A Method for Constructing an Intelligent Algorithm Platform for Urban Rail Transit Data

Abstract

The present invention discloses a method for constructing an intelligent algorithm platform for urban rail transit data. The intelligent algorithm platform provides services for algorithm models, and the services of the algorithm models include: S1, obtaining the urban rail transit data to be predicted and corresponding fields, and performing auxiliary calibration and cleaning on the urban rail transit data; S2, performing feature engineering on the cleaned urban rail transit data to obtain a feature training set, and training various algorithm models according to the feature training set; S3, extracting various algorithm models related to the fields from the algorithm platform according to the corresponding fields to form a set, and performing algorithm loading on the set composed of the various algorithm models to obtain an algorithm integration model; S4, inputting the cleaned urban rail transit data into the algorithm integration model to obtain prediction result data. Common algorithms are provided and pre-set in the system. One can directly select this algorithm for training operations without paying attention to model development.

Description

Technical Field

[0001] The present invention relates to the field of public transportation, and particularly to a method for constructing an intelligent algorithm platform for urban rail transit data. Background Art

[0002] Currently, the urban rail transit industry is developing rapidly, and the amount of information data is constantly expanding. Data processing has evolved from the original single-data processing and multi-data processing to the current era of big data processing.

[0003] With the rise of big data technology, the advantages of data are becoming more and more significant, and the scope of influence is becoming wider and wider. How to make good use of this data and extract valuable information from the vast amount of data is the core work of algorithm engineers. However, in the work of algorithm engineers, due to various engineering requirements, it is often necessary to establish engineering projects for different projects to provide different computing environments for the algorithms. This makes the focus of algorithm engineers' work often need to be placed on engineering rather than algorithms. As a result, algorithm engineers have a lot of repetitive work, and their work efficiency is greatly affected. Therefore, how to separate algorithms from engineering, so that algorithm engineers can focus their energy on algorithms rather than engineering, thereby improving the work efficiency of algorithm engineers and avoiding unnecessary repetitive work, has become an urgent problem to be solved in this field.

[0004] As an important part of the closed-loop chain of the transportation platform, work efficiency and user experience are the core competitiveness of transportation services. With the increase in the number of passengers, the increase in rail lines, and the complexity of traffic scenarios, various algorithms in the rail transit scenario are also facing increasing challenges under the goals of being faster (algorithms need to be iterated and launched quickly), better (the business increasingly relies on machine learning algorithms to produce positive effects), and more accurate (various predictions of algorithms such as passenger flow need to accurately approximate the true value). Summary of the Invention

[0005] The purpose of the present invention is to provide a method for constructing an intelligent algorithm platform for urban rail transit data, which provides prediction services for multiple machine learning algorithm models for users of urban rail transit. The data to be predicted can be directly from other platforms in the urban rail transit system, and the prediction results can be directly applied to other platforms in urban rail transit, facilitating user use. It is used to solve the problem that existing algorithm engineers need to perform cumbersome engineering development and cannot focus their limited energy on the iteration of algorithm strategies.

[0006] A method for constructing an intelligent algorithm platform for urban rail transit data, wherein the intelligent algorithm platform provides services for algorithm models, and the services for algorithm models include:

[0007] S1. Obtain the urban rail transit data to be predicted and the corresponding fields, and perform auxiliary calibration and cleaning on the urban rail transit data to obtain the cleaned urban rail transit data and the corresponding fields;

[0008] S2. Perform feature engineering on the cleaned urban rail transit data to obtain a feature training set, and train various algorithm models according to the feature training set. The feature engineering includes statistical feature engineering, graph feature engineering, and deep feature engineering;

[0009] S3. Extract various algorithm models related to the fields from the algorithm platform according to the corresponding fields to form a set, and perform algorithm loading on the set of various algorithm models. Algorithm loading is to load the algorithm models in a specific order set to cooperate to form an algorithm integration model;

[0010] S4. Input the cleaned urban rail transit data into the algorithm integration model to obtain prediction result data.

