Vehicle defrosting method and system and vehicle

Abstract

The invention relates to a vehicle defrosting method and system and a vehicle, and belongs to the technical field of vehicle electronic control. The defrosting method comprises the steps that when a vehicle is in a flameout state, state data and environment sensing data of the vehicle are obtained; detecting whether the vehicle window is covered by frost and snow based on the environment sensing data; when it is detected that the vehicle window is covered by frost and snow, the expected vehicle using time of the user is calculated, and the vehicle using probability of the user within the expected vehicle using time is predicted; when the vehicle using probability is larger than a triggering threshold value, a defrosting strategy is obtained based on the environment sensing data, the state data and the expected vehicle using time; based on the defrosting strategy, the vehicle is controlled to execute the operation of removing frost and snow on vehicle windows. Triggering and dynamic scheduling of the defrosting behavior are achieved, invalid energy consumption is avoided, and meanwhile it is ensured that a user obtains a clear view before actually using a vehicle. The whole process does not need manual intervention, and the safety and convenience of vehicle use in winter are remarkably improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of vehicle electronic control, in particular to a vehicle defrosting method and system and a vehicle. BACKGROUND

[0002] In cold seasons, especially in high-latitude or high-altitude areas, when the vehicle is parked outdoors, the ambient temperature at night or in the early morning often drops below freezing point, and the water vapor in the air will condense on the low-temperature window surface, directly forming a frost layer. In addition, when it snows, snowflakes will directly accumulate on the window. Moreover, after the vehicle is turned off, the moisture remaining in the vehicle may also cool and condense on the window, further increasing the degree of frosting.

[0003] The snow or frost on the window can seriously hinder the user's vision and directly affect driving safety. Once the front windshield is covered with frost and snow, the user cannot clearly observe the road ahead, traffic signals, and the surrounding environment, which will greatly increase the risk of accidents during starting, lane changing, and braking. Therefore, the user must remove the frost and snow on the window before driving.

[0004] At present, the common defrosting and snow removal methods mainly include manual removal or turning on the vehicle's warm air heating system. Although manual removal has low cost, it has poor experience in severe cold weather and has the risk of scratching the glass or paint. Using engine waste heat or air conditioning warm air for defrosting requires starting the vehicle first, and the response is relatively slow. Moreover, regardless of the existing technology, it is difficult to clean in time. In order to remove the frost and snow, the user has to go out and manually handle it in advance, which not only consumes time and effort, but also may face health risks due to exposure to low-temperature environments, which is particularly disadvantageous for the elderly or the weak. SUMMARY

[0005] The present application provides a vehicle defrosting method, system and vehicle, which aims to automatically identify the frost and snow state on the window through the camera and other devices of the vehicle, and combine the vehicle start time prediction model to calculate the user's driving time, and then obtain a defrosting strategy that adapts to the current environment and user habits, so as to control the vehicle to perform defrosting operation at the right time. The present application effectively solves the problems of not timely cleaning of snow or frost on the window, relying on manual intervention, and increasing the user's burden in winter driving in the prior art, and significantly improves the safety and user experience of winter driving.

[0006] To achieve the above-mentioned purpose, the present application adopts the following technical solutions: The present application provides a vehicle defrosting method, comprising: acquiring state data and environmental perception data of the vehicle when the vehicle is in an off state; detecting whether the window is covered with frost and snow based on the environmental perception data; calculate an expected vehicle use time of the user when the vehicle window is detected to be covered by frost and snow, and predict a vehicle use probability of the user at the expected vehicle use time; generate a defrosting strategy based on the environmental perception data, the state data and the expected vehicle use time when the vehicle use probability is greater than a triggering threshold; control the vehicle to perform an operation of removing frost and snow on the vehicle window based on the defrosting strategy.

