An event-driven intelligent demand response and dispatching method and device for electric official vehicles

Abstract

This invention discloses an event-driven intelligent demand response and scheduling method and device for electric official vehicles. The method collects historical vehicle usage event data for semantic modeling, extracts multi-dimensional features, and constructs structured feature vectors. It then uses the K-means algorithm for clustering to form three types of event databases. Next, it performs classification modeling using multi-distribution adaptive fitting and a CIA (Conformity Analysis) optimization strategy, outputting the statistical characteristics and distribution of the specified events. Finally, it generates a highly representative set of vehicle usage event scenarios using Monte Carlo sampling and backward scenario reduction. The method collects historical spot market prices and external environmental information to construct a set of input variables. It then processes the input variables into a conditional time series using a generative adversarial network to output future electricity price scenarios. Finally, it constructs a multi-day scheduling objective function, solves it using a robust optimization algorithm, and outputs charging periods and scheduling sequences. This invention solves problems such as low scheduling efficiency, high charging costs, and insufficient adaptation to grid demand response.

Description

Technical Field

[0001] This invention relates to the field of power demand response and vehicle dispatching technology, specifically to an event-driven intelligent demand response and dispatching method and apparatus for electric official vehicles. Background Technology

[0002] Driven by my country's green development strategy, electric vehicles, with their core advantages of zero emissions and low energy consumption, have gradually become the mainstream choice for official travel. Due to the relatively fixed usage scenarios, standardized management, and traceable data of official vehicles, a vast amount of data on vehicle application records, driving trajectories, power consumption, and charging behavior has been accumulated, providing a data foundation for intelligent dispatching. Currently, existing electric official vehicle dispatching systems mostly integrate technologies such as GPS positioning and IoT sensing, and some have introduced big data and cloud computing methods, forming a standardized dispatching model of "online application - process approval - vehicle matching - route planning - task feedback." The core functions include real-time vehicle positioning, remaining power monitoring, and basic dispatching, which improves the convenience of official travel to a certain extent. Meanwhile, with the advancement of electricity market reforms, the grid side has introduced demand response policies such as peak-valley time-of-use pricing, providing opportunities for official vehicles to reduce operating costs by optimizing charging times.

[0003] However, existing technologies still have significant shortcomings, making it difficult to fully unleash the intelligent and economic potential of official vehicle dispatching. On the one hand, the depth of data mining is insufficient. Existing systems only perform simple classification and statistics on vehicle usage events, failing to delve into the temporal patterns, mileage characteristics, and correlations of different events. This results in the underutilization of the value of massive amounts of data, and dispatching decisions lack precise data support. On the other hand, dispatching strategies lack specificity. They fail to develop differentiated solutions based on the urgency and predictability of official vehicle events, instead using a uniform dispatching logic to allocate vehicles and plan charging times. This not only easily leads to conflicts between vehicle idleness and urgent tasks but also frequently results in charging behavior coinciding with off-peak electricity pricing, failing to fully respond to electricity market price signals and adding extra operating costs. Furthermore, existing dispatching technologies largely borrow from the dispatching logic of private cars and car-sharing services, failing to fully adapt to the unique advantages of official vehicle management regulations and the absence of privacy concerns, resulting in insufficient adaptability of dispatching solutions to actual official travel needs.

[0004] Therefore, there is an urgent need for an event-driven intelligent demand response and scheduling method for electric official vehicles to solve the problems of low scheduling efficiency, high charging costs, and insufficient adaptation to grid demand response in existing technologies. Summary of the Invention

[0005] To address this, the present invention provides an event-driven intelligent demand response and scheduling method and apparatus for electric official vehicles, which solves the problems of low scheduling efficiency, high charging costs, and insufficient adaptation to grid demand response caused by the lack of in-depth data mining, lack of event differentiation strategies, and failure to adapt to the characteristics of official scenarios in existing electric official vehicle scheduling technologies.

[0006] To achieve the above objectives, the present invention provides the following technical solution: an event-driven intelligent demand response and scheduling method for electric official vehicles, characterized in that it includes:

[0007] By collecting historical vehicle usage event data, semantic modeling is performed on unstructured text information to extract multi-dimensional features; structured feature vectors are constructed based on the multi-dimensional features; and the structured feature vectors are clustered using the K-means algorithm to form an event database.

[0008] Based on the event database, the vehicle usage time and travel mileage of the set events are classified and modeled by multi-distribution adaptive fitting and CIA selection strategy, and the statistical characteristics and distribution of the set events are output. Based on the statistical characteristics and distribution of the set events, a set of highly representative vehicle usage event scenarios is generated by Monte Carlo sampling strategy and backward scenario reduction method.

[0009] By collecting historical prices in the spot market and information about the external environment, a set of input variables is constructed; the input variables in the set of input variables are then processed by a conditional time series generator adversarial network to output future electricity price scenarios.

[0010] Based on the set of highly representative vehicle usage scenarios and the future electricity price scenario, and combined with vehicle battery level and charging pile distribution information, a multi-day scheduling objective function is constructed. Based on a differentiated constraint strategy, a robust optimization algorithm is used to solve the multi-day scheduling objective function, and the charging time period and scheduling order are output.

[0011] As a preferred solution for an event-driven intelligent demand response and scheduling method for electric official vehicles, when performing semantic modeling on unstructured text information, the original text is de-stopped, segmented, and a keyword semantic dependency matrix is ​​constructed. High-dimensional semantic vectors are extracted from the keyword semantic dependency matrix using a pre-trained language model BERT. The multi-dimensional features include: event urgency, predictability, monthly occurrence frequency, and average mileage. The event urgency is divided into three levels: ordinary, important, and urgent. The predictability is divided into three levels: high, medium, and low.

[0012] As a preferred solution for an event-driven intelligent demand response and scheduling method for electric official vehicles, the objective function of the K-means algorithm is expressed as follows during the clustering of the structured feature vectors:

[0013]

[0014] In the formula, Let be the centroid of the k-th type of event cluster; Let be the feature vector of event i; N is the total number of events.

[0015] As a preferred solution for an event-driven intelligent demand response and scheduling method for electric official vehicles, this method classifies and models the vehicle usage time and travel mileage of a given event through multi-distribution adaptive fitting and CIA selection strategies. It constructs a mixed probability density function by weighting normal distribution, log-normal distribution, and Poisson distribution; evaluates the fitting effect of the mixed probability density function using the AIC criterion, calculating the AIC value for each mixed probability density function; selects several optimal fitting candidate models based on the AIC values; evaluates and filters each optimal fitting candidate model using a utility function to obtain the optimal fitting model; and outputs the statistical characteristics and distribution of the given event using the optimal fitting model.

[0016] As a preferred solution for an event-driven intelligent demand response and scheduling method for electric official vehicles, in the process of generating the highly representative set of vehicle usage event scenarios through Monte Carlo sampling strategy and backward scenario reduction method, the Monte Carlo sampling strategy randomly selects a set number of samples from the statistical features and distribution to generate an initial set of vehicle usage event scenarios; based on the initial set of vehicle usage event scenarios, the Wasserstein distance between any two scenarios is calculated through backward scenario reduction method, and redundant scenarios corresponding to the minimum probability distance are iteratively deleted until the number of scenarios reaches the set target, thereby generating the highly representative set of vehicle usage event scenarios.

