily sequence data v d as an example, set the same initial weight where n is the number of samples;

[0022] The predictor defined based on the Boosting algorithm is k m (x i ), and the weights of each predictor are α m , obtaining the multi-predictor fusion model C m (x i ), C m-1 (x i ) = α1k1(x i ) + α2k2(x i ) + … + α m-1 k m-1 (x i );

[0023] Calculate the maximum error on the training data set: E m = max|y i - k m (x i )|, and the data set is {(x1, y1), (x2, y2),...(x N , y N ),}; Then calculate the relative error of each sample, Furthermore, calculate the regression error rate, where w is the sample, and then calculate the weights of the weak predictors, Finally, update the sample weight distribution,

[0024] Fuse the time series predictors to obtain the output of the multi-time series predictor, forming a multi-predictor fusion model.

[0025] Preferably, inputting the second-level cumulative train traction energy consumption data of the train to be analyzed into the multi-predictor fusion model, and judging the traction energy consumption state of the train to be analyzed according to the output value of the multi-predictor fusion model, includes:

[0026] Design an abnormality degree evaluation function score u = qa + 5qb + 10qc + 20qd, where a, b, c, and d respectively represent the number of weeks when the actual single consumption of the line deviates from the typical value of single consumption by less than 5%, greater than 5% and less than 10%, greater than 10% and less than 20%, and more than 20%, a + b + c + d = 7, and q is the weight value of a, b, c, and d;

[0027] Input the second-level cumulative train traction energy consumption data of the train to be analyzed into the multi-predictor fusion model, calculate the abnormality degree evaluation function value score u within the unit statistical period according to the output value of the multi-predictor fusion model within the unit statistical period, and use the abnormality degree evaluation function value score u and the set threshold μ dCompare. If the abnormal degree evaluation function value score u >μ d , it is determined that the traction energy consumption of the train to be analyzed is abnormal; otherwise, it is determined that the traction energy consumption of the train to be analyzed is normal.

[0028] As can be seen from the technical solutions provided by the embodiments of the present invention above, the method of the present invention can effectively evaluate the traction energy consumption level of urban rail trains and perform preliminary anomaly detection, providing reference opinions for promoting energy optimization.

[0029] Additional aspects and advantages of the present invention will be given in part in the following description, which will become apparent from the following description, or can be understood through the practice of the present invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0030] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings required for the description of the embodiments will be briefly introduced below. Obviously, the drawings in the following description are only some embodiments of the present invention. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative efforts.

[0031] Figure 1 It is a processing flow chart of a method for detecting abnormal traction energy consumption of trains based on Boosting and multi-predictor fusion provided by an embodiment of the present invention;

[0032] Figure 2 It is a flow chart of a Boosting algorithm provided by an embodiment of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0033] The embodiments of the present invention will be described in detail below. Examples of the embodiments are shown in the drawings, where the same or similar reference numerals represent the same or similar elements or elements with the same or similar functions throughout. The embodiments described below by referring to the drawings are exemplary and are only used to explain the present invention and should not be construed as limiting the present invention.

[0034] Those skilled in the art can understand that, unless specifically stated otherwise, the singular forms "a", "an", "the" and "said" used herein may also include the plural forms. It should be further understood that the term "comprising" used in the specification of the present invention means the presence of the described features, integers, steps, operations, elements and / or components, but does not exclude the presence or addition of one or more other features, integers, steps, operations, elements, components and / or their groups. It should be understood that when we say that an element is "connected" or "coupled" to another element, it can be directly connected or coupled to other elements, or there may also be intermediate elements. In addition, the "connection" or "coupling" used herein may include wireless connection or coupling. The phrase "and / or" used herein includes any unit and all combinations of one or more of the associated listed items.

[0035] Those skilled in the art can understand that, unless otherwise defined, all terms (including technical terms and scientific terms) used herein have the same meaning as the general understanding of those of ordinary skill in the art to which the present invention belongs. It should also be understood that terms such as those defined in a general dictionary should be understood to have a meaning consistent with the meaning in the context of the prior art, and will not be interpreted with an idealized or overly formal meaning unless defined as here.