[0007] In the above embodiment, the application obtains the state data and the environmental perception data of the vehicle when the vehicle is in an off state, judges whether the vehicle window is covered by frost and snow based on the environmental perception data, calculates an expected vehicle use time of the user when the vehicle window is detected to be covered by frost and snow, and predicts a vehicle use probability of the user at the expected vehicle use time. When the vehicle use probability is greater than a triggering threshold, the system generates a defrosting strategy based on the environmental perception data, the state data and the expected vehicle use time, and automatically performs a removing operation, thereby constructing an intelligent closed-loop control mechanism of perception-decision-execution. The mechanism realizes triggering and dynamic scheduling of the defrosting behavior, avoids invalid energy consumption, and ensures that the user obtains a clear view before actual vehicle use. The whole process does not need manual intervention, significantly improves the safety and convenience of winter vehicle use, and more importantly, through deep integration of user behavior prediction and the environmental state of the vehicle, the defrosting service is changed from passive response to intelligent prediction, thereby providing the user with a more energy-saving, more seamless and more active vehicle use experience. In addition, the application also discloses a vehicle defrosting system applied to a vehicle, the defrosting system comprising: a data acquisition module configured to obtain state data and environmental perception data of the vehicle when the vehicle is in an off state; a frost and snow detection module configured to detect whether the vehicle window is covered by frost and snow based on the environmental perception data; an expected vehicle use time calculation module configured to calculate an expected vehicle use time of the user when the vehicle window is detected to be covered by frost and snow; a vehicle use probability prediction module configured to predict a vehicle use probability of the user at the expected vehicle use time based on the expected vehicle use time; a strategy generation module configured to generate a defrosting strategy based on the environmental perception data, the state data and the expected vehicle use time when the vehicle use probability is greater than a triggering threshold; an execution control module configured to control the vehicle to perform an operation of removing frost and snow on the vehicle window based on the defrosting strategy.

[0008] In addition, the application also provides a vehicle comprising the defrosting system. BRIEF DESCRIPTION OF DRAWINGS

[0009] Figure 1 is a flowchart of a vehicle defrosting method provided by the application; Figure 2 is a flowchart of a method for calculating an expected use time of a user provided by an embodiment of the present application; Figure 3 is a flowchart of a method for predicting a use probability of a user at an expected use time provided by an embodiment of the present application; Figure 4 is a system architecture schematic diagram of a defrosting system provided by an embodiment of the present application; Figure 5 is a functional strategy flowchart of a defrosting system provided by an embodiment of the present application; Figure 6 is an architecture schematic diagram of a cloud platform server provided by an embodiment of the present application; Figure 7 is an architecture schematic diagram of data preprocessing of a defrosting system provided by an embodiment of the present application Figure 1 ; Figure 8 is an architecture schematic diagram of data preprocessing of a defrosting system provided by an embodiment of the present application Figure 2 ; Figure 9 is a decision flowchart of a defrosting system provided by an embodiment of the present application; Figure 10 is a feedback adjustment flowchart of a defrosting system provided by an embodiment of the present application. DETAILED DESCRIPTION

[0010] The meaning of "and / or" appearing throughout the text is to include three parallel solutions. For example, "A and / or B" includes A solution, or B solution, or A and B solutions.

[0011] In the following, the present application will be specifically described through exemplary embodiments. However, it should be understood that the elements, structures and features in one embodiment can also be beneficially combined into other embodiments without further description As shown in the accompanying Figures 1 to 3 , the present application provides a vehicle defrosting method applied to a vehicle, and the defrosting method comprises: S1, acquiring state data and environment perception data of the vehicle when the vehicle is in an off state.

[0012] It is worth mentioning that the vehicle will not be completely powered off after the engine is turned off, but will automatically enter a low-power monitoring state, namely the sentinel mode. This mode is triggered by the electronic control unit (ECU) of the vehicle when it detects that the ignition switch is off and the vehicle is in the parking state. After entering the sentinel mode, the vehicle control system will turn off non-critical functional modules to reduce overall power consumption, and only keep necessary sensors running, including cameras, ultrasonic radars, and environmental sensors, to collect real-time perception data of the external environment. At the same time, the vehicle communication module maintains 4G or 5G network connection to ensure that the perception data can be uploaded to the cloud in a timely manner and support interaction with the remote server. Through the above design, while ensuring environmental monitoring capability, the system power consumption is significantly optimized, providing stable and low-power data support for subsequent frost and snow identification, user travel intention prediction, and intelligent defrosting strategy generation.

[0013] In some embodiments of the present application, the state data of the vehicle includes vehicle ID, vehicle location information, vehicle window area, battery state of charge (SOC), vehicle parking state, vehicle door state, and ignition switch state, etc. These data are used to reflect the current physical properties and operating state of the vehicle, and can be obtained from the vehicle control system through the vehicle CAN bus.