[0017] As a preferred embodiment of an event-driven intelligent demand response and scheduling method for electric official vehicles, the conditional time series generative adversarial network includes a generator and a discriminator; the generator outputs a predicted electricity price sequence based on the input variables and random noise; the discriminator outputs the probability of the authenticity of the electricity price sequence based on the predicted electricity price sequence; and the prediction error is minimized through adversarial training to output the future electricity price scenario.

[0018] The objective function expression for adversarial training is:

[0019]

[0020] In the formula, This is a gradient penalty term; To generate the target value function for the adversarial network; The expected score is the score of the true sample distribution; To generate the expected score of the sample distribution; represents the weight coefficient of the gradient penalty term.

[0021] As a preferred embodiment of an event-driven intelligent demand response and scheduling method for electric official vehicles, the expression for the multi-day scheduling objective function is:

[0022]

[0023] In the formula, For charging costs; The penalty cost is represented by w; the uncertain parameters of the scenario are represented by u; the set of uncertain parameters is represented by x; and the decision variables are represented by x (departure time, charging time, etc.).

[0024] As a preferred solution for an event-driven intelligent demand response and scheduling method for electric official vehicles, the differentiated constraint strategy is as follows: for emergency events, a rigid time window constraint is adopted to ensure that the task completion time does not exceed the deadline; for temporary official events, a flexible start window constraint is adopted to allow the task start time to be adjusted within a preset time range; and for routine events, a planning cycle constraint is adopted to ensure that all tasks are completed within a preset cycle.

[0025] This invention also provides an event-driven intelligent demand response and scheduling device for electric official vehicles, based on the above-mentioned event-driven intelligent demand response and scheduling method for electric official vehicles, comprising:

[0026] The vehicle usage data analysis module is used to collect historical vehicle usage event data, perform semantic modeling on unstructured text information, and extract multi-dimensional features; construct structured feature vectors based on the multi-dimensional features; and cluster the structured feature vectors using the K-means algorithm to form an event database.

[0027] The vehicle usage demand prediction module is used to classify and model the vehicle usage time and travel mileage of a set event based on the event database through multi-distribution adaptive fitting and CIA optimization strategy, and output the statistical characteristics and distribution of the set event; based on the statistical characteristics and distribution of the set event, a set of highly representative vehicle usage event scenarios is generated through Monte Carlo sampling strategy and backward scenario reduction method.

[0028] The electricity price forecasting module is used to construct a set of input variables by collecting historical prices in the spot market and external environmental information; the input variables in the set of input variables are processed by a generative adversarial network based on time series conditions to output future electricity price scenarios;

[0029] The bus optimization scheduling module is used to construct a multi-day scheduling objective function based on the set of highly representative vehicle usage events and the future electricity price scenario, combined with vehicle battery level and charging pile distribution information; based on a differentiated constraint strategy, the multi-day scheduling objective function is solved by a robust optimization algorithm, and the charging time period and scheduling order are output.

[0030] As a preferred solution for an event-driven intelligent demand response and scheduling device for electric official vehicles, the vehicle data analysis module performs semantic modeling on unstructured text information by removing stops, segmenting words, and constructing a keyword semantic dependency matrix. A high-dimensional semantic vector is extracted from the keyword semantic dependency matrix using a pre-trained language model BERT. The multi-dimensional features include: event urgency, predictability, monthly frequency, and average mileage. Event urgency is categorized into three levels: ordinary, important, and urgent. Predictability is categorized into three levels: high, medium, and low.

[0031] As a preferred solution for an event-driven intelligent demand response and scheduling device for electric official vehicles, in the vehicle usage data analysis module, during the clustering of the structured feature vectors using the K-means algorithm, the objective function of the K-means algorithm is expressed as:

[0032]

[0033] In the formula, Let be the centroid of the k-th type of event cluster; Let be the feature vector of event i; N is the total number of events.

[0034] As a preferred solution for an event-driven intelligent demand response and scheduling device for electric official vehicles, the vehicle demand prediction module, in the process of classifying and modeling the vehicle usage time and travel mileage of a set event through multi-distribution adaptive fitting and CIA selection strategy, constructs a mixed probability density function by weighting a normal distribution, log-normal distribution, and Poisson distribution; evaluates the fitting effect of the mixed probability density function using the AIC criterion, and calculates the AIC value corresponding to each mixed probability density function; selects several optimal fitting candidate models based on the AIC values; evaluates and filters each optimal fitting candidate model using a utility function to obtain the optimal fitting model; and outputs the statistical characteristics and distribution of the set event through the optimal fitting model.

[0035] As a preferred solution for an event-driven intelligent demand response and scheduling device for electric official vehicles, the vehicle demand prediction module generates a highly representative set of vehicle usage events by using a Monte Carlo sampling strategy and a backward scenario reduction method. The Monte Carlo sampling strategy randomly selects a set number of samples from the statistical features and distribution to generate an initial set of vehicle usage events. Based on this initial set, the backward scenario reduction method is used to calculate the Wasserstein distance between any two scenarios, iteratively deleting redundant scenarios corresponding to the minimum probability distance until the number of scenarios reaches a set target, thus generating the highly representative set of vehicle usage events.

[0036] As a preferred embodiment of an event-driven intelligent demand response and scheduling device for electric official vehicles, the electricity price prediction module includes a conditional time series generative adversarial network comprising a generator and a discriminator; the generator outputs a predicted electricity price sequence based on the input variables and random noise; the discriminator outputs the probability of the authenticity of the electricity price sequence based on the predicted electricity price sequence; and the prediction error is minimized through adversarial training to output the future electricity price scenario.

[0037] The objective function expression for adversarial training is:

[0038]

[0039] In the formula, This is a gradient penalty term; To generate the target value function for the adversarial network; The expected score is the score of the true sample distribution; To generate the expected score of the sample distribution; represents the weight coefficient of the gradient penalty term.

[0040] As a preferred embodiment of an event-driven intelligent demand response and scheduling device for electric official vehicles, the expression for the multi-day scheduling objective function in the vehicle optimization scheduling module is as follows:

[0041]

[0042] In the formula, For charging costs; The penalty cost is represented by w; the uncertain parameters of the scenario are represented by u; the set of uncertain parameters is represented by x; and the decision variables are represented by x (departure time, charging time, etc.).

[0043] As a preferred solution for an event-driven intelligent demand response and dispatching device for electric official vehicles, the differentiated constraint strategy in the vehicle optimization dispatching module is as follows: rigid time window constraints are adopted for emergency events to ensure that the task completion time does not exceed the deadline; flexible start window constraints are adopted for temporary official events to allow adjustment of the task start time within a preset time range; and planned cycle constraints are adopted for routine events to ensure that all tasks are completed within a preset cycle.