[0036] For the convenience of understanding the embodiments of the present invention, the following will further explain and illustrate with several specific embodiments in conjunction with the drawings, and each embodiment does not constitute a limitation on the embodiments of the present invention.

[0037] The present invention proposes a method for detecting abnormal train traction energy consumption based on Boosting and multi-predictor fusion, which realizes an accurate description of the normal operating states of various research objects in the urban rail system at different time scales. The proposed abnormal detection method combines the interpretability of the threshold method and the accuracy of the prediction algorithm.

[0038] The processing flow of a method for detecting abnormal train traction energy consumption based on Boosting and multi-predictor fusion provided by an embodiment of the present invention is as Figure 1 shown, and includes the following processing steps:

[0039] Step 1: Process the train traction energy consumption data accumulated in seconds to form time series data sets at time scales such as daily, weekly, monthly and yearly.

[0040] Step 2: Construct multiple time series predictors such as ARIMA (Autoregressive Integrated Moving Average model), polynomial fitting and Prophet.

[0041] Step 3: Verify and validate the single time series predictor, and use the Boosting algorithm to adjust the prediction result weights of each time series predictor to form a multi-predictor fusion model.

[0042] Step 4: Use the output value of the multi-predictor fusion model as the basis for judging the abnormality of the train traction energy consumption, and combine it with the threshold to design an abnormality degree evaluation function. The energy consumption with a certain abnormal score is defined as abnormal energy consumption.

[0043] The operation process of the above Step 1 is as follows:

[0044] Step 101: Sort out the power consumption objects included in the urban rail transit system from macro to micro. The power consumption objects include three-level research objects such as lines, trains, and traction auxiliary power units.

[0045] Step 102: For the three-level research objects, perform a difference operation on the original second-level cumulative traction energy consumption data to obtain daily, weekly, monthly, and annual time series data v d , v w , v m , v y .

[0046] The operation process of the above Step 2 is as follows:

[0047] Step 201: Construct an ARIMA model to verify the stability of the time series.

[0048] Step 202: Select the model order according to the truncation and trailing situation of the partial correlation function. Calculate the autocorrelation function (Autocorrelation Function, ACF) and the partial autocorrelation function (Partial Autocorrelation Function, PACF), and draw the correlation function diagram. Determine the order p of the AR model and the order q of the MA model according to the truncation and trailing properties of the ACF and PACF diagrams.

[0049] Step 203: Verify the fitting residuals of the ARIMA model and output the prediction results.

[0050] Step 204: Select the polynomial order and construct a polynomial fitting model.

[0051] Step 205: Verify the fitting result of the polynomial fitting model and optimize the model.

[0052] Step 206: Decompose the time series into seasonal, trend, and holiday terms, and construct a Prophet prediction model.

[0053] Step 207: Select the segmentation points and use piecewise linear functions to fit the trend terms. Use Fourier series to fit the seasonal terms and add holiday terms according to the time series scale.

[0054] Step 208: Integrate each sub-model and verify to obtain the Prophet time series prediction model.

[0055] The operation process of step 3 is as follows:

[0056] Step 301: Verify and validate a single time series predictor, and compare and analyze the prediction output results of each single time series predictor.

[0057] Step 302: Taking the daily sequence data v d as an example, set the same initial weight where n is the number of samples.

[0058] Step 303: Figure 2 is a flowchart of a Boosting algorithm provided by an embodiment of the present invention. The predictor defined based on the Boosting algorithm is k m (x i ), the weight of each predictor is α m , and finally a multi-predictor fusion model C m (x i ) is obtained. Among them, C m-1 (x i ) = α1k1(x i ) + α2k2(x i ) + … + α m-1 k m-1 (x i ). Single time series predictors such as ARIMA, polynomial fitting, and Prophet are respectively weak predictors defined in the Boosting algorithm.