[0014] In some embodiments, the environmental perception data includes image data of the vehicle window area, environmental temperature, environmental humidity, external wind speed, external air pressure, and ground snow depth, etc., which are obtained through the vehicle's own sensors.

[0015] S2, detecting whether the vehicle window is covered with frost and snow based on the environmental perception data.

[0016] Specifically, the vehicle's own high-definition camera is used to collect real-time images of the vehicle window area to obtain image data of the vehicle window area, and to detect whether there is snow accumulation or frost formation on the vehicle window based on the image data.

[0017] S3, when detecting that the vehicle window is covered with frost and snow, estimating the expected vehicle use time of the user and predicting the vehicle use probability of the user at the expected vehicle use time.

[0018] The expected vehicle use time refers to the time point at which the user is most likely to use the vehicle in the future based on the user's historical vehicle start behavior, usually represented by a specific time (such as "07:30").

[0019] In addition, the vehicle use probability refers to the degree of possibility that the user actually performs the vehicle operation at the expected vehicle use time point, quantitatively expressed in numerical form (usually a real number or percentage between 0 and 1), reflecting the influence of current environmental conditions, schedule, etc. on the user's travel intention.

[0020] S4, when the probability of use is greater than the triggering threshold, obtaining a defrosting strategy based on the environmental perception data, the state data, and the expected use time.

[0021] S5, based on the defrosting strategy, controlling the vehicle to perform an operation of removing frost and snow from the vehicle window.

[0022] In some embodiments, the method of detecting whether the vehicle window is covered by frost or snow comprises: S21, extracting surface covering features from the image data; Specifically, the vehicle window area (especially the front windshield) is located from the image collected by the high-definition camera, and the visual features of the surface covering in the area are extracted. The surface covering features include but are not limited to color distribution, texture characteristics, reflection properties, light transmission, and edge continuity, etc., which are used to represent the physical form and optical performance of the covering, and provide a basis for distinguishing frost, snow or other interference (such as water stains, dust) in the subsequent.

[0023] S22, when the surface covering features exhibit non-uniform distribution, high reflectivity, and discrete morphology, it is determined that the vehicle window is covered by snow; S23, when the surface covering features exhibit partial light transmission, mirror reflection characteristics, and continuous thin layer with blurred edge transition, it is determined that the vehicle window is covered by frost.

[0024] In the above embodiments, the present application extracts surface covering features based on image data of the vehicle window area, and distinguishes frost and snow according to their distribution morphology, optical characteristics, and structural continuity, significantly improving the accuracy and scene adaptability of frost and snow identification, and avoiding the problem of mismatching defrosting mode caused by misjudging frost as snow or vice versa. Therefore, the system can call the most suitable removal strategy (such as mechanical scraping for snow accumulation and thermal melting for frost) for different types of coverings, ensuring the defrosting effect while avoiding energy waste and execution risks caused by invalid or inefficient operations, thereby improving the overall defrosting efficiency, energy efficiency level, and system intelligence level.

[0025] In some embodiments, the method of detecting whether the vehicle window is covered by frost or snow further comprises: S24, after determining that the vehicle window is covered by snow, if the ambient temperature is not higher than the snow accumulation temperature threshold, it is confirmed that the vehicle window is in a snow accumulation state.

[0026] The snow accumulation temperature threshold is a critical temperature for distinguishing whether dry snow or wet snow can exist stably on the surface of the vehicle window, and is usually set in the range of 0°C to 2°C. The threshold can be calibrated or dynamically adjusted according to the climate characteristics of the geographical area where the vehicle is located, historical weather data, or user preferences, to ensure that the visual recognition result is confirmed as an effective snow accumulation state only under reasonable temperature conditions, avoiding misjudgment caused by temporary snow melting or sleet.

[0027] S25, after determining that the vehicle window is covered by frost, if the ambient temperature is lower than the frost temperature threshold and the ambient humidity is higher than the frost humidity threshold, it is determined that the vehicle window is in a frost state.