[0044] The present invention has the following advantages:

[0045] First, data utilization is more efficient: We deeply mine the value of massive amounts of data on official vehicles, and through semantic modeling and clustering classification, we accurately capture the patterns and characteristics of different vehicle usage events, providing precise data support for dispatching decisions.

[0046] Second, the scheduling strategy is more precise: differentiated modeling and constraint strategies are formulated for the three types of events to balance emergency response and cost optimization, and avoid vehicle idleness and task conflicts.

[0047] Third, the economic benefits are more significant: by combining CTSGAN price prediction with off-peak charging optimization, charging costs are greatly reduced, while robust optimization improves vehicle utilization and reduces operating expenses.

[0048] Fourth, it has stronger adaptability to different scenarios: it fully conforms to the characteristics of official vehicle management regulations and data integrity, while taking into account the power grid demand response requirements, and its adaptability and practicality are outstanding.

[0049] Fifth, higher decision reliability: The model is screened by dual verification of AIC criteria and utility function, combined with scenario reduction and robust optimization, to ensure that the scheduling scheme is anti-interference and stable and reliable. Attached Figure Description

[0050] To more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings in the following description are merely exemplary, and those skilled in the art can derive other embodiments based on the provided drawings without creative effort.

[0051] The structures, proportions, sizes, etc. illustrated in this specification are only for the purpose of assisting those skilled in the art in understanding and reading the content disclosed herein, and are not intended to limit the conditions under which the present invention can be implemented. Therefore, they have no substantial technical significance. Any modifications to the structure, changes in the proportions, or adjustments to the size, without affecting the effects and objectives that the present invention can produce, should still fall within the scope of the technical content disclosed in the present invention.

[0052] Figure 1This is a flowchart illustrating an event-driven intelligent demand response and scheduling method for electric official vehicles provided in Embodiment 1 of the present invention.

[0053] Figure 2 This is a schematic diagram illustrating the specific implementation process of an event-driven intelligent demand response and scheduling method for electric official vehicles provided in Embodiment 1 of the present invention;

[0054] Figure 3 This is a schematic diagram illustrating the distribution of vehicle dispatch time for vehicle use events in one possible embodiment of Embodiment 1 of the present invention;

[0055] Figure 4 This is a schematic diagram of the travel distance distribution of official vehicle events in one possible embodiment of Embodiment 1 of the present invention;

[0056] Figure 5 This is a schematic diagram of the distribution range of driving distances for official vehicle events in one possible embodiment of Embodiment 1 of the present invention;

[0057] Figure 6 This is a schematic diagram illustrating the proportion of events related to official vehicle use in one possible embodiment of Embodiment 1 of the present invention.

[0058] Figure 7 This is a schematic diagram illustrating the effect of generating a government vehicle event scene in one possible embodiment of Embodiment 1 of the present invention.

[0059] Figure 8 This is a schematic diagram of a spot electricity price simulation in one possible embodiment provided in Embodiment 1 of the present invention;

[0060] Figure 9 This is a schematic diagram illustrating the adjustment of the official vehicle charging plan in one possible embodiment of Embodiment 1 of the present invention;

[0061] Figure 10 This is a schematic diagram comparing key indicators before and after bus dispatch optimization in one possible embodiment of Embodiment 1 of the present invention;

[0062] Figure 11 This is a schematic diagram of the architecture of an event-driven intelligent demand response and scheduling device for electric official vehicles provided in Embodiment 2 of the present invention. Detailed Implementation

[0063] The following specific embodiments illustrate the implementation of the present invention. Those skilled in the art can easily understand other advantages and effects of the present invention from the content disclosed in this specification. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. 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.

[0064] Example 1

[0065] See Figure 1 and Figure 2 Embodiment 1 of the present invention provides an event-driven intelligent demand response and scheduling method for electric official vehicles, comprising the following steps:

[0066] S1. By collecting historical vehicle usage event data, semantic modeling is performed on unstructured text information to extract multi-dimensional features; a structured feature vector is constructed based on the multi-dimensional features; the structured feature vector is clustered using the K-means algorithm to form an event database;

[0067] S2. Based on the event database, the vehicle usage time and travel mileage of the set events are classified and modeled by multi-distribution adaptive fitting and CIA selection strategy, and the statistical characteristics and distribution of the set events are output. Based on the statistical characteristics and distribution of the set events, a set of highly representative vehicle usage event scenarios is generated by Monte Carlo sampling strategy and backward scenario reduction method.

[0068] S3. By collecting historical prices in the spot market and external environment information, a set of input variables is constructed; the input variables in the set of input variables are processed by a conditional time series generator adversarial network to output the future electricity price scenario;

[0069] S4. Based on the set of highly representative vehicle usage events and the future electricity price scenario, and combined with vehicle battery level and charging pile distribution information, construct a multi-day scheduling objective function; based on a differentiated constraint strategy, solve the multi-day scheduling objective function using a robust optimization algorithm, and output the charging time period and scheduling order.

[0070] In this embodiment, in step S1, unstructured text information is semantically modeled and multi-dimensional features are extracted by collecting historical vehicle usage event data; a structured feature vector is constructed based on the multi-dimensional features; and the structured feature vector is clustered using the K-means algorithm to form an event database.

[0071] Specifically, historical data on official vehicle usage events are collected, and the original vehicle usage request text s is processed by removing stops and segmenting words to obtain the word sequence structure as shown in equation (1):

[0072] (1)

[0073] In the formula, S is the word sequence structure; This is the word segmentation vector.

[0074] And based on expert knowledge, a keyword semantic dependency matrix is ​​constructed as shown in equation (2):

[0075] (2)

[0076] In the formula, D is the dependency matrix.

[0077] The pre-trained language model BERT is used to extract sentence vectors, converting the text into a high-dimensional semantic vector h:

[0078] (3)

[0079] A multi-task attribute classification model is constructed. The semantic recognition task includes event urgency, delay tolerance, predictability, and task type, such as emergency repair / inspection / official business / delivery. A multi-task neural network (MTL) is used to output multiple attributes simultaneously.

[0080] (4)

[0081] In the formula, Let be the predicted probability distribution of the k-th semantic attribute; This is the weight matrix corresponding to the k-th semantic attribute classifier; is the bias vector corresponding to the k-th semantic attribute classifier; h is the text semantic feature vector extracted based on BERT.

[0082] The losses for each task are shown in equation (5):

[0083] (5)

[0084] In the formula, The loss function value for the task of recognizing the k-th semantic attribute; The total number of categories contained in the k-th semantic attribute; The true label for the c-th category in the k-th attribute; This represents the predicted probability of the c-th category in the k-th attribute.

[0085] Form a structured event feature vector; for each event Constructing feature vectors:

[0086] (6)

[0087] In the formula, x is the eigenvector; The urgency level of the event (1 = normal, 2 = important, 3 = urgent). For the predictability of events; Event frequency; This represents the average mileage traveled.

[0088] Then, in order to build a generalizable event type library, K-means clustering is used to cluster the structured event vectors based on the urgency and delayability of the events. The objective function is:

[0089] (7)

[0090] In the formula, Let be the centroid of the k-th type of event cluster; Let be the feature vector of event i; N is the total number of events.