[0059] Step 304: To calculate the deviation between the output value and the actual value of each time series predictor, first calculate the maximum error on the training data set: E m = max|y i -k m (x i )|, the data set is {(x1, y1), (x2, y2),...(x N , y N ),}; then calculate the relative error of each sample, and then calculate the regression error rate, where w is the sample. Then calculate the weight of the weak predictor, and finally update the sample weight distribution,

[0060] Finally, integrate the output of each time series predictor to obtain the output of the multi-time series predictor, form a multi-predictor fusion model, and use this multi-predictor fusion model as the abnormal judgment basis for the corresponding time scale.

[0061] The operation process of step 4 is as follows:

[0062] Step 401: Design an abnormal degree evaluation function score u = qa + 5qb + 10qc + 20qd, where a, b, c, and d respectively represent the number of weeks in which the actual single consumption of the line deviates from the typical value of single consumption by less than 5%, greater than 5% and less than 10%, greater than 10% and less than 20%, and more than 20% (a + b + c + d = 7), and q is the weight value of a, b, c, and d. The abnormal degree is evaluated based on the amplitude and frequency exceeding the threshold within a unit statistical period.

[0063] Step 402: Input the second-level cumulative train traction energy consumption data of the train to be analyzed into the above multi-predictor fusion model, and use the output value of the multi-predictor fusion model as the basis for judging the abnormality of the train traction energy consumption, and evaluate the energy consumption level of the research object. Calculate the abnormal degree evaluation function value score within a unit statistical period according to the output value of the multi-predictor fusion model within a unit statistical period u and compare the abnormal degree evaluation function value score u with the set threshold μ d . If the abnormal degree evaluation function value score u > μ d , it is determined that the traction energy consumption of the train to be analyzed is abnormal; otherwise, it is determined that the traction energy consumption of the train to be analyzed is normal.

[0064] In summary, the beneficial effects of the train traction energy consumption abnormal detection method based on Boosting and multi-predictor fusion in the embodiments of the present invention are as follows:

[0065] Based on the train traction energy consumption, the analysis level is deepened, the energy consumption level is evaluated from the perspective of various research objects, the analysis breadth and accuracy are improved, and a reasonable mechanism is provided for abnormal traceability.

[0066] The basis for abnormal judgment proposed by the present invention relies on a variety of time series prediction models and Boosting algorithm to optimize the weights. The obtained prediction values are weighted and verified, which conforms to the actual energy consumption level of the engineering site and is convenient for the staff to understand and use.

[0067] Those of ordinary skill in the art can understand that the drawings are only schematic diagrams of an embodiment, and the modules or processes in the drawings are not necessarily essential for implementing the present invention.

[0068] From the description of the above embodiments, those skilled in the art can clearly understand that the present invention can be implemented by means of software plus a necessary general hardware platform. Based on such an understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art can be embodied in the form of a software product. This computer software product can be stored in a storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions for causing a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods described in various embodiments or some parts of the embodiments of the present invention.

[0069] Each embodiment in this specification is described in a progressive manner. For the same or similar parts among the embodiments, reference can be made to each other. The key point of each embodiment is to illustrate the differences from other embodiments. In particular, for the device or system embodiments, since they are basically similar to the method embodiments, the description is relatively simple. For the relevant parts, reference can be made to the partial description of the method embodiments. The device and system embodiments described above are only illustrative. The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they can be located in one place or distributed to multiple network units. Some or all of the modules can be selected according to actual needs to achieve the purpose of the solution of this embodiment. Those of ordinary skill in the art can understand and implement it without creative efforts.

[0070] As mentioned above, the above are only the preferred specific embodiments of the present invention, but the protection scope of the present invention is not limited thereto. Any changes or substitutions that can be easily thought of by those skilled in the art within the technical scope disclosed by the present invention should be covered by the protection scope of the present invention. Therefore, the protection scope of the present invention should be subject to the protection scope of the claims.