[0011] Further, the algorithm integration model includes algorithm models for multi-level processing, and the multi-level processing is performed in the order of priority: set the priority of the first-level processing algorithm model >... > the priority of the N-level processing algorithm model, where N is an integer greater than or equal to 1.

[0012] Further, each level of processing algorithm model includes multiple optimal algorithm models, and each optimal algorithm model is obtained by optimally evaluating multiple algorithm models in the corresponding algorithm model class by calling a pre-trained evaluation model. The algorithm model class is multiple algorithm models that use different methods to process data to obtain the same or substantially the same results. The processing includes dimensionality reduction, association, clustering, classification, and regression. The evaluation model is used to evaluate the ability of the algorithm model to process and predict data.

[0013] Further, the processing is classification, and the discrimination model specifically includes the following steps:

[0014] S301. Obtain the test data set corresponding to the urban rail transit data to be predicted from the preset sample database;

[0015] S302. Input the test data set into multiple algorithm models in the algorithm model class to obtain multiple test result sets, and establish a confusion matrix for the test result sets. Calculate the corresponding evaluation indicators through the confusion matrix. The evaluation indicators include accuracy, precision, recall rate, ROC curve, AUC, and F1 harmonic mean;

[0016] S303. The algorithm model with the optimal evaluation indicator is the optimal algorithm model.

[0017] Furthermore, there are at least two of the multiple optimal algorithm models, and the multiple optimal algorithm models are connected in parallel or in series to form the processing of the corresponding stage: the multiple optimal algorithm models in the first-level processing are connected in parallel or in series to form the first-level stage processing; the multiple optimal algorithm models in the N-level processing are connected in parallel or in series to form the N-level stage processing, where N is an integer greater than or equal to 1.

[0018] Furthermore, the algorithm platform includes 3D models and images, operation and maintenance efficiency verification, and train operation index verification:

[0019] 3D models and images include: image stitching algorithm, OCR algorithm, GANS image generation;

[0020] Operation and maintenance efficiency verification includes: model drift and update algorithm, metric measurement algorithm, anomaly detection algorithm;

[0021] Train operation index verification includes: time-series sequence classification prediction algorithm, time-series sequence regression prediction algorithm, multi-objective optimization algorithm, clustering algorithm.

[0022] Furthermore, the urban rail transit data is the inbound or outbound passenger flow data of urban rail transit at time t;

[0023] The corresponding field is: the value of the inbound or outbound passenger flow of urban rail transit at the window of time t + 1;

[0024] The algorithm set: the first-level processing is a clustering algorithm model, and the second-level processing is a long short-term memory neural network model;

[0025] Perform clustering model processing on the spatial distribution characteristics to extract the line characteristics, station characteristics, and section passenger flow characteristics of different subway stations. These three characteristics are the spatial characteristics; perform clustering model processing on the time distribution characteristics, extract the passenger flow distribution characteristics of each day in a week, then divide the passenger flow distribution characteristics of each day into multiple time periods, and extract the passenger flow distribution characteristics of each time period. These two distribution characteristics are the time characteristics;

[0026] Input the urban rail transit passenger flow data combined with the time characteristics and spatial characteristics into the long short-term memory neural network model to obtain the predicted value of the inbound or outbound passenger flow of urban rail transit at the window of time t + 1.

[0027] Furthermore, the urban rail transit data is the image of the inspection parts of rail transit vehicles:

[0028] The corresponding field is: the suspected fault diagram of parts;

[0029] The algorithm set: the first-level processing is a classification algorithm model, and the second-level processing is a fault detection model;

[0030] The images of the inspection parts of the rail transit vehicle are processed by a classification model to obtain the rail transit vehicle images labeled with tags classified according to structure and function. The test samples of the inspection parts are input into the fault detection model for detection to obtain a set of suspected fault images.

