A novel ETC issuance amount prediction method, system, device and storage medium

Abstract

The application discloses a novel ETC issuance amount prediction method, system, device and a storage medium, through collecting multi-dimensional data and constructing ETC issuance amount characteristics, the relationship between the ETC issuance amount and various factors is comprehensively and accurately reflected, a scientific data foundation is laid for a prediction model, and the problem of lacking scientificity due to incomplete data is overcome;Secondly, the LightGBM model is selected, which has strong nonlinear fitting capability, can automatically learn complex patterns and feature interactions, accurately describes the relationship to improve the prediction accuracy, the model uses multiple technologies to reduce the amount of calculation and memory occupation, can quickly train large-scale data, provide results in time and reduce the hardware requirements;In addition, based on the model training and verification of the constructed characteristics, the model adjusts the parameters to minimize the error in the training, evaluates the performance and finds problems, so as to continuously optimize the model, make it better adapt to actual data, further reduce the prediction deviation and improve the prediction accuracy.

Description

Technical Field

[0001] This invention relates to the field of ETC issuance prediction technology, and in particular to a novel ETC issuance prediction method, system, device and storage medium. Background Technology

[0002] With the continuous improvement of transportation infrastructure and the increasing level of intelligence, Electronic Toll Collection (ETC) systems have been widely used in highway toll collection, parking lot toll collection, and other fields. The popularization of ETC technology has not only significantly improved vehicle traffic efficiency and reduced congestion at toll stations, but also reduced the cost of manual toll collection and human error, thereby improving the overall level of traffic operation and management. Therefore, accurately predicting the issuance volume of ETC is crucial for ETC issuing institutions, traffic management departments, and related enterprises to formulate scientific and reasonable issuance plans, resource allocation strategies, and market promotion plans.

[0003] Current technologies for predicting ETC issuance volume mainly rely on empirical judgment and simple statistical methods, such as linear extrapolation based on issuance volume data from previous years, or estimation based on the number of potential users obtained from market research. However, these methods have many limitations: empirical judgment is easily affected by personal subjective factors and lacks scientific rigor and accuracy; simple statistical methods often ignore the complex relationship between ETC issuance volume and various influencing factors, such as the growth of car ownership, fluctuations in traffic flow, and market competition, leading to a large deviation between the predicted results and the actual situation. Summary of the Invention

[0004] In view of this, the present invention proposes a novel method, system, device and storage medium for predicting ETC issuance volume, which can effectively solve the defects of existing technologies, such as lack of scientificity and accuracy, and large deviation between prediction results and actual situation.

[0005] The technical solution of this invention is implemented as follows:

[0006] A novel method for predicting ETC issuance volume, specifically including:

[0007] Obtain forecasts of ETC issuance demand;

[0008] Collect multidimensional data and construct ETC issuance characteristics based on the multidimensional data;

[0009] Based on the demand for ETC issuance and the characteristics of ETC issuance, the LightGBM model was selected as the prediction model.

[0010] The LightGBM model was trained and validated based on the characteristics of ETC issuance volume to obtain an ETC issuance volume prediction model.

[0011] Based on the ETC issuance prediction model, the daily ETC issuance prediction value is obtained.

[0012] As a further optional solution to the novel ETC issuance prediction method, the step of obtaining the ETC issuance prediction demand specifically includes:

[0013] Obtain the business background and objectives, forecasting tasks, and functional and non-functional requirements for ETC issuance volume forecasting;

[0014] Based on the business background and objectives of ETC issuance forecasting, forecasting tasks, and functional and non-functional requirements, the demand for ETC issuance is integrated to obtain the ETC issuance demand.

[0015] As a further optional solution to the novel ETC issuance prediction method, the collection of multidimensional data and the construction of ETC issuance characteristics based on the multidimensional data specifically include:

[0016] Collect historical daily ETC issuance volume, monthly new vehicle registration volume, monthly MTC traffic volume, monthly ETC usage rate, holiday and weekend calendar information, promotional activities, weather, and regional economic indicators from various channels.

[0017] The historical daily ETC issuance volume, monthly new vehicle registration volume, monthly MTC traffic volume, monthly ETC usage rate, holiday and weekend calendar information, promotional activities, weather, and regional economic indicators are preprocessed to obtain data with unified dimensions.

[0018] By constructing features from data of a unified dimension, we obtain ETC issuance volume features that include time features, business time sequence features, event identifier features, external correlation features, and channel attribute features.