[0091] Based on the clustering results, the events are divided into emergency repair events, routine inspection events, and temporary official business events.

[0092] Among them, emergency repair incidents (high urgency, low predictability): These generally occur suddenly, have no obvious regularity, are usually not notified to the official vehicle management system in advance, and require a quick response, exhibiting the characteristics of high urgency and low predictability.

[0093] Routine inspection events (low urgency, high predictability): These events involve the daily work of government vehicles. Generally, the vehicle is used in advance and there is a relatively clear task completion plan. They generally do not require prediction and are characterized by low urgency and high predictability.

[0094] Temporary official business events (medium urgency, medium predictability): These events are generally sudden official business events, usually notified in advance, and the events can be delayed to a relatively high degree, exhibiting the characteristics of medium urgency and medium predictability.

[0095] Ultimately, the clustering output forms an event feature library, which serves as input for subsequent modeling.

[0096] In this embodiment, in step S2, based on the event database, the vehicle usage time and travel mileage of the set events are classified and modeled by multi-distribution adaptive fitting and CIA optimization strategy, and the statistical characteristics and distribution of the set events are output; based on the statistical characteristics and distribution of the set events, a set of highly representative vehicle usage event scenarios is generated by Monte Carlo sampling strategy and backward scenario reduction method.

[0097] Specifically, a multi-distribution fitting strategy is adopted for the vehicle usage time and mileage requirements of different types of events, and the optimal model is selected through AIC. Then, Monte Carlo sampling is used to generate future demand scenarios.

[0098] First, a mixture probability density function f(X) composed of a weighted combination of m basic distributions is fitted to the uncertain characteristic data:

[0099] (8)

[0100] In the formula, As weight, Based on the probability distribution (such as normal, lognormal, Poisson, etc.); These are the distribution parameters.

[0101] The CIA (Computer-Aided Institutional Analytical) optimal classification modeling strategy is adopted. Based on the fitted distribution and event characteristics, a utility function U is introduced for classification evaluation, and the optimal model parameters are selected. Output the statistical characteristics and distribution of each type of event, such as car usage time and travel mileage. .

[0102] Specifically, the fitting effect of the mixed probability density function is evaluated by the AIC criterion, and the AIC value corresponding to each mixed probability density function is calculated; several optimal fitting candidate models are selected based on the AIC values; each optimal fitting candidate model is evaluated and screened by the utility function to obtain the optimal fitting model; and the statistical characteristics and distribution of the set event are output through the optimal fitting model.

[0103] In this embodiment, during the process of generating the highly representative set of car-use event scenarios using a Monte Carlo sampling strategy and a backward scenario reduction method, a set number of samples are randomly selected from the statistical features and distribution using the Monte Carlo sampling strategy to generate an initial set of car-use event scenarios. Based on the initial set of car-use event scenarios, the Wasserstein distance between any two scenarios is calculated using the backward scenario reduction method, and redundant scenarios corresponding to the minimum probability distance are iteratively deleted until the number of scenarios reaches the set target, thereby generating the highly representative set of car-use event scenarios.

[0104] Specifically, based on the characteristic distribution of each type of event, a Monte Carlo sampling strategy is adopted to sample the distribution. Randomly selected from A set of sample points is used to generate an initial set of car usage event scenarios. :

[0105] (9)

[0106] Furthermore, the backward scene reduction method is employed to calculate the Wasserstein distance between any two scenes. .

[0107] Iterative calculation for each scene minimum probability distance :

[0108] (10)

[0109] Delete Minimal scene and its probability Transfer to the nearest scene .

[0110] Repeat the above steps until the number of scenarios reaches the preset target, and finally output a highly representative set of car use event scenarios.

[0111] In this embodiment, in step S3, a set of input variables is constructed by collecting historical prices in the spot market and external environmental information; the input variables in the set of input variables are processed by a time series generator adversarial network to output the future electricity price scenario.

[0112] Specifically, this step mainly targets the needs of large government agencies participating in the spot market. In response to the volatility of spot prices, a spot price prediction framework based on Conditional Time Series Generative Adversarial Network (CTSGAN) is constructed as a price reference for subsequently formulating the optimal charging strategy.

[0113] First, collect historical price data from the spot market. and external environment information Construct the set of input variables :

[0114] (11)

[0115] In the formula, Historical electricity price series; For temperature; This refers to wind speed.

[0116] The network consists of a generator and a discriminator. The generator G takes random noise and condition C as input and outputs a predicted electricity price sequence. The discriminator inputs an electricity price sequence. And condition C, output probability .

[0117] The objective function for adversarial training is as follows:

[0118] (12)

[0119] In the formula, This is a gradient penalty term; To generate the target value function for the adversarial network; The expected score is the score of the true sample distribution; To generate the expected score of the sample distribution; represents the weight coefficient of the gradient penalty term.

[0120] Ultimately, a set of future electricity price scenarios generated based on condition C is obtained.

[0121] In this embodiment, in step S4, based on the set of highly representative vehicle usage events and the future electricity price scenario, and combined with vehicle battery power and charging pile distribution information, a multi-day scheduling objective function is constructed; based on the differentiated constraint strategy, the multi-day scheduling objective function is solved by a robust optimization algorithm, and the charging time period and scheduling order are output.

[0122] Specifically, the goal of this step is to establish a multi-day, multi-vehicle scheduling model that minimizes total operating costs while ensuring priority response to emergencies, taking into account the worst-case scenario of uncertainty in vehicle usage events and electricity prices.

[0123] Considering the characteristics of different types of events, the models for different events are shown in Table 1:

[0124] Table 1. Multi-type event modeling methods

[0125] In the scheduling model, all uncertain parameters It can be represented as the mean. Add an uncertainty set Deviation within :

[0126] (13)

[0127] In the formula, Let represent the disturbance level of the j-th type of uncertainty factor in the i-th scenario.

[0128] The objective function includes charging costs and penalty terms for multiple types of events:

[0129] (14)

[0130] In the formula, For charging costs; The penalty cost is represented by w; the uncertain parameters of the scenario are represented by u; the set of uncertain parameters is represented by x; and the decision variables are represented by x (departure time, charging time, etc.).

[0131] Charging costs are directly affected by uncertain electricity prices. The impact reflects consideration of dynamic electricity pricing, enabling vehicles to charge during off-peak hours as much as possible.

[0132] (15)

[0133] In the formula, The amount of charge generated by vehicle v during period t.

[0134] The penalty cost reflects the consequences of failing to respond or complete a task in a timely manner, and the penalty coefficient is used to characterize the urgency of different types of events:

[0135] (16)

[0136] In the formula, Penalty coefficient for the amount of time taken to complete an emergency event beyond the deadline. Larger; The penalty coefficient is calculated based on the amount of time a temporary official task starts exceeding the maximum cache time. Moderate; Penalty coefficient for exceeding the planned end time for daily events. Lower; Penalties for exceeding deadlines in emergency situations; The penalty cost for temporary official tasks starting when the maximum cache time is exceeded; Penalty costs for completing routine events beyond the planned end time.