[0031] Further, the urban rail transit data is the integrated monitoring alarm data and parameter configuration of rail transit:

[0032] The corresponding fields are: alarm text messages;

[0033] The algorithm set: the first-level processing is a classification algorithm model, the second-level processing is a purification algorithm model, and the third-level processing algorithm is a decision-making algorithm model;

[0034] The collected alarm data and equipment and parameter configuration are used as inputs for processing by the classification algorithm model. The alarm data belonging to the same equipment or monitoring object are classified and used as inputs for the next-level purification algorithm model. The purification algorithm model processes and refines the alarm data according to the level and time sequence, generates concise and pure data, and then uses it as the input for the next-level decision-making algorithm model. The decision-making algorithm model generates the content of the text message through the concise and pure data, including the determination of alarm content, suggestions, handling measures, request for feedback content, and recipients, and finally generates an output message.

[0035] An algorithm service device includes:

[0036] One or more memories,

[0037] One or more processors,

[0038] Multiple modules, the modules are stored in the memory and executed by the processor. The modules include:

[0039] A data reception calibration module, used to obtain the urban rail transit data to be predicted and the corresponding fields, and perform auxiliary calibration and cleaning on the urban rail transit data to obtain the cleaned urban rail transit data and the corresponding fields;

[0040] A feature extraction module, used to perform feature engineering on the cleaned urban rail transit data to obtain a feature training set, and train various algorithm models according to the feature training set. The feature engineering includes statistical feature engineering, graph feature engineering, and deep feature engineering;

[0041] An algorithm extraction and loading module, used to extract various algorithm models related to the fields from the algorithm platform according to the corresponding fields to form a set, and perform algorithm loading on the set composed of various algorithm models. The algorithm loading is to load the algorithm models according to a specific order set to form an algorithm integration model;

[0042] A prediction result return module, which is used to input the cleaned urban rail transit data into the algorithm integration model to obtain prediction result data.

[0043] Currently, whether in the field of traditional machine learning or the current popular field of deep learning, supervised learning based on training samples with clear labels or results is still a major model training method. Especially in the field of deep learning, more data is needed to improve the model effect. Currently, there are already some large-scale public datasets, such as ImageNet, COCO and other datasets in the field of images that can provide references. For most enterprise developers, we need to use the actual business data in the professional field to customize the AI model application service to ensure that it can be better applied in the business. Therefore, the collection and annotation of business scenario data are essential important links in the actual AI model development process.

[0044] The present invention makes data prediction processing more accurate, consistent, complete and efficient by establishing an intelligent algorithm platform, including providing integrated data access, cleaning and transformation, model construction and model application services.

[0045] The intelligent algorithm platform includes the service of algorithm models and the training of algorithm models. The training of algorithm models is for backend algorithm engineers, providing multiple machine learning algorithms for users to dynamically select and use, automatically applying complex data calculations to the analysis process of massive data, covering the whole process of algorithm engineers' research, development, online, and evaluation of algorithm effects: data processing, feature engineering, model training, model evaluation, model release, online prediction and effect evaluation.

[0046] The service of algorithm models is for front-end users, providing services for machine learning algorithm models; obtaining the prediction data uploaded by service users; calling the corresponding machine learning algorithm model according to the prediction request to execute the prediction service; and sending the prediction result to the service users.

[0047] The beneficial effects of the present invention are:

[0048] 1. For different urban rail transit data, matching algorithm models are provided, and common algorithms are provided and pre-set in the system. One can directly select this algorithm for training tasks without paying attention to model development.

[0049] 2. Without coding, directly use the pre-set algorithm to train the model. Support common frameworks (such as TensorFlow, PyTorch, etc.), without the need for users to configure the algorithm framework themselves, saving development costs.

[0050] 3. The intelligent algorithm middle platform can carry various algorithm capabilities, greatly promoting the efficiency of data science, algorithm modeling and training, model version control, model deployment and other links in the algorithm development and application processes, and enabling the algorithm capabilities to exert their due value within this framework. BRIEF DESCRIPTION OF THE DRAWINGS

[0051] Figure 1 Schematic diagram of the service method of the algorithm model of the present invention;

[0052] Figure 2 Schematic diagram of the service process of the algorithm model of the present invention;

[0053] Figure 3 Schematic diagram of the algorithm service device of the present invention;

[0054] Figure 4 Schematic diagram of the three major categories of toolboxes for feature engineering of the present invention; DETAILED DESCRIPTION OF THE EMBODIMENTS

[0055] The present invention will be further described in detail below in conjunction with the embodiments and the drawings, but the embodiments of the present invention are not limited thereto.