[0019] As a further optional solution to the novel ETC issuance prediction method, the step of training and validating the LightGBM model based on ETC issuance characteristics to obtain an ETC issuance prediction model specifically includes:

[0020] The time series cross-validation method is used to dynamically divide the ETC issuance characteristics into training and validation sets;

[0021] The LightGBM model was trained based on the training set to obtain the initial ETC issuance prediction model;

[0022] The initial ETC issuance prediction model was validated using the validation set to obtain the final ETC issuance prediction model.

[0023] As a further optional solution to the novel ETC issuance prediction method, the step of dynamically dividing the ETC issuance characteristics into a training set and a validation set using a time series cross-validation method specifically includes:

[0024] Obtain the time-series characteristics of ETC issuance volume;

[0025] Based on the time-series characteristics of ETC issuance volume, a time-series cross-validation method is adopted to formulate dynamic partitioning rules;

[0026] The ETC issuance volume characteristics are divided based on dynamic partitioning rules to obtain training and validation sets.

[0027] As a further alternative to the novel ETC issuance prediction method, an early stopping mechanism is used to determine the number of iterations during the training of the LightGBM model based on the training set.

[0028] As a further optional solution to the novel ETC issuance prediction method, the prediction method also includes:

[0029] The daily ETC issuance forecast is visualized using trend comparison charts and data tables.

[0030] A novel ETC issuance prediction system includes:

[0031] The acquisition module is used to obtain the predicted demand for ETC issuance.

[0032] The collection and construction module is used to collect multidimensional data and construct ETC issuance characteristics based on the multidimensional data;

[0033] The model selection module is used to select the LightGBM model as the prediction model based on the ETC issuance volume prediction demand and the characteristics of ETC issuance volume.

[0034] The model training module is used to train and validate the LightGBM model based on ETC issuance characteristics to obtain an ETC issuance prediction model.

[0035] The prediction module is used to predict the issuance of ETC based on the ETC issuance prediction model, and obtain the daily predicted value of ETC issuance.

[0036] A computing device includes a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the steps of any of the novel ETC issuance prediction methods described above.

[0037] A computer-readable storage medium storing a computer program that, when executed by a processor, implements the steps of any of the novel ETC issuance prediction methods described above.

[0038] The beneficial effects of this invention are as follows: By collecting multidimensional data and constructing ETC issuance characteristics based on this data, the relationship between ETC issuance and various factors can be reflected more comprehensively and accurately, providing a solid and scientific data foundation for subsequent prediction models and overcoming the lack of scientific rigor caused by incomplete data in existing technologies. Secondly, by selecting the LightGBM model as the prediction model, the LightGBM model, as an efficient implementation of gradient boosting decision trees, has strong nonlinear fitting capabilities and can automatically learn complex patterns and interactions between features in the data, thereby more accurately characterizing the relationship between ETC issuance and various features and improving prediction accuracy. Simultaneously, the LightGBM model employs a histogram-based decision tree algorithm and a depth-limited Leaf-... The Wise leaf growth strategy and techniques such as mutual exclusion feature binding significantly reduce computational load and memory consumption, enabling rapid training on large-scale data. This allows the model to complete training in a shorter time, providing timely prediction results for decision-making, while reducing hardware resource requirements and improving prediction efficiency. Furthermore, the LightGBM model is trained and validated based on the constructed ETC issuance features. Through the training process, the LightGBM model can continuously adjust its parameters according to the feature data to minimize prediction errors. The validation phase can evaluate the performance of the LightGBM model and promptly identify problems. In this way, the model can be continuously optimized to better adapt to real-world data, further improving prediction accuracy and reducing the deviation between prediction results and actual conditions. Attached Figure Description

[0039] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0040] Figure 1 This is a flowchart illustrating a novel ETC issuance prediction method according to the present invention.