Claims

1. A method for detecting abnormal train traction energy consumption based on Boosting and multi-predictor fusion, characterized in that Including: Processing the second-level cumulative train traction energy consumption data to form a time series data set at daily, weekly, monthly, and annual time scales; Constructing multiple time series predictors, verifying and validating each time series predictor, and using the Boosting algorithm to adjust the prediction result weights of each time series predictor to form a multi-predictor fusion model; Inputting the second-level cumulative train traction energy consumption data of the train to be analyzed into the multi-predictor fusion model, and judging the traction energy consumption status of the train to be analyzed according to the output value of the multi-predictor fusion model; The constructing of multiple time series predictors includes: Constructing an ARIMA model, verifying the stability of the time series, selecting the model order according to the truncation and trailing of the partial autocorrelation function, verifying the fitting residuals of the ARIMA model, and outputting the prediction result; Selecting the polynomial order, constructing a polynomial fitting model, verifying the fitting result of the polynomial fitting model, and optimizing the polynomial fitting model; Decomposing the time series into seasonal, trend, and holiday terms, constructing a Prophet prediction model, selecting break points to fit the trend term with a piecewise linear function, fitting the seasonal term with a Fourier series, adding holiday terms according to the time series scale, integrating each sub-model and verifying to obtain a Prophet time series prediction model; The verifying and validating of each time series predictor and using the Boosting algorithm to adjust the prediction result weights of each time series predictor to form a multi-predictor fusion model includes: Verifying and validating a single time series predictor, and comparatively analyzing the prediction output results of each single time series predictor; Taking the daily sequence data v d as an example, the same initial weights are set where n is the number of samples; The predictor defined based on the Boosting algorithm is k m (x i ), and the weights of each predictor are α m , obtaining the multi-predictor fusion model C m (x i ), C m-1 (x i ) = α1k1(x i ) + α2k2(x i ) + … + α m-1 k m-1 (x i ); Calculate the maximum error on the training dataset: E m = max|y i - k m (x i )|, where the dataset is {(x1, y1), (x2, y2),...(x N , y N )}; then calculate the relative error of each sample, and then calculate the regression error rate, where w is the sample, and then calculate the weight of the weak predictor, finally update the sample weight distribution, Fusing the outputs of each time series predictor to obtain the output of the multi-time series predictor, and forming a multi-predictor fusion model.

2. The method according to claim 1, characterized in that, The processing of the second-level cumulative train traction energy consumption data to form a time series data set at daily, weekly, monthly, and annual time scales includes: Sorting out the power consumption objects included in the urban rail transit system from macro to micro, and the power consumption objects include three-level research objects: lines, trains, and traction auxiliary power units; For the three-level research object, the difference processing is performed on the originally second-accumulated train traction energy consumption data to obtain the daily time series data v d , the weekly time series data v w , the monthly time series data v m and the annual time series data v y .

3. The method according to claim 1, wherein The inputting the second-level cumulative train traction energy consumption data of the train to be analyzed into the multi-predictor fusion model and judging the traction energy consumption status of the train to be analyzed according to the output value of the multi-predictor fusion model includes: Design anomaly degree evaluation function score u = qa + 5qb + 10qc + 20qd, where a, b, c, and d respectively represent the number of weeks in which the actual single consumption of the circuit deviates from the typical value of single consumption by less than 5%, more than 5% and less than 10%, more than 10% and less than 20%, and more than 20%. a + b + c + d = 7, and q is the weight value of a, b, c, and d; Input the train traction energy consumption data accumulated in seconds of the train to be analyzed into the multi-predictor fusion model, and calculate the abnormal degree evaluation function value score within the unit statistical period according to the output value of the multi-predictor fusion model within the unit statistical period u , and the abnormal degree evaluation function value score u is compared with the set threshold μ d . If the abnormal degree evaluation function value score u > μ d , it is determined that the traction energy consumption of the train to be analyzed is abnormal; otherwise, it is determined that the traction energy consumption of the train to be analyzed is normal.

Images