[0056] In the description of the present invention, it should be noted that the orientation or positional relationship indicated by the terms "center", "upper", "lower", "left", "right", "vertical", "longitudinal", "lateral", "horizontal", "inner", "outer", "front", "rear", "top", "bottom", etc. is based on the orientation or positional relationship shown in the drawings, or the orientation or positional relationship in which the product of the present invention is habitually placed during use. It is only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore should not be construed as a limitation to the present invention.

[0057] In the description of the present invention, it should also be noted that unless otherwise clearly specified and defined, the terms "set", "provided with", "installed", "connected", "connected to" should be understood in a broad sense. For example, it can be a fixed connection, a detachable connection, or an integral connection; it can be a mechanical connection or an electrical connection; it can be directly connected or indirectly connected through an intermediate medium, and it can be the communication inside two elements. For those of ordinary skill in the art, the specific meanings of the above terms in the present invention can be understood according to specific circumstances.

[0058] The intelligent algorithm platform is a one-stop data modeling and analysis platform built based on a cloud platform, big data computing services, a deep machine learning framework, and related software and hardware support. The functions of the algorithm platform can include several functional modules such as data import and export, data preprocessing, data statistical analysis, data mining, model development, model training, model release, and model management.

[0059] The system architecture of the intelligent algorithm platform can be divided into the service of algorithm models and the training of algorithm models. The service of algorithm models can provide unified prediction services, facilitating users to manage prediction services and providing basic management and scheduling functions such as creation, expansion, contraction, and deletion.

[0060] The algorithm model training provides binary classification evaluation, multi-classification evaluation, regression model evaluation, and clustering model evaluation, meeting the evaluation and analysis of classification, regression, and clustering algorithm models. It supports running on CPU and GPU of model services. It supports the seamless docking of PMML (Predictive Model Markup Language) models, TensorFlow models, and custom models. It supports the secure release of new models through mechanisms such as deployment mechanism (flow splitting by percentage), version control, and quick rollback, and provides intelligent operation and maintenance monitoring charts.

[0061] The algorithm model training also includes the human-in-the-loop stage, which includes randomly extracting a small number of samples from the validation dataset for human-in-the-loop annotation and forming sample training, training and feeding back the small number of already annotated samples to the model, and adjusting the model according to the feedback.

[0062] The algorithm model training also includes managing the running algorithm types and corresponding versions. By pre-configuring the startup version and the list of historical available versions for different types of algorithms, and monitoring the running status of the algorithm in real time during operation, if an exception occurs, it will roll back to the previous stable version.

[0063] The overall architecture of the service of algorithm models can be roughly divided into two layers, the API layer and the execution layer. API layer: There are two types of online prediction service APIs: the prediction service API and the prediction request API. According to the different characteristics and requirements of each type, there are different designs in the architecture. Prediction service API: Responsible for the creation, deployment, deletion, modification, etc. of the prediction service. Prediction request API: Processes the prediction requests sent by the client and returns the prediction results. Execution layer: All computing resources are managed through the cloud management application. Each node in the cluster of the cloud management application can be a cloud server or a physical machine.

[0064] In one example, the algorithm platform includes a prediction platform, which facilitates users to manage prediction services and provides basic management and scheduling functions such as creation, expansion, contraction, and deletion. The prediction platform has a general prediction runtime library: providing a unified prediction API service. The prediction platform supports model formats of various frameworks, supports running on CPU and GPU, and supports various model accelerations.