[0041] Figure 2 This is a schematic diagram of the composition of a novel ETC issuance prediction system according to the present invention;

[0042] Figure 3 This is a schematic diagram of the composition of a computing device according to the present invention. Detailed Implementation

[0043] The technical solutions in the embodiments of the present invention will be clearly and completely described below. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of them. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0044] refer to Figures 1 to 3 A novel method for predicting ETC issuance volume, specifically including:

[0045] Demand for ETC issuance forecasts includes:

[0046] Collect key information: First, obtain the business background and objectives of ETC issuance forecasting, and clarify the business decisions and operational goals that the forecasting work should support; second, define the forecasting task, including the forecasting object (such as the issuance volume of Guangdong Tong Card ETC), time granularity (such as daily), forecasting period (such as short-term 7-30 days, medium-term 1 year), etc.; finally, determine functional and non-functional requirements, such as the forecasting error rate needing to be controlled within a certain range, the model needing to be interpretable, and supporting subsequent addition of features or channels, etc.

[0047] Constructing ETC issuance volume requirements: Based on the collected key information, a system for constructing ETC issuance volume requirements is carried out. In this process, it is necessary to comprehensively consider the specific requirements of the business background and objectives for the forecasting work, the requirements of the forecasting task for data granularity and time range, and the impact of functional and non-functional requirements on model performance and scalability, so as to form a comprehensive, specific and actual business-compliant ETC issuance volume requirements document.

[0048] Specifically, by comprehensively collecting key information such as business background and objectives and building accurate prediction models, solid data support can be provided for business decisions regarding ETC issuance volume. For example, based on prediction results for different regions and time periods, more targeted marketing strategies can be formulated to improve the success rate of market expansion. After clarifying the operational objectives supported by the prediction work, the prediction model can be optimized around these objectives, enabling production, sales, service and other operational processes to be accurately arranged according to the predicted ETC issuance volume, ensuring that various operational indicators are achieved smoothly and improving overall operational efficiency.

[0049] Reasonable definition of data granularity and time range, as well as consideration of functional and non-functional requirements in forecasting tasks, help enterprises accurately grasp resource needs. For example, based on the predicted peak and trough of issuance volume, human, material, and financial resources can be rationally allocated to avoid resource waste or shortages and reduce operating costs. For ETC equipment manufacturers or issuing institutions, accurate issuance volume forecasts enable refined inventory management, allowing them to arrange production plans in advance based on forecast results, ensuring that inventory levels meet market demand without tying up too much capital and improving capital turnover.

[0050] The defined non-functional requirements, such as the control range of the prediction error rate, enable the prediction results to have a certain reliability assessment standard. When the prediction error exceeds the control range, a risk warning can be issued in a timely manner, prompting enterprises to pay attention to market changes or problems with the model and take corresponding measures to reduce risks.

[0051] Collect multidimensional data and construct ETC issuance characteristics based on the multidimensional data, specifically including:

[0052] Data is collected from sources such as the ETC issuance system, the highway network toll collection and settlement system, the official website of the car owner's home, and the public calendar API. This includes, but is not limited to, historical daily ETC issuance volume, monthly new car registration volume, monthly MTC traffic volume, monthly ETC usage rate, accurate holiday and weekend calendar information, promotional activities, weather, and regional economic indicators.

[0053] The raw data is cleaned and integrated, missing values, outliers and isolated values ​​are processed, the date format is converted and sorted in ascending order by date, the ETC usage rate is converted into a percentage, and data from different sources and with different granularities are unified to the "date" dimension.

[0054] The structural features include time features (year, quarter, month, week, day, day of the week), business time sequence features (issue volume of the previous day, issue volume of the previous working day, average issue volume of the past 7 days), event identifier features (whether it is a holiday, whether it is one day before or after a holiday, whether it is a weekend, whether it is a promotion day), external correlation features (current month's MTC vehicle traffic, previous month's MTC vehicle traffic, current month's new vehicle registrations, previous month's new vehicle registrations, current month's ETC usage rate), and channel attribute features (channel type, channel category).

[0055] Specifically, data was collected from multiple sources, including the ETC issuance system and the highway network toll collection and settlement system, covering historical daily ETC issuance volume across various channels and monthly new vehicle registration volume. This extensive data collection method ensured the comprehensiveness of the data, enabling a more complete reflection of market dynamics and business conditions related to ETC issuance, and providing rich material for subsequent analysis. The raw data was cleaned and integrated, handling missing values, outliers, and isolated values, effectively removing noise and errors. At the same time, date formats were converted and sorted in ascending order by date, and the ETC usage rate was converted to a percentage and data dimensions were unified, ensuring data consistency and accuracy, making the data more suitable for analysis and modeling requirements.