[0028] wherein the frost temperature threshold is a typical upper limit of temperature for water vapor to form frost by condensation, which is generally set below 0°C (e.g. -2°C to 0°C); the frost humidity threshold is the minimum relative humidity condition required for frost formation, which is generally set between 70% and 90%. The two thresholds together constitute a meteorologically reasonable criterion for frost formation, and their specific values can be calibrated according to sensor accuracy, regional climate differences or actual measurement statistics, thereby effectively excluding false positive identification due to condensed water, fog or low humidity and low temperature environment, and improving the accuracy of frost determination.

[0029] Finally, combined with the determination results of steps S24 and S25, the system can output a confirmation conclusion of whether the vehicle window is covered by frost or snow, and clearly specify the specific type of the covering; that is, it is determined to be in a snow state, a frost state, or no effective covering. The result is the final output of the method for detecting whether the vehicle window is covered by frost or snow, and is used to trigger the subsequent defrosting decision process.

[0030] In the above embodiments, the present application further combines environmental temperature and humidity to verify and confirm the determination result based on the preliminary determination of the frost and snow type based on image features, effectively improving the reliability of frost and snow identification and avoiding misjudgment due to image interference, light changes or surface stains. By setting temperature and humidity conditions that are consistent with the physical cause, such as confirming the frost state only in a low-temperature and high-humidity environment and confirming the snow state in a low-temperature condition, the system can combine visual recognition results with meteorological reasonableness to exclude misjudgments that do not conform to the actual physical law. This design significantly enhances the robustness and accuracy of frost and snow detection, providing a reliable basis for subsequent selection of an appropriate defrosting mode.

[0031] In some embodiments, the method of estimating the user's expected vehicle driving time comprises: S301, obtaining historical vehicle starting time data of the user.

[0032] wherein the historical vehicle starting time data refers to the specific time point at which the user starts the vehicle each time in the past period of time (such as the last 7 days or 30 days), which is used to reflect his daily travel habits. This data can be stored locally on the vehicle (such as a vehicle storage unit), and can also be synchronized to a cloud server, which can be flexibly called according to the system architecture.

[0033] S302, inputting the historical vehicle starting time data into a pre-trained vehicle starting time prediction model.

[0034] The vehicle start time prediction model is a machine learning model based on time series analysis, such as a recurrent neural network (RNN), a long short-term memory network (LSTM, a recurrent neural network suitable for time series modeling), or a Transformer, and the input is a historical vehicle start time sequence of the user, and the output is the most likely start time in the future. The model is trained in the offline stage by a large amount of historical start time data of users, can learn the travel time regularity of individual users, and model periodic behaviors (such as implicit time patterns such as daily commuting and weekend travel), thereby outputting the most likely vehicle use time within the next 24 hours, i.e., the expected vehicle use time.

[0035] S303, taking the time output by the vehicle start time prediction model as the expected vehicle use time of the user.

[0036] In the above embodiment, the application obtains the historical vehicle start time data of the user, and automatically calculates the expected vehicle use time of the user by using the pre-trained vehicle start time prediction model, effectively overcoming the problems of cumbersome operation and insufficient adaptability caused by the traditional defrosting scheme relying on manual reservation or fixed time triggering. In the cold winter scene, the vehicle owner does not need to go out in advance to remove frost and snow, nor does he need to actively set the departure plan on the mobile phone. The system can automatically identify the high-probability travel time based on the long-term and real vehicle use behavior of the user, and intelligently predict the defrosting opportunity. This mechanism completely transfers the preparatory labor originally borne by the user to the vehicle background to automatically complete, significantly reducing the use burden and time cost in bad weather; at the same time, since the calculation of the expected vehicle use time continues to evolve dynamically with the user's life rhythm, it can accurately fit the individual travel habits, so that the defrosting service changes from passive response to active adaptation, thereby providing the user with a more intelligent, smooth and personalized winter vehicle experience while ensuring driving safety.

[0037] In some embodiments, the method of predicting the vehicle use probability of the user at the expected vehicle use time comprises: S311, obtaining date data.

[0038] The date data includes the day of the week corresponding to the current date, whether it is a statutory holiday or a special calendar event (such as a make-up day), and is used to reflect the difference in travel inclination of the user under different date types. For example, weekdays usually correspond to a higher commuting probability, while statutory holidays or make-up days often result in the cancellation or time shift of travel plans, resulting in a lower vehicle use probability.