[0065] A method for constructing an intelligent algorithm platform for urban rail transit data, where the intelligent algorithm platform provides services for algorithm models, and the services of the algorithm models include:

[0066] S1. Obtain the urban rail transit data to be predicted and the corresponding fields, and perform auxiliary calibration and cleaning on the urban rail transit data to obtain the cleaned urban rail transit data and the corresponding fields;

[0067] S2. Perform feature engineering on the cleaned urban rail transit data to obtain a feature training set, and train various algorithm models according to the feature training set. The feature engineering includes statistical feature engineering, graph feature engineering, and deep feature engineering;

[0068] S3. Extract various algorithm models related to the fields from the algorithm platform according to the corresponding fields to form a set, and perform algorithm loading on the set composed of various algorithm models. Algorithm loading is to load the algorithm models in a specific order set to cooperate to form an algorithm integration model;

[0069] S4. Input the cleaned urban rail transit data into the algorithm integration model to obtain prediction result data.

[0070] Auxiliary calibration, also known as the active learning algorithm, usually consists of two modules:

[0071] The first module is the anomaly detection module. This module finds a series of data in the urban rail transit data that is least similar to the pattern of normal data as the data of possible anomaly samples;

[0072] The second module is the anomaly retrieval module. This module retrieves the fault samples of the urban rail transit data by means of an efficient time series retrieval algorithm and pushes the anomaly samples to the operation and maintenance personnel for calibration.

[0073] The realization of this function solves the situation of wasting a large amount of manpower in the process of making data labels in the development of rail transit intelligent applications, transforms the existing manual annotation into automatic annotation, and greatly improves the development efficiency of rail transit intelligent applications.

[0074] Feature engineering

[0075] Due to the characteristics of the rail transit industry itself, the data has the characteristics of being structured, highly coupled, and strong in principle. In the process of implementing traditional feature engineering methods, a large amount of valuable data is often generated for direct use by other business systems. Therefore, the functional modules of the three major toolboxes of feature engineering (statistical features, graph features, and deep features).

[0076] Feature construction refers to the artificial construction of new features from the original data. In this process, developers identify some features with rail transit physics or business logic significance from the original data. Generally, mixed attributes or combined attributes are used to create new features, or the original features are decomposed or segmented to create new features.

[0077] The object of feature extraction is the raw data, and its purpose is to automatically construct new features and transform the original features into a set of variables with obvious physical significance, statistical significance or kernel. For example, by transforming the feature values to reduce the number of values of a certain feature in the original data.

[0078] Feature selection aims to select a subset of the most statistically significant features from the feature set to achieve the effect of dimensionality reduction.

[0079] Furthermore, the algorithm integration model includes algorithm models for multi-level processing, and the multi-level processing is carried out in the order of priority: set the priority of the first-level processing algorithm model >... > the priority of the N-level processing algorithm model, where N is an integer greater than or equal to 1.

[0080] Furthermore, each level of processing algorithm model includes multiple optimal algorithm models, and each optimal algorithm model is obtained by optimally evaluating multiple algorithm models in the corresponding algorithm model class by calling a pre-trained evaluation model. The algorithm model class is multiple algorithm models that use different methods to process data to obtain the same or substantially the same results. The processing includes dimensionality reduction, association, clustering, classification and regression. The evaluation model is used to evaluate the ability of the algorithm model to process and predict data.

[0081] Furthermore, the processing is classification, and the discrimination model specifically includes the following steps:

[0082] S301. Obtain a test data set corresponding to the urban rail transit data to be predicted from a preset sample database;

[0083] S302. Input the test data set into multiple algorithm models in the algorithm model class to obtain multiple test result sets, establish a confusion matrix for the test result sets, and calculate corresponding evaluation indicators through the confusion matrix. The evaluation indicators include accuracy, precision, recall rate, ROC curve, AUC, and F1 harmonic mean;

[0084] The F1 harmonic mean is the harmonic value of precision and recall. Since it is closer to the smaller of the two numbers, the F value is maximized when precision and recall are close. Many evaluation metrics for recommendation systems use the F value. 2 / F1 = 1 / Precision + 1 / Recall, where Precision is precision and Recall is recall.