[0056] The system constructs various feature types, including time features (year, quarter, month, etc.), business time series features (issue volume of the previous day, etc.), event identifier features (whether it is a holiday, etc.), external correlation features (MTC traffic volume in the current month, etc.), and channel attribute features (channel type, etc.). These features characterize the influencing factors of ETC issuance volume from different perspectives, enabling a more comprehensive and in-depth exploration of the patterns behind the data and providing strong support for accurate prediction of ETC issuance volume. The constructed features are closely related to the ETC issuance business. For example, business time series features reflect the time continuity and short-term fluctuations of issuance volume, event identifier features consider the impact of special periods such as holidays and promotional days on issuance volume, and external correlation features reflect the connection with other relevant business indicators. This close integration with the business makes the features more practical and can better serve business decision-making.

[0057] Based on the business objective of high-precision prediction and analysis of the main factors affecting ETC issuance volume, as clearly stated in the ETC issuance volume requirements, and considering the structured data characteristics contained in the ETC issuance volume features (including numerical features such as ETC issuance volume and MTC traffic flow) and categorical features such as day of the week and whether it is a holiday), the LightGBM model was selected as the prediction model. The selection criteria are as follows: the LightGBM model has a feature importance ranking function, meeting the analysis requirements of the main factors affecting issuance volume; it can directly support categorical feature input without complex preprocessing, effectively avoiding dimensionality explosion and computational efficiency degradation; it employs mutually exclusive feature binding technology to reduce feature dimensionality and improve training speed; its histogram-based decision tree algorithm reduces computation and memory usage; and its depth-limited leaf-wise growth strategy and one-sided gradient sampling technology further improve training efficiency and control model complexity. Therefore, in handling high-dimensional, mixed-type time series regression prediction problems, it achieves the best overall performance in terms of accuracy, efficiency, interpretability, and engineering feasibility.

[0058] The LightGBM model was trained and validated based on ETC issuance characteristics to obtain an ETC issuance prediction model, which specifically includes:

[0059] The time series cross-validation method is used to dynamically divide the ETC issuance characteristics into training and validation sets;

[0060] The LightGBM model was trained based on the training set to obtain the initial ETC issuance prediction model;

[0061] The initial ETC issuance prediction model was validated using the validation set to obtain the final ETC issuance prediction model.

[0062] Specifically, using time-series cross-validation to dynamically divide the training and validation sets allows for more rational use of data, enabling the model to fully learn the characteristics and patterns of ETC issuance volume in different time periods. The LightGBM model itself is characterized by high efficiency and high accuracy. Combined with reasonable training and validation methods, it can significantly improve the accuracy of ETC issuance volume prediction and reduce prediction errors. Time-series cross-validation simulates the scenario of actual data changing over time, allowing the model to be exposed to diverse data patterns during training and validation in different time periods. This helps the model learn more general feature representations and enhances its adaptability to unknown future data, thereby improving the model's generalization ability and enabling it to maintain good predictive performance in different market environments.

[0063] The time series cross-validation method fully leverages the value of ETC issuance feature data, ensuring that each part of the data plays a role in the training and validation process, avoiding data waste, improving data utilization efficiency, and providing more information for model training from limited data. The LightGBM model is fast and efficient during training. Combined with a reasonable training and validation process, it can rationally allocate computing resources, reduce unnecessary computational consumption, reduce computational costs while ensuring model performance, and improve the overall efficiency of resource utilization.

[0064] By training and validating the model multiple times on different training and validation sets, the model can reduce the impact of unreasonable data partitioning or data noise, making the prediction results more stable and reliable. This avoids model performance fluctuations caused by a single data partitioning, providing enterprises with continuous and stable prediction services. This training and validation method has a clear process, which facilitates model monitoring and maintenance. When model performance declines, the problem can be quickly located to determine whether it is due to data changes or the model itself, so as to update and optimize the model in a timely manner and ensure the long-term stable operation of the model.

[0065] It should be noted that after obtaining the final ETC issuance prediction model, the performance of the final ETC issuance prediction model is evaluated through root mean square error and feature importance analysis.

[0066] In some embodiments, the method of dynamically dividing the ETC issuance characteristics into a training set and a validation set using time series cross-validation specifically includes:

[0067] The time series characteristics (time dependence, periodic fluctuation characteristics and trend changes) of ETC issuance volume are obtained, and its non-independent and identically distributed characteristics as time series data are identified. The periodic fluctuation characteristics include the issuance volume change pattern with daily, weekly and monthly periods, and the trend changes include long-term increase or decrease trends affected by policy adjustments or market promotion activities.