[0039] In some embodiments, the date data can be customized by the user on the mobile phone client.

[0040] S312, based on the vehicle location information, obtaining weather forecast data corresponding to the expected vehicle use time.

[0041] Specifically, if the estimated time of vehicle use is 07:30 on the same day, then based on the vehicle's current location (such as the latitude and longitude obtained through GPS or Beidou positioning), the weather forecast data for the time period around 07:30 at that location will be queried from the cloud-based meteorological service interface.

[0042] The weather forecast data includes temperature, snowfall, wind speed, visibility, and meteorological risk level. The meteorological risk level is used to characterize the impact of specific weather phenomena (such as blizzards, freezing rain, fog, or road icing) on ​​travel safety. It is generated and provided by meteorological service providers based on multi-factor fusion analysis and serves as a quantitative indicator that comprehensively reflects the severity of severe weather.

[0043] S313. Input the expected vehicle usage time, date data, weather forecast data, and ground snow depth into the pre-trained vehicle usage probability prediction model.

[0044] The vehicle usage probability prediction model is a machine learning-based regression model used to output an estimate of the probability that a user will actually travel at the expected usage time. During the offline phase, this model is jointly trained using a large amount of historical user behavior data and environmental context features. It can comprehensively evaluate the impact of multiple factors on users' travel intentions, including date type (such as weekday or holiday), vehicle's current location, weather forecast for the corresponding time period (such as temperature, snowfall, wind speed, visibility, and meteorological risk level), and ground snow depth.

[0045] In some embodiments, the vehicle usage probability prediction model employs the eXtreme Gradient Boosting Regression Model. By modeling the aforementioned features, it outputs a continuous value between 0 and 1, serving as a probability value quantifying the likelihood of a user using the vehicle during the expected usage time. A higher probability value indicates a greater likelihood of the user actually traveling, which can be used to trigger subsequent defrosting decisions.

[0046] S314. Determine the probability value output by the vehicle usage probability prediction model as the probability of a user using the vehicle at the expected usage time.

[0047] In the above embodiments, when predicting the actual travel probability of a user, this application not only relies on historical behavior to estimate the travel time, but also comprehensively incorporates ground snow depth, date data, and corresponding weather forecast data as prediction basis. These features are input into a pre-trained vehicle usage probability prediction model to obtain a probability assessment result that is closer to the actual travel intention. This mechanism effectively avoids misjudgment of vehicle usage intention caused by ignoring environmental, weather, and schedule factors, and significantly improves the accuracy of defrosting trigger decisions. The system can intelligently identify high-probability real travel scenarios and only generate and execute defrosting strategies when it is truly necessary. While ensuring that necessary services are available in a timely manner, it prevents high-power operations from being initiated when the user has no actual travel needs, thereby greatly reducing ineffective energy consumption, improving the overall energy utilization efficiency of the vehicle, especially beneficial for maintaining the range of electric vehicles in winter parking conditions, and further enhancing the reliability and user trust of the intelligent defrosting service.

[0048] In some embodiments, the trigger threshold is calculated by multiplying the amplification factor by the meteorological risk level and adding one, then multiplying the resulting value by the base threshold. The result is the trigger threshold.

[0049] Specifically, the system employs a dynamic threshold adjustment mechanism, adaptively calculating trigger thresholds based on real-time weather risk levels to balance user experience and energy efficiency. The trigger threshold... The calculation formula is:

[0050] in Set a basic trigger threshold (for example, set it to 0.6, which means that under normal conditions, the defrosting operation will be triggered when the probability of vehicle use exceeds 60%). This is an amplification factor (e.g., a value of 0.5, used to control the degree of influence of risk level on trigger threshold). The risk level is obtained from the meteorological bureau's API and is queried in real time via the cloud.