[0085] For classification problems, the evaluation metrics used are: Accuracy, Confusion Matrix, Precision (Precision Rate), Recall (Recall Rate), FβScore, and AUCKS.

[0086] For regression problems, the evaluation metrics used are: Mean Absolute Error, Mean Squared Error, Root Mean Squared Error, and Coefficient of Determination.

[0087] For clustering problems, the evaluation metrics used are: Rand Index, Mutual Information, and Silhouette Coefficient.

[0088] S303. The algorithm model with the optimal evaluation metrics is the optimal algorithm model.

[0089] Furthermore, there are at least two of the multiple optimal algorithm models. The multiple optimal algorithm models are connected in parallel or in series to form the processing of the corresponding stage: the multiple optimal algorithm models in the first-level processing are connected in parallel or in series to form the first-level stage processing; the multiple optimal algorithm models in the N-level processing are connected in parallel or in series to form the N-level stage processing, where N is an integer greater than or equal to 1.

[0090] Furthermore, the algorithm platform includes 3D models and image categories, operation and maintenance efficiency verification categories, and driving index verification categories:

[0091] The 3D models and image categories include: image stitching algorithms, OCR algorithms, and GANS image generation;

[0092] The operation and maintenance efficiency verification categories include: model drift and update algorithms, metric measurement algorithms, and anomaly detection algorithms;

[0093] The driving index verification categories include: time-series sequence classification prediction algorithms, time-series sequence regression prediction algorithms, multi-objective optimization algorithms, and clustering algorithms.

[0094] Example 1

[0095] The urban rail transit data is the inbound or outbound passenger flow data of urban rail transit at time t;

[0096] The corresponding field is: the value of the passenger flow entering or leaving the urban rail transit in the time period t+1;

[0097] The algorithm set: the primary processing is a clustering algorithm model, and the secondary processing is a long short-term memory neural network model;

[0098] Perform clustering model processing on the spatial distribution characteristics to extract the line characteristics, station characteristics, and cross-section passenger flow characteristics of different subway stations. These three characteristics are the spatial characteristics; perform clustering model processing on the time distribution characteristics, extract the passenger flow distribution characteristics of each day in a week, then divide the passenger flow distribution characteristics of each day into multiple time periods, and extract the passenger flow distribution characteristics of each time period. These two distribution characteristics are the time characteristics;

[0099] Input the urban rail transit passenger flow data combined with the time characteristics and spatial characteristics into the long short-term memory neural network model to obtain the predicted value of the passenger flow entering or leaving the urban rail transit in the window of the time period t+1.

[0100] Embodiment 2

[0101] The urban rail transit data is the image of the inspection part of the rail transit vehicle:

[0102] The corresponding field is: the suspected fault diagram of the parts;

[0103] The algorithm set: the primary processing is a classification algorithm model, and the secondary processing is a fault detection model;

[0104] Perform classification model processing on the image of the inspection part of the rail transit vehicle to obtain the labeled rail transit vehicle image classified according to the structure and function, and input the test samples of the inspection part into the fault detection model for detection to obtain the set of suspected fault images.

[0105] Embodiment 3

[0106] The urban rail transit data is the integrated monitoring alarm data and parameter configuration of the rail transit:

[0107] The corresponding field is: the alarm text message;

[0108] The algorithm set: the primary processing is a classification algorithm model, the secondary processing is a purification algorithm model, and the tertiary processing is a decision algorithm model;

[0109] The collected alarm data, devices, and parameter configurations are used as inputs for processing by a classification algorithm model. The alarm data belonging to the same device or monitoring object is classified and used as the input for the next-level purification algorithm model. The purification algorithm model processes and refines the alarm data according to levels, chronological order, and duplicate judgment, etc., to generate concise and pure data, which is then used as the input for the next-level decision-making algorithm model. The decision-making algorithm model generates the content of the short message through the concise and pure data, including the determination of the alarm content, suggestions, handling measures, request for feedback content, and recipients, and finally generates the output message.