[0068] Based on the time-series characteristics of ETC issuance volume, a time-series cross-validation method is adopted, and a dynamic partitioning rule is formulated. Specifically, the historical ETC issuance volume data is divided into K consecutive time periods as folds in chronological order, where K≥2. During each validation, the first i folds (i=1,2,...,K-1) are selected as the training set, and the (i+1)th fold is selected as the validation set. K validations are performed in a rolling manner to ensure that the time order of the training set and the validation set is strictly irreversible and to avoid future information leakage.

[0069] The ETC issuance volume characteristics are divided based on dynamic partitioning rules to obtain training and validation sets.

[0070] Specifically, time dependence analysis: by calculating the autocorrelation coefficient and other methods, it was found that the daily ETC issuance volume is strongly correlated with the issuance volume of the previous 1-3 days, indicating that the data has time dependence.

[0071] Periodic fluctuation characteristics identification: Using algorithms such as Fourier transform, the issuance volume change pattern with weekly and monthly cycles is identified. For example, the issuance volume is relatively high on weekdays (Monday to Friday) and decreases on weekends (Saturday and Sunday); a small peak in issuance volume occurs in the latter half of each month due to factors such as concentrated processing by some users.

[0072] Trend analysis: Combining policy documents and market promotion activity records, the impact of vigorously promoting the ETC policy is analyzed. The overall issuance volume shows a long-term growth trend, but the growth rate may fluctuate at certain policy adjustment points or after the end of market promotion activities.

[0073] We choose to divide the historical ETC issuance data into K=5 consecutive time periods as Fold.

[0074] In the first validation, the first i=1 folds (i.e. the first 1 / 5 of the data time period) are selected as the training set, and the i+1=2 folds (i.e. the second 1 / 5 of the data time period) are selected as the validation set.

[0075] During the second validation, the first two folds are selected as the training set, the third fold as the validation set, and so on, with K=5 validations performed in a rolling manner. At the same time, strict data isolation measures are taken to ensure that the time order of the training set and the validation set is irreversible in each validation, thus avoiding information leakage.

[0076] Based on the dynamic partitioning rules established above, the ETC issuance data for the past three years will be divided into corresponding training and validation sets for subsequent model training and validation.

[0077] Thus, the time-series cross-validation method fully considers the time-series characteristics of ETC issuance data, such as time dependence, periodicity, and trend. By dynamically dividing the training and validation sets, the model can learn and validate on data from different time periods, making full use of various feature information in the data and avoiding the loss of feature information that may be caused by traditional random data partitioning. For example, after identifying features with a weekly cycle, dynamic partitioning ensures that the model is trained and validated on data combinations from different weeks, better capturing the changing patterns within the cycle and improving the efficiency and rationality of data utilization. Secondly, rolling multiple validations can more comprehensively evaluate the model's performance under different time periods and data combinations, avoiding the randomness and bias that may exist in a single validation, making the evaluation of performance indicators such as model prediction accuracy and stability more accurate and reliable. For example, during multiple validation processes, the model's performance in different market trend stages (growth, stability, and volatility) can be observed, thus more accurately judging the model's applicability in various real-world scenarios. Furthermore, strictly ensuring the irreversible temporal order of the training and validation sets effectively avoids the risk of future information leaking into the training process. This is crucial for time series prediction problems, as future data should not affect the model's learning and prediction of past and current data. For instance, in ETC issuance prediction, if information about future marketing activities leaks into the training set, the model may over-rely on this information, leading to inaccurate predictions when there are no similar promotional activities in practical applications. Dynamic partitioning rules effectively avoid this problem, ensuring the model's reliability and practicality.

[0078] In some embodiments, an early stopping mechanism is used during the training of the LightGBM model based on the training set. When the model’s performance on the validation set no longer improves for several consecutive rounds, the training is automatically stopped to prevent overfitting and to determine the optimal number of iterations.

[0079] Specifically, set early stopping parameters: determine the validation set, divide the historical data according to a certain ratio (e.g., 80% for training set and 20% for validation set), set the number of early stopping rounds (patience), for example, set it to 5 rounds, and determine the performance evaluation metric. Here, the mean absolute percentage error (MAPE) is selected as the metric to measure the model's performance on the validation set.