[0051] This application obtains structured weather forecast data by calling meteorological bureau APIs (such as the interfaces opened by the China Meteorological Administration) or commercial meteorological service interfaces. The raw information returned by these interfaces includes the current and future temperature, precipitation type, precipitation intensity, wind speed, relative humidity, and meteorological disaster warning level. Precipitation type is an enumerated value, such as no precipitation, rain, snow, sleet, or freezing rain; precipitation intensity is in millimeters per hour; wind speed is in meters per second; and meteorological warning levels are represented by the four levels of blue, yellow, orange, and red in the national standard. Furthermore, based on the meteorological warning levels, this application directly converts them into standardized meteorological risk level values ​​(range [0,1]) through a preset mapping table, where a blue warning corresponds to 0.4, a yellow warning to 0.6, an orange warning to 0.9, and a red warning to 0.95, with higher values ​​indicating higher weather risk.

[0052] In other words, under severe weather conditions, the system will proactively raise the trigger threshold, thereby reducing the risk of accidentally triggering defrosting operations. For example, when the base threshold... = 0.6, amplification factor = 0.5, and the current meteorological risk level is medium ( When (= 0.6), the dynamic threshold is calculated as follows: = 0.6 (1 + 0.5 0.6) = 0.78. This means that the system will only initiate defrosting when the predicted probability of vehicle use exceeds 78%. Since high-risk weather typically significantly reduces a user's actual willingness to travel, even if their historical habits indicate a possibility of vehicle use, the system can effectively avoid unnecessary defrosting when the predicted probability of vehicle use is artificially high but the actual likelihood of travel is low. This reduces energy waste and improves the overall vehicle energy efficiency and intelligence level.

[0053] It should be noted that, at the vehicle usage probability prediction level, environmental parameters such as weather and snow depth are used as features input into the prediction model, reflecting users' actual behavioral patterns under similar conditions in historical data, and outputting objective probability values. At the trigger threshold adjustment level, the meteorological risk level is used as an independent variable to dynamically adjust the trigger boundary. Its core purpose is to reduce the risk of false triggering by raising the decision threshold under severe weather conditions with high prediction uncertainty. When the weather is severe, on the one hand, the vehicle usage probability prediction model reduces the vehicle usage probability, and on the other hand, this application automatically raises the trigger threshold standard. The two work together to reduce the overall false judgment rate.

[0054] In the above embodiments, this application dynamically generates a trigger threshold adapted to the current meteorological risk by multiplying the amplification factor by the meteorological risk level, adding one, and then multiplying it by the basic trigger threshold. This enables the defrosting decision-making mechanism to respond to weather warning information. This dynamic adjustment mechanism can increase the trigger threshold accordingly as the meteorological risk level increases, thereby effectively suppressing false defrosting triggers caused by low user travel intentions under severe weather conditions. At the same time, it maintains a low trigger threshold when meteorological conditions are stable, ensuring that necessary services are executed in a timely manner. Thus, while taking into account user experience and driving safety, it significantly improves energy utilization efficiency, enhances the rationality of system decision-making and energy efficiency robustness in complex meteorological environments, and is particularly beneficial for maintaining the range of electric vehicles when parked in winter.

[0055] In some embodiments, the defrosting strategy includes the expected execution duration, the defrosting start time, and the defrosting control instructions.

[0056] The estimated execution time refers to the anticipated time required to complete effective defrosting. Its duration is influenced by factors such as the type of covering (e.g., snow or frost), snow depth, ambient temperature, outside wind speed, target defrosting temperature, and the available power of the vehicle's heating system. The defrosting start time is calculated by working backwards from the user's expected usage time to ensure the defrosting operation is completed smoothly before the user actually uses the vehicle. The defrosting end time is typically set slightly earlier than or aligned with the expected usage time, providing the user with a buffer. Defrosting control commands include specific control signals sent to actuators such as windshield wipers, the vehicle's air conditioning system, and window defrosting wires. These signals are used to start or stop the defrosting function, adjust heating power, airflow, or air delivery mode, thereby achieving on-demand, precise, and energy-efficient defrosting operation.

[0057] In the above embodiments, this application can reverse-plan the start timing and execution rhythm of the defrosting operation based on the user's expected vehicle usage time. This strategy ensures that the defrosting operation is completed on time before the user actually uses the vehicle, avoiding energy waste caused by starting too early and preventing delays that could affect travel preparations, thus achieving synergistic optimization of timeliness and energy efficiency. This improves the automation level, execution reliability, and energy utilization efficiency of the defrosting process, providing users with a seamless, timely, and energy-saving intelligent defrosting service. In some embodiments, the expected execution time is determined based on ambient temperature, outside wind speed, target defrost temperature, and window area.