[0110] Embodiment 4

[0111] The purpose of this embodiment is to provide an algorithm service device, including:

[0112] One or more memories,

[0113] One or more processors,

[0114] Multiple modules, which are stored in the memory and executed by the processor. The modules include:

[0115] A data reception calibration module, which is used to obtain the urban rail transit data to be predicted and the corresponding fields, and perform auxiliary calibration and cleaning on the urban rail transit data to obtain the cleaned urban rail transit data and the corresponding fields;

[0116] A feature extraction module, which is used to perform feature engineering on the cleaned urban rail transit data to obtain a feature training set, and train various algorithm models according to the feature training set. The feature engineering includes statistical feature engineering, graph feature engineering, and deep feature engineering;

[0117] An algorithm extraction and loading module, which is used to extract various algorithm models related to the fields from the algorithm platform according to the corresponding fields to form a set, and perform algorithm loading on the set composed of the various algorithm models. The algorithm loading is to load the algorithm models in a specific order set to cooperate to form an algorithm integration model;

[0118] A prediction result return module, which is used to input the cleaned urban rail transit data into the algorithm integration model to obtain prediction result data.

[0119] As mentioned above, it is only a preferred embodiment of the present invention, and does not impose any form of limitation on the present invention. According to the technical essence of the present invention, within the spirit and principle of the present invention, any simple modification, equivalent replacement, and improvement made to the above embodiments still fall within the protection scope of the technical solution of the present invention.

Claims

1. A method for constructing an intelligent algorithm platform for urban rail transit data, characterized in that, the intelligent algorithm platform provides services for algorithm models, and the services of the algorithm models include: S1. Obtain the urban rail transit data to be predicted and the corresponding fields, and perform auxiliary calibration and cleaning on the urban rail transit data to obtain the cleaned urban rail transit data and the corresponding fields; S2. Perform feature engineering on the cleaned urban rail transit data to obtain a feature training set, and train various algorithm models according to the feature training set. The feature engineering includes statistical feature engineering, graph feature engineering, and deep feature engineering; S3. Extract various algorithm models related to the fields from the algorithm platform according to the corresponding fields to form a set, and perform algorithm loading on the set of various algorithm models. Algorithm loading is to load the algorithm models in a specific order set to form an algorithm integration model; The algorithm integration model includes algorithm models for multi-level processing, and the multi-level processing is performed in the order of priority: Set the priority of the first-level processing algorithm model >... > the priority of the N-level processing algorithm model, where N is an integer greater than or equal to 1; Each level of processing algorithm model includes multiple optimal algorithm models. Each optimal algorithm model is obtained by optimally evaluating multiple algorithm models in the corresponding algorithm model class by calling a pre-trained evaluation model. The algorithm model class is multiple algorithm models that use different methods to process data to obtain the same or substantially the same results. The processing includes dimensionality reduction, association, clustering, classification, and regression. The evaluation model is used to evaluate the ability of the algorithm model to process and predict data; When the processing is classification, obtaining the optimal algorithm model specifically includes the following steps: S301. Obtain a test data set corresponding to the urban rail transit data to be predicted from a preset sample database; S302. Input the test data set into multiple algorithm models in the algorithm model class to obtain multiple test result sets, establish a confusion matrix for the test result sets, and calculate the corresponding evaluation indicators through the confusion matrix. The evaluation indicators include accuracy, precision, recall rate, ROC curve, AUC, and F1 harmonic mean; S303. The algorithm model with the optimal evaluation indicator is the optimal algorithm model; There are at least two of the multiple optimal algorithm models, and the multiple optimal algorithm models are connected in parallel or in series to form the processing of the corresponding stage: the multiple optimal algorithm models in the first-level processing are connected in parallel or in series to form the first-level stage processing; the multiple optimal algorithm models in the N-level processing are connected in parallel or in series to form the N-level stage processing, where N is an integer greater than or equal to 1; S4. Input the cleaned urban rail transit data into the algorithm integration model to obtain predicted result data.