[0137] In this embodiment, considering the completion time constraints of differentiated tasks, event windows for the three types of events are modeled separately.

[0138] Emergency events employ rigid event windows and are validated using the worst-case upper bound of uncertain parameters to ensure task completion time. It will not exceed the minimum time limit :

[0139] (17)

[0140] In the formula, The start time for completing the task; The time of arrival at the mission location; The time required to complete the task.

[0141] Temporary official tasks adopt a flexible start window model, which allows for flexible adjustment of task start time within the window period to serve overall cost optimization.

[0142] (18)

[0143] In the formula, The earliest completion time; This is the deadline for the task.

[0144] Routine events only need to be completed within the planned timeframe.

[0145] At the same time, it is essential to ensure that the vehicle's State of Charge (SOC) remains sufficient in all usage scenarios:

[0146] (19)

[0147] (20)

[0148] In the formula, The vehicle's battery consumption at time t depends on the mission's travel distance L. Let V be the battery charge of vehicle v at time t; , These are the minimum and maximum battery levels. For charging efficiency; The charging power for the vehicle; For vehicle battery capacity; This is an event execution variable.

[0149] To ensure that each vehicle usage event can be assigned to one and only one vehicle for execution, and that the event can only be initiated once, a unique constraint is set for task assignment:

[0150]

[0151] Through this differentiated modeling and robust transformation, the model ensures the highest priority and feasibility for emergency events, while leveraging the flexibility of routine and temporary official events to minimize empty running rates and charging costs, thus achieving the invention objective of efficient and economical dispatch.

[0152] In one possible embodiment, in order to verify the official vehicle demand forecasting model, the model's effectiveness in dimensions such as "improvement rate of official vehicle use efficiency", "reduction rate of official vehicle operating electricity costs" and "timeliness of response to official travel" is evaluated by combining electricity price fluctuations and the frequency of official event types. This provides a supporting basis for forming an official vehicle time-sharing scheduling strategy that adapts to the demand for official travel.

[0153] I. Analysis of the Official Vehicle Incident:

[0154] Due to their controllable official travel routes and relatively regular nighttime charging habits, official vehicles exhibit high responsiveness and significant adjustability as load adjustment targets. The official vehicle fleet of the Distribution Network Asset Department in Futian District, Shenzhen, experiences numerous usage incidents, high frequency of use, and diverse task types, offering considerable dispatch flexibility and thus representing a typical official vehicle usage scenario.

[0155] The first-level potential assessment of official vehicles was carried out based on the real operation data of official vehicles in the power distribution asset department of Futian District, Shenzhen from October to December 2024. Through the classification and analysis of vehicle use events, it was verified that on the premise of ensuring the rigid demand for official trips, the official vehicle fleet has significant adjustment potential. This assessment included a total of 10 vehicles, namely Yue BDK4728, Yue BAH6988, Yue BAG8881, Yue BB12270, Yue BBB5809, Yue BDD8063, Yue BAM0283, Yue BAG6896, Yue BB06183. By analyzing the frequencies of different types of vehicle use events, 507 vehicle use events can be divided into daily, necessary, and emergency types according to the characteristics of vehicle use events. The specific definitions and classifications are shown in Table 2:

[0156] type definition event quantity Daily This refers to specific business targets that need to be met within a certain timeframe. "Inspection", "Check", "Payment Reminder", "Meter Check", "Visit", "Document Delivery", "Routine", "Interim Inspection", "Morning Exercise", "Survey", "Work", "Verification", "Vehicle", "Meter Check", "Electricity Bill Collection", "Car Wash", "On-site Survey", "Legal Publicity", "Business Expansion On-site Survey", "Cable Relocation and Power Supply", "On-site Survey", "Attending the Distribution Network Department's Distribution Video Surveillance Construction Promotion Meeting" 328 Emergency Unpredictable, sudden situations requiring vehicle use within a short period of time "Emergency Repair", "Fault", "Power Supply", "Sudden Event", "Emergency Response", "Urgent", "Troubleshooting" 19 Necessary Generally, you need to use the car on a fixed date, which can be known in advance. "Meeting," "Evaluation," "Training," "Reception," "Examination," "Acceptance," "Operation," "Test," "Power Supply Guarantee," "Special Project," "Customer Issues," "Outing," "On Duty," "Replacement," "Power Supply," "Legal Publicity," "District Committee Meeting," "Business Trip," "Delivering Documents," "Sending Personnel," "Visiting the National Key Project Application Department and Distribution Network Department to Discuss Equipment Relocation to Ground for Communication," "Attending the Distribution Network Department's Distribution Video Monitoring Construction Promotion Meeting," "Inspection of the Bauhinia Demonstration Project" 160

[0157] Table 2 Semantic Table of Official Vehicle Use Events

[0158] From the perspective of the proportion of event quantity, daily events account for 64.7% of the total, necessary events account for 31.6%, and emergency events only account for 3.7%. The vast majority of tasks are non-emergency official duties, with strong schedulability and response potential.

[0159] Among them, the driving duration statistics of each type of event are as Figure 3 shown, and basically concentrated within 8 hours. Although there are some high-duration outliers in daily tasks, the overall still shows strong time concentration, which is conducive to achieving group response and time shift.

[0160] The driving mileage statistics of each type of event are as Figure 4 and Figure 5 shown, and basically distributed within 10 - 50 km within 8 hours. Among them, about 68.2% of the tasks have a driving distance within 30 kilometers, and only 22.9% of the tasks exceed 100 kilometers. The mileage of emergency tasks is generally short. Although the daily tasks have a large span, they are still mainly medium and short distances, with obvious spatial planning.

[0161] Figure 6 shows the proportion of various official vehicle use events in terms of time and space. Among them, the driving mileage and time of emergency events are generally short, while the driving time and mileage of daily events are relatively long.

[0162] Overall, the official vehicle fleet shows highly schedulable characteristics in terms of travel time, driving distance, and task structure. The vast majority of tasks have the potential for time shift and response adjustment. Therefore, this fleet can be used as a stable and flexible adjustment resource in the power distribution system to effectively support load optimization control on the basis of ensuring necessary official trips.

[0163] II. Prediction of Vehicle Use Events:

[0164] like Figure 7 As shown, during the demonstration period, the prediction model achieved a 95% accuracy rate in predicting demand for fixed-frequency events such as "daily office commuting" and "cross-departmental meetings," and a 100% adaptability rate in responding to demand for sudden events such as "emergency official business." The bus dispatch scenario generated by the model can cover 100% of the official travel time demand, 98.6% of the travel mileage, and a comprehensive coverage of 99.3%.

[0165] III. Market Price Forecast:

[0166] like Figure 8 As shown, the proposed Conditional Time Series Generative Adversarial Network (CTSGAN) method predicts electricity price scenarios. With multiple days of electricity price scenario information and weather information as input, the average relative error of the obtained electricity price scenario is 10.79%, of which the prediction error during the peak period is 14.56%. However, the overall electricity price trend is basically predicted accurately, which can support the next step of optimization.