[0080] Model training and monitoring: Start the training process of the LightGBM model. After each round of training, evaluate the model using the validation set, calculate the MAPE value, and record the change of the MAPE value of the model on the validation set after each round of training.

[0081] Early Stop Judgment and Execution: Continuously monitor changes in the MAPE value. If the MAPE value of the model on the validation set does not decrease (i.e., performance does not improve) in 5 consecutive training rounds, the early stop mechanism is triggered, and the training process of the model is automatically stopped. At this time, the number of training rounds recorded is the optimal number of iterations, and the model corresponding to the number of rounds is used as the final ETC issuance prediction model.

[0082] Thus, during model training, as the number of iterations increases, the model may gradually overlearn from noisy and outlier data in the training set, resulting in excellent performance on the training set but declining performance on new data (validation set and real-world application data), i.e., overfitting. The early stopping mechanism monitors the model's performance on the validation set and stops training when performance no longer improves after several consecutive iterations, effectively preventing overfitting of the training data and giving the model better generalization ability, enabling it to more accurately predict ETC issuance. Secondly, without the early stopping mechanism, achieving good model performance might require a large number of iterative training iterations, consuming significant computational resources, including C++. Early stopping consumes more PU and memory, and also increases training time. However, it can stop training in time when the model performance no longer improves, avoiding unnecessary calculations, improving training efficiency, and reducing computational costs. This is especially effective in training large-scale data and complex models. In addition, early stopping automatically determines the optimal number of iterations for the model, avoiding the tedious process of manually trying different number of iterations to determine the best model. This not only saves manpower and time costs, but also ensures that the number of iterations found is the number that makes the model achieve relatively optimal performance on the validation set under the current training conditions. This ensures the performance and quality of the model and provides a reliable guarantee for accurately predicting ETC issuance.

[0083] In some embodiments, the prediction method further includes:

[0084] The daily ETC issuance forecast is visualized using trend comparison charts and data tables.

[0085] Specifically, collect the actual daily ETC issuance data for this month, and organize it with the corresponding forecast values. Arrange the data in chronological order to form a data table containing three columns: date, actual issuance, and forecast issuance.

[0086] Trend comparison chart creation: Choose a suitable visualization tool, such as the Matplotlib library in Python or the professional data analysis software Tableau; plot the actual issuance volume and the predicted issuance volume as two separate curves with the date on the horizontal axis and the issuance volume of ETC on the vertical axis. Different colors can be used to distinguish the two curves, for example, the actual issuance volume can be represented by a blue curve and the predicted issuance volume by an orange curve; add elements such as legends, titles, and axis labels to the chart to make the information clearer and easier to understand.

[0087] Data table presentation: The organized data table containing date, actual issuance volume, and predicted issuance volume is displayed directly. It can be placed next to or below the trend comparison chart for easy comparison and viewing of specific values.

[0088] Thus, the trend comparison chart clearly presents the changing trends of actual and predicted ETC issuance over time. By observing the closeness of the two curves, business personnel can intuitively judge the accuracy of the prediction model. If the two curves are basically consistent and the values ​​are close, it indicates that the prediction effect is good; if there is a large deviation, it suggests that the model may need further optimization. The data table provides specific numerical information, which facilitates accurate comparison between predicted and actual values ​​and calculation of error indicators, such as mean absolute error (MAE) and root mean square error (RMSE), to further quantify the accuracy of the prediction. Secondly, for the managers of ETC issuing institutions, the intuitive visualization helps to quickly understand market dynamics and predictions. Based on the trend comparison chart and data table, managers can promptly identify abnormal changes in issuance, analyze the reasons, and formulate corresponding strategies. For example, if the predicted issuance is consistently lower than the actual demand, it may be necessary to increase marketing efforts or optimize the issuance process. When formulating production plans, accurate prediction visualization results can help to rationally arrange the production and inventory of ETC equipment, avoid inventory backlog or shortages, and improve resource utilization efficiency.

[0089] A novel ETC issuance prediction system includes:

[0090] The acquisition module is used to obtain the predicted demand for ETC issuance.

[0091] The collection and construction module is used to collect multidimensional data and construct ETC issuance characteristics based on the multidimensional data;

[0092] The model selection module is used to select the LightGBM model as the prediction model based on the ETC issuance volume prediction demand and the characteristics of ETC issuance volume.

[0093] The model training module is used to train and validate the LightGBM model based on ETC issuance characteristics to obtain an ETC issuance prediction model.