2. The method for constructing an intelligent algorithm platform for urban rail transit data according to claim 1, characterized in that, the algorithm platform includes 3D model and image category, operation and maintenance efficiency verification category, and train operation index verification category: The 3D model and image category includes: image stitching algorithm, OCR algorithm, GANS image generation algorithm; The operation and maintenance efficiency verification category includes: model drift and update algorithms, metric measurement algorithms, and anomaly detection algorithms; The train operation index verification category includes: time-series sequence classification prediction algorithms, time-series sequence regression prediction algorithms, multi-objective optimization algorithms, and clustering algorithms.

3. A method for constructing an intelligent algorithm platform for urban rail transit data according to claim 1, characterized in that, the urban rail transit data is the inbound or outbound passenger flow data of urban rail transit at time t; the corresponding field is: the value of the inbound or outbound passenger flow of urban rail transit in the window at time t + 1; the algorithm integration model: the primary processing is a clustering algorithm model, and the secondary processing is a long short-term memory neural network model; Perform clustering model processing on the spatial distribution characteristics to extract the line characteristics, station characteristics, and cross-section passenger flow characteristics of different subway stations. These three characteristics are the spatial characteristics; Perform clustering model processing on the time distribution characteristics. Extract the passenger flow distribution characteristics of each day in a week, then divide the passenger flow distribution characteristics of each day into multiple time periods, and extract the passenger flow distribution characteristics of each time period. These two distribution characteristics are the time characteristics; Input the urban rail transit passenger flow data combined with the time characteristics and spatial characteristics into the long short-term memory neural network model to obtain the predicted value of the inbound or outbound passenger flow of urban rail transit in the window at time t + 1.

4. A method for constructing an intelligent algorithm platform for urban rail transit data according to claim 1, characterized in that, the urban rail transit data is the image of the inspection part of the rail transit vehicle: the corresponding field is: the suspected fault diagram of the parts; the algorithm integration model: the primary processing is a classification algorithm model, and the secondary processing is a fault detection model; Perform classification model processing on the image of the inspection part of the rail transit vehicle to obtain a labeled rail transit vehicle image classified according to structure and function. Input the test samples of the inspection part into the fault detection model for detection to obtain a set of suspected fault images.

5. A method for constructing an intelligent algorithm platform for urban rail transit data according to claim 1, characterized in that, the urban rail transit data is the integrated monitoring alarm data and parameter configuration of the rail transit: the corresponding field is: alarm text messages; the algorithm integration model: the primary processing is a classification algorithm model, the secondary processing is a purification algorithm model, and the tertiary processing is a decision-making algorithm model; Use the collected alarm data and equipment and parameter configurations as inputs for classification algorithm model processing, classify the alarm data belonging to the same equipment or monitoring object, and use it as the input for the next-level purification algorithm model; the purification algorithm model processes and refines the alarm data according to the level and time sequence to generate concise and pure data, and then use it as the input for the next-level decision-making algorithm model; the decision-making algorithm model generates the content of the text message through the concise and pure data, including the determination of the alarm content, suggestions, handling measures, request for feedback content, and sending object, and finally generates the output message.

6. An algorithm service device, characterized in that, including: one or more memories, one or more processors, Multiple modules for implementing a method for constructing an intelligent algorithm platform for urban rail transit data as described in claim 1, the modules being stored in a memory and executed by a processor, the modules including: A data reception calibration module for obtaining urban rail transit data to be predicted and corresponding fields, and performing auxiliary calibration cleaning on the urban rail transit data to obtain the cleaned urban rail transit data and corresponding fields; A feature extraction module for performing feature engineering on the cleaned urban rail transit data to obtain a feature training set, and training various algorithm models according to the feature training set, the feature engineering including statistical feature engineering, graph feature engineering, and deep feature engineering; An algorithm extraction and loading module for extracting various algorithm models related to the fields from the algorithm platform according to the corresponding fields to form a set, and performing algorithm loading on the set composed of the various algorithm models, the algorithm loading being to load the algorithm models in a specific order set to cooperate to form an algorithm integration model; A prediction result return module for inputting the cleaned urban rail transit data into the algorithm integration model to obtain prediction result data.

Images