[0167] IV. Optimization Algorithm Practice:

[0168] The proposed optimization algorithm enables the transfer and scheduling of multiple official vehicles during different dates and charging periods. During a three-day demonstration period, Figure 9 The data shows the charging schedule and power allocation for each vehicle on different dates. Charging activity has clearly shifted to periods of low grid load and has not conflicted with the needs of official travel.

[0169] The Gantt chart clearly shows that the charging behavior of the eight official vehicles effectively shifted during the three-day demonstration period: for example, vehicle Yue BAG8896 was charged during multiple off-peak hours, such as 05:00 on day 1, 04:00 on day 2, and 23:00 on day 3, with a single charge reaching 23.8–55.5 kWh. The bar chart of daily charging time distribution further shows that in the later stages of the scheduling adjustment, the number of charges was relatively evenly distributed throughout the night, and the number of vehicles charging was generally kept at a low level, successfully avoiding concentrated charging loads at night.

[0170] like Figure 10 As shown in Table 3, the comparison of key indicators before and after the scheduling optimization shows that the proposed scheduling method significantly increases the charging volume during off-peak hours by 120 kWh, reflecting the effective transfer of load from peak and flat periods to off-peak periods. At the same time, the average daily charging cost is reduced by 51.1 yuan, resulting in a total arbitrage income increase of 154.2 yuan.

[0171] Electricity Market Types Peak Shift Ratio Transfer load Arbitrage income / yuan Electricity spot market 100% 120kwh 154.2

[0172] Table 3. Electricity Spot Market Charging Load Transfer for Government Vehicles

[0173] This scheduling result shows that, while ensuring the needs of official vehicles are met, the charging process for official vehicles can be significantly shifted to periods with lower electricity prices, thus reducing charging costs.

[0174] The application scenarios of this invention are as follows:

[0175] When applied to the dispatching of official vehicles in government agencies and public institutions, this invention can formulate differentiated strategies for various scenarios such as meeting attendance and official visits, ensuring priority response to emergency tasks, while optimizing charging plans in line with off-peak electricity prices to reduce fleet operating costs.

[0176] When applied to special dispatching by power supply bureaus and power operation and maintenance companies, this invention can accurately match the characteristics of tasks such as fault repair and line inspection, ensure the timeliness of emergency response through rigid constraints, and simultaneously adapt to the grid demand response requirements to reduce charging expenses.

[0177] When applied to the management of official travel for large state-owned enterprises / central enterprises, this invention can integrate data from scenarios such as office commuting and customer reception, improve vehicle utilization through intelligent scheduling, avoid idle waste, and help enterprises achieve their goals of cost reduction and efficiency improvement.

[0178] When applied to load aggregators' optimization decisions, this invention can integrate the resources of multiple fleets of government vehicles, participate in the grid demand response market through off-peak charging and flexible scheduling, and provide core technical support for aggregators to obtain arbitrage profits.

[0179] When applied to centralized official travel services in industrial parks / new districts, this invention can realize the sharing and scheduling of vehicle resources within the region, adapt to scenarios such as joint official business in industrial parks and government outreach to rural areas, and optimize regional charging layout and travel efficiency.

[0180] When applied to the dispatch of special vehicles by emergency management departments, this invention ensures the response to emergencies such as disaster reconnaissance and material transportation through rigid time window constraints, while rationally planning charging periods to ensure stable vehicle range.

[0181] It should be noted that the method of this disclosure embodiment can be executed by a single device, such as a computer or server. The method of this embodiment can also be applied to a distributed scenario, where multiple devices cooperate to complete the task. In such a distributed scenario, one of these devices may execute only one or more steps of the method of this disclosure embodiment, and the multiple devices will interact with each other to complete the method described.

[0182] It should be noted that the above description describes some embodiments of this disclosure. Other embodiments are within the scope of the appended claims. In some cases, the actions or steps recorded in the claims can be performed in a different order than that shown in the above embodiments and still achieve the desired result. Furthermore, the processes depicted in the drawings do not necessarily require a specific or sequential order to achieve the desired result. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0183] Example 2

[0184] See Figure 11 Embodiment 2 of the present invention also provides an event-driven intelligent demand response and scheduling device for electric official vehicles, comprising:

[0185] The vehicle usage data analysis module 001 is used to collect historical vehicle usage event data, perform semantic modeling on unstructured text information, extract multi-dimensional features, construct structured feature vectors based on the multi-dimensional features, and cluster the structured feature vectors using the K-means algorithm to form an event database.

[0186] The vehicle demand prediction module 002 is used to classify and model the vehicle usage time and travel mileage of a set event based on the event database through multi-distribution adaptive fitting and CIA optimization strategy, and output the statistical characteristics and distribution of the set event; based on the statistical characteristics and distribution of the set event, a set of highly representative vehicle usage event scenarios is generated through Monte Carlo sampling strategy and backward scenario reduction method.

[0187] The electricity price prediction module 003 is used to construct a set of input variables by collecting historical prices in the spot market and external environmental information; the input variables in the set of input variables are processed by a generative adversarial network for time series conditions to output future electricity price scenarios;

[0188] The bus optimization scheduling module 004 is used to construct a multi-day scheduling objective function based on the set of highly representative vehicle usage events and the future electricity price scenario, combined with vehicle battery power and charging pile distribution information; based on the differentiated constraint strategy, the multi-day scheduling objective function is solved by a robust optimization algorithm, and the charging time period and scheduling order are output.

[0189] In this embodiment, the vehicle data analysis module 001 performs semantic modeling on unstructured text information by removing stops, segmenting words, and constructing a keyword semantic dependency matrix from the original text; high-dimensional semantic vectors are extracted from the keyword semantic dependency matrix using the pre-trained language model BERT; the multi-dimensional features include: event urgency, predictability, monthly occurrence frequency, and average mileage; the event urgency is divided into three levels: ordinary, important, and urgent; the predictability is divided into three levels: high, medium, and low.

[0190] In this embodiment, in the vehicle usage data analysis module 001, during the clustering of the structured feature vectors using the K-means algorithm, the objective function of the K-means algorithm is expressed as:

[0191]

[0192] In the formula, Let be the centroid of the k-th type of event cluster; Let be the feature vector of event i; N is the total number of events.

[0193] In this embodiment, in the vehicle demand prediction module 002, during the process of classifying and modeling the vehicle usage time and travel mileage of a set event through multi-distribution adaptive fitting and CIA selection strategy, a mixed probability density function is constructed by weighting the normal distribution, log-normal distribution, and Poisson distribution; the fitting effect of the mixed probability density function is evaluated by the AIC criterion, and the AIC value corresponding to each mixed probability density function is calculated; several optimal fitting candidate models are selected based on the AIC values; each optimal fitting candidate model is evaluated and screened by the utility function to obtain the optimal fitting model; and the statistical characteristics and distribution of the set event are output through the optimal fitting model.