[0094] The prediction module is used to predict the issuance of ETC based on the ETC issuance prediction model, and obtain the daily predicted value of ETC issuance.

[0095] A computing device includes a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the steps of any of the novel ETC issuance prediction methods described above.

[0096] A computer-readable storage medium storing a computer program that, when executed by a processor, implements the steps of any of the novel ETC issuance prediction methods described above.

[0097] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A novel method for predicting ETC issuance volume, characterized in that, Specifically, it includes: Obtain forecasts of ETC issuance demand; Collect multidimensional data and construct ETC issuance characteristics based on the multidimensional data; Based on the demand for ETC issuance and the characteristics of ETC issuance, the LightGBM model was selected as the prediction model. The LightGBM model was trained and validated based on the characteristics of ETC issuance volume to obtain an ETC issuance volume prediction model. Based on the ETC issuance prediction model, the daily ETC issuance prediction value is obtained.

2. The novel ETC issuance prediction method according to claim 1, characterized in that, The requirement to obtain ETC issuance forecasts specifically includes: Obtain the business background and objectives, forecasting tasks, and functional and non-functional requirements for ETC issuance volume forecasting; Based on the business background and objectives of ETC issuance forecasting, forecasting tasks, and functional and non-functional requirements, the requirements are integrated to obtain the ETC issuance forecasting requirements.

3. The novel ETC issuance prediction method according to claim 1, characterized in that, The collection of multidimensional data and the construction of ETC issuance characteristics based on the multidimensional data specifically include: Collect historical daily ETC issuance volume, monthly new vehicle registration volume, monthly MTC traffic volume, monthly ETC usage rate, holiday and weekend calendar information, promotional activities, weather, and regional economic indicators from various channels. The historical daily ETC issuance volume, monthly new vehicle registration volume, monthly MTC traffic volume, monthly ETC usage rate, holiday and weekend calendar information, promotional activities, weather, and regional economic indicators are preprocessed to obtain data with unified dimensions. By constructing features from data of a unified dimension, we obtain ETC issuance volume features that include time features, business time sequence features, event identifier features, external correlation features, and channel attribute features.

4. The novel ETC issuance prediction method according to claim 1, characterized in that, The process of training and validating the LightGBM model based on ETC issuance volume characteristics to obtain an ETC issuance volume prediction model specifically includes: The time series cross-validation method is used to dynamically divide the ETC issuance characteristics into training and validation sets; The LightGBM model was trained based on the training set to obtain the initial ETC issuance prediction model; The initial ETC issuance prediction model was validated using the validation set to obtain the final ETC issuance prediction model.

5. The novel ETC issuance prediction method according to claim 4, characterized in that, The method of dynamically dividing ETC issuance characteristics into training and validation sets using time-series cross-validation specifically includes: Obtain the time-series characteristics of ETC issuance volume; Based on the time-series characteristics of ETC issuance volume, a time-series cross-validation method is adopted to formulate dynamic partitioning rules; The ETC issuance volume characteristics are divided based on dynamic partitioning rules to obtain training and validation sets.

6. The novel ETC issuance prediction method according to claim 5, characterized in that, The training process of the LightGBM model based on the training set employs an early stopping mechanism to determine the number of iterations.

7. The novel ETC issuance prediction method according to claim 1, characterized in that, The prediction method further includes: The daily ETC issuance forecast is visualized using trend comparison charts and data tables.

8. A novel ETC issuance prediction system, characterized in that, include: The acquisition module is used to obtain the predicted demand for ETC issuance. The collection and construction module is used to collect multidimensional data and construct ETC issuance characteristics based on the multidimensional data; The model selection module is used to select the LightGBM model as the prediction model based on the ETC issuance volume prediction demand and the characteristics of ETC issuance volume. The model training module is used to train and validate the LightGBM model based on ETC issuance characteristics to obtain an ETC issuance prediction model. The prediction module is used to predict the issuance of ETC based on the ETC issuance prediction model, and obtain the daily predicted value of ETC issuance.

9. A computing device, characterized in that, The method includes a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the steps of the novel ETC issuance prediction method according to any one of claims 1-7.

10. A computer-readable storage medium, characterized in that, The storage medium stores a computer program, which, when executed by a processor, implements the steps of the novel ETC issuance prediction method according to any one of claims 1-7.

Images