[0194] In this embodiment, in the process of generating the highly representative set of car usage event scenarios through the Monte Carlo sampling strategy and the backward scenario reduction method in the car usage demand prediction module 002, a set number of samples are randomly selected from the statistical features and distributions through the Monte Carlo sampling strategy to generate an initial set of car usage event scenarios; based on the initial set of car usage event scenarios, the Wasserstein distance between any two scenarios is calculated through the backward scenario reduction method, and redundant scenarios corresponding to the minimum probability distance are iteratively deleted until the number of scenarios reaches the set target, thereby generating the highly representative set of car usage event scenarios.

[0195] In this embodiment, the electricity price prediction module 003 includes a conditional time series generative adversarial network (GAN) comprising a generator and a discriminator. The generator outputs a predicted electricity price sequence based on the input variables and random noise. The discriminator outputs the probability of the authenticity of the electricity price sequence based on the predicted electricity price sequence. The prediction error is minimized through adversarial training, and the future electricity price scenario is output.

[0196] The objective function expression for adversarial training is:

[0197]

[0198] In the formula, This is a gradient penalty term; To generate the target value function for the adversarial network; The expected score is the score of the true sample distribution; To generate the expected score of the sample distribution; represents the weight coefficient of the gradient penalty term.

[0199] In this embodiment, the expression for the multi-day scheduling objective function in the bus optimization scheduling module 004 is:

[0200]

[0201] In the formula, For charging costs; To incur penalties;

[0202] In this embodiment, the differentiated constraint strategy in the bus optimization scheduling module 004 is as follows: for emergency events, a rigid time window constraint is adopted to ensure that the task completion time does not exceed the deadline; for temporary official events, a flexible start window constraint is adopted to allow the task start time to be adjusted within a preset time range; and for routine events, a planning cycle constraint is adopted to ensure that all tasks are completed within a preset cycle.

[0203] It should be noted that the information interaction and execution process between the modules of the above system are based on the same concept as the method embodiment in Embodiment 1 of this application, and the resulting technical effects are the same as those in the method embodiment of this application. For details, please refer to the description in the method embodiment shown above in this application, and it will not be repeated here.

[0204] Example 3

[0205] Embodiment 3 of the present invention provides a non-transitory computer-readable storage medium storing program code for an event-driven intelligent demand response and scheduling method for electric official vehicles. The program code includes instructions for executing the event-driven intelligent demand response and scheduling method for electric official vehicles according to Embodiment 1 or any possible implementation thereof.

[0206] Computer-readable storage media can be any available medium that a computer can access, or a data storage device such as a server or data center that integrates one or more available media. The available medium can be magnetic media (e.g., floppy disks, hard disks, magnetic tapes), optical media (e.g., DVDs), or semiconductor media (e.g., solid-state drives (SSDs)).

[0207] Example 4

[0208] Embodiment 4 of the present invention provides an electronic device, including: a memory and a processor;

[0209] The processor and the memory communicate with each other via a bus; the memory stores program instructions that can be executed by the processor, and the processor can execute an event-driven intelligent demand response and scheduling method for electric official vehicles according to Embodiment 1 or any possible implementation thereof.

[0210] Specifically, a processor can be implemented in hardware or software. When implemented in hardware, the processor can be a logic circuit, an integrated circuit, etc. When implemented in software, the processor can be a general-purpose processor that reads software code stored in memory. This memory can be integrated into the processor or located outside the processor and exist independently.

[0211] In the above embodiments, implementation can be achieved, in whole or in part, through software, hardware, firmware, or any combination thereof. When implemented in software, it can be implemented, in whole or in part, as a computer program product. The computer program product includes one or more computer instructions. When the computer program instructions are loaded and executed on a computer, all or part of the processes or functions described in the embodiments of the present invention are generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable system. The computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, the computer instructions can be transmitted from one website, computer, server, or data center to another via wired (e.g., coaxial cable, fiber optic, digital subscriber line (DSL)) or wireless (e.g., infrared, wireless, microwave, etc.) means.

[0212] It is obvious to those skilled in the art that the modules or steps of the present invention described above can be implemented using general-purpose computing systems. They can be centralized on a single computing system or distributed across a network of multiple computing systems. Optionally, they can be implemented using program code executable by a computing system, thereby storing them in a storage system for execution by the computing system. In some cases, the steps shown or described can be performed in a different order than those presented herein, or they can be fabricated as separate integrated circuit modules, or multiple modules or steps can be fabricated as a single integrated circuit module. Thus, the present invention is not limited to any particular combination of hardware and software.

[0213] Although the present invention has been described in detail above with general descriptions and specific embodiments, modifications or improvements can be made to it, which will be obvious to those skilled in the art. Therefore, all such modifications or improvements made without departing from the spirit of the present invention fall within the scope of protection claimed by the present invention.

Claims

1. An event-driven intelligent demand response and scheduling method for electric official vehicles, characterized in that, include: By collecting historical vehicle usage data, semantic modeling is performed on unstructured text information to extract multi-dimensional features; Construct a structured feature vector based on the aforementioned multi-dimensional features; The structured feature vectors are clustered using the K-means algorithm to form an event database; Based on the event database, the vehicle usage time and travel mileage of the set events are classified and modeled by multi-distribution adaptive fitting and CIA selection strategy, and the statistical characteristics and distribution of the set events are output. Based on the statistical characteristics and distribution of the set events, a set of highly representative vehicle usage event scenarios is generated by Monte Carlo sampling strategy and backward scenario reduction method. By collecting historical prices in the spot market and information about the external environment, a set of input variables is constructed; the input variables in the set of input variables are then processed by a conditional time series generator adversarial network to output future electricity price scenarios. Based on the set of highly representative vehicle usage events and the future electricity price scenario, and combined with vehicle battery power and charging pile distribution information, a multi-day scheduling objective function is constructed. Based on the differentiated constraint strategy, the multi-day scheduling objective function is solved by a robust optimization algorithm, and the charging time period and shift order are output.

2. The event-driven intelligent demand response and scheduling method for electric official vehicles according to claim 1, characterized in that, When performing semantic modeling on unstructured text information, the original text is removed from stops, segmented into words, and a keyword semantic dependency matrix is ​​constructed. High-dimensional semantic vectors are extracted from the keyword semantic dependency matrix using the pre-trained language model BERT. The multi-dimensional features include: event urgency, predictability, monthly frequency of occurrence, and average mileage; The urgency of the events is divided into three levels: ordinary, important, and urgent; the predictability is divided into three levels: high, medium, and low.

3. The event-driven intelligent demand response and scheduling method for electric official vehicles according to claim 2, characterized in that, In the process of clustering the structured feature vectors using the K-means algorithm, the objective function of the K-means algorithm is expressed as: ； In the formula, Let be the centroid of the k-th type of event cluster; Let be the feature vector of event i; N is the total number of events.

4. The event-driven intelligent demand response and scheduling method for electric official vehicles according to claim 3, characterized in that, In the process of classifying and modeling the vehicle usage time and travel mileage of a given event through multi-distribution adaptive fitting and CIA optimization strategy, a mixed probability density function is constructed by weighting the normal distribution, log-normal distribution and Poisson distribution. The fitting effect of the mixed probability density function is evaluated by the AIC criterion, and the AIC value corresponding to each mixed probability density function is calculated. Based on the AIC value, several optimal fitting candidate models are selected; each optimal fitting candidate model is evaluated and screened using a utility function to obtain the optimal fitting model; and the statistical characteristics and distribution of the set event are output using the optimal fitting model.

5. The event-driven intelligent demand response and scheduling method for electric official vehicles according to claim 4, characterized in that, In the process of generating the highly representative set of car use event scenarios through Monte Carlo sampling strategy and backward scenario reduction method, a set number of samples are randomly drawn from the statistical features and distribution through Monte Carlo sampling strategy to generate the initial set of car use event scenarios. Based on the initial set of car-use event scenarios, the Wasserstein distance between any two scenarios is calculated using the backward scenario reduction method. Redundant scenarios corresponding to the minimum probability distance are iteratively deleted until the number of scenarios reaches the set target, thereby generating the highly representative set of car-use event scenarios.

6. The event-driven intelligent demand response and scheduling method for electric official vehicles according to claim 5, characterized in that, The conditional time series generative adversarial network includes a generator and a discriminator; the generator outputs a predicted electricity price sequence based on the input variables and random noise; the discriminator outputs the probability of the authenticity of the electricity price sequence based on the predicted electricity price sequence. The prediction error is minimized through adversarial training, and the future electricity price scenario is output. The objective function expression for adversarial training is: ； In the formula, This is a gradient penalty term; To generate the target value function for the adversarial network; The expected score is the score of the true sample distribution; To generate the expected score of the sample distribution; represents the weight coefficient of the gradient penalty term.

7. The event-driven intelligent demand response and scheduling method for electric official vehicles according to claim 6, characterized in that, The expression for the multi-day scheduling objective function is: ； In the formula, For charging costs; Let w be the penalty cost; w be the uncertain parameter of the scenario; u be the set of uncertain parameters; and x be the decision variable.

8. The event-driven intelligent demand response and scheduling method for electric official vehicles according to claim 7, characterized in that, The differentiated constraint strategy is as follows: for emergency events, a rigid time window constraint is adopted to ensure that the task completion time does not exceed the deadline; for temporary official events, a flexible start window constraint is adopted to allow the task start time to be adjusted within a preset time range; and for routine events, a planning cycle constraint is adopted to ensure that all tasks are completed within a preset cycle.

9. An event-driven intelligent demand response and scheduling device for electric official vehicles, employing the event-driven intelligent demand response and scheduling method for electric official vehicles as described in any one of claims 1-8, characterized in that, include: The vehicle usage data analysis module is used to perform semantic modeling on unstructured text information and extract multi-dimensional features by collecting historical vehicle usage event data; A structured feature vector is constructed based on the aforementioned multi-dimensional features; the structured feature vector is then clustered using the K-means algorithm to form an event database. The vehicle usage demand prediction module is used to classify and model the vehicle usage time and travel mileage of a set event based on the event database through multi-distribution adaptive fitting and CIA optimization strategy, and output the statistical characteristics and distribution of the set event; based on the statistical characteristics and distribution of the set event, a set of highly representative vehicle usage event scenarios is generated through Monte Carlo sampling strategy and backward scenario reduction method. The electricity price forecasting module is used to construct a set of input variables by collecting historical prices in the spot market and external environmental information; the input variables in the set of input variables are processed by a generative adversarial network based on time series conditions to output future electricity price scenarios; The bus optimization scheduling module is used to construct a multi-day scheduling objective function based on the set of highly representative vehicle usage events and the future electricity price scenario, combined with vehicle battery power and charging pile distribution information; Based on the differentiated constraint strategy, the multi-day scheduling objective function is solved by a robust optimization algorithm, and the charging time period and shift order are output.

10. The event-driven intelligent demand response and scheduling device for electric official vehicles according to claim 9, characterized in that, In the vehicle usage data analysis module, when performing semantic modeling on unstructured text information, the original text is de-stopped, segmented, and a keyword semantic dependency matrix is ​​constructed. High-dimensional semantic vectors are extracted from the keyword semantic dependency matrix using the pre-trained language model BERT; the multi-dimensional features include: event urgency, predictability, monthly frequency of occurrence, and average mileage; the event urgency is divided into three levels: ordinary, important, and urgent; the predictability is divided into three levels: high, medium, and low. In the vehicle usage data analysis module, during the clustering of the structured feature vectors using the K-means algorithm, the objective function of the K-means algorithm is expressed as: ； In the formula, Let be the centroid of the k-th type of event cluster; Let N be the feature vector of event i; N is the total number of events. In the vehicle demand prediction module, during the classification and modeling of vehicle usage time and travel mileage for a given event using multi-distribution adaptive fitting and CIA selection strategies, a mixed probability density function is constructed by weighting normal distribution, log-normal distribution, and Poisson distribution. The fitting effect of the mixed probability density function is evaluated using the AIC criterion, and the AIC value corresponding to each mixed probability density function is calculated. Several optimal fitting candidate models are selected based on the AIC values. Each optimal fitting candidate model is evaluated and screened using a utility function to obtain the optimal fitting model. The statistical characteristics and distribution of the given event are output using the optimal fitting model. In the vehicle demand prediction module, during the process of generating the highly representative vehicle event scenario set using the Monte Carlo sampling strategy and the backward scenario reduction method, a set number of samples are randomly selected from the statistical features and distribution using the Monte Carlo sampling strategy to generate an initial vehicle event scenario set; based on the initial vehicle event scenario set, the Wasserstein distance between any two scenarios is calculated using the backward scenario reduction method, and redundant scenarios corresponding to the minimum probability distance are iteratively deleted until the number of scenarios reaches the set target, thereby generating the highly representative vehicle event scenario set. In the electricity price prediction module, the conditional time series generative adversarial network includes a generator and a discriminator; the generator outputs a predicted electricity price sequence based on the input variables and random noise; the discriminator outputs the probability of the authenticity of the electricity price sequence based on the predicted electricity price sequence; and the prediction error is minimized through adversarial training to output the future electricity price scenario. The objective function expression for adversarial training is: ； In the formula, This is a gradient penalty term; To generate the target value function for the adversarial network; The expected score is the score of the true sample distribution; To generate the expected score of the sample distribution; represents the weight coefficient of the gradient penalty term.

11. In the bus optimization scheduling module, the expression for the multi-day scheduling objective function is: ； In the formula, For charging costs; The penalty cost is represented by w; the uncertain parameters of the scenario are represented by u; the set of uncertain parameters is represented by x; and the decision variable is represented by x. In the bus optimization scheduling module, the differentiated constraint strategy is as follows: for emergency events, a rigid time window constraint is adopted to ensure that the task completion time does not exceed the deadline; for temporary official events, a flexible start window constraint is adopted to allow the task start time to be adjusted within a preset time range; for routine events, a planning cycle constraint is adopted to ensure that all tasks are completed within a preset cycle.

Images