[0058] Specifically, the system calculates the preheating time required for effective defrosting using an empirical thermodynamic formula that comprehensively considers key environmental variables affecting the vehicle's heat exchange efficiency. Estimated execution time. The calculation formula is:

[0059] ΔT represents the temperature difference between the target defrost temperature and the current ambient temperature (unit: °C). The larger the temperature difference, the longer the preheating time required. The target defrost temperature is derived from historical big data statistics reported by the vehicle. Specifically, it is the in-vehicle air conditioning temperature that users usually set in winter scenarios, used to characterize the user's expectations for the comfort of the windows and cabin after defrosting.

[0060] In addition, A is the total area of ​​the windows (unit: m²), which affects the efficiency of heating / cooling the interior space and reflects the size of the heated surface; W is the wind speed outside the vehicle (unit: m / s). The higher the wind speed, the faster the heat dissipates from the vehicle body, thus prolonging the preheating process; the coefficients α, β and γ are empirical parameters calibrated through real vehicle experiments, used to quantify the contribution weight of each factor to the preheating time, and can be taken as 0.5, 0.2 and 0.1 respectively under typical configuration.

[0061] For example, in extremely cold regions, when the ambient temperature is -20℃ (at which point the interior temperature is usually close to the ambient temperature) and the target defrost temperature is set to 15℃, the temperature difference ΔT is 35℃. If the vehicle has a large window area (e.g., 4m²) and is in windy weather (wind speed 5m / s), substituting the aforementioned empirical formula and parameters, the preheating time can be calculated to be approximately 18.8 minutes. Based on this estimation, the system can reasonably plan the defrost start time before the user's expected vehicle usage time, ensuring that the windows are clear and usable before use, while avoiding energy waste caused by running the heating device too early or for too long. This effectively improves the overall vehicle energy efficiency while ensuring driving comfort.

[0062] In some embodiments, the defrosting start time is determined based on the expected vehicle usage time and the estimated execution duration.

[0063] Specifically, the system uses the user's expected vehicle usage time as a baseline, working backwards from the estimated execution time to calculate the defrosting start time. It also considers allowing for a certain margin to ensure the defrosting operation is completed smoothly before the user actually uses the vehicle. For example, if it's predicted the user will use the vehicle at 7:30 AM, and the estimated execution time under current environmental conditions is 18 minutes, the defrosting start time will be set at 7:12 AM. Further considering safety margins and the decision-making and response time required after sending a notification to the user, the system can advance the start time by 5 to 10 minutes to address fluctuations in heating efficiency, sudden low temperatures, or other environmental uncertainties. Through this mechanism, the system effectively avoids energy waste caused by premature start-up while ensuring effective window defrosting and a good user experience, achieving precise control of defrosting timing and energy efficiency optimization.

[0064] In the above embodiments, this application uses ambient temperature, outside wind speed, target defrosting temperature, and window area as key parameters to calculate the estimated execution time required for defrosting. Based on this time and the user's expected usage time, the defrosting start time is determined in reverse, ensuring that the start and end times of the defrosting operation match actual usage needs. This mechanism fully considers external heat exchange conditions (such as low temperatures intensifying frost formation and wind speed accelerating heat dissipation) and vehicle physical characteristics (such as window area affecting heating load), avoiding insufficient defrosting or excessive energy consumption caused by using fixed durations or empirical estimations. Therefore, the defrosting operation of this application can complete the defrosting task with minimal necessary energy consumption and the most compact time window while meeting visibility requirements. This ensures immediate availability when the user gets in the vehicle and significantly improves the precision of energy scheduling and overall vehicle energy efficiency, making it particularly suitable for winter applications of electric vehicles that are sensitive to energy management. On the one hand, it ensures sufficient execution time in high heat loss scenarios such as cold and windy conditions to guarantee defrosting effectiveness; on the other hand, it shortens the running time under mild conditions to reduce unnecessary energy consumption. This mechanism significantly improves the energy efficiency and user experience of the defrosting process, and enhances the intelligence and precision of the vehicle's thermal management.

[0065] In some embodiments, the defrosting control command includes at least one of a mechanical defrosting command, a hot air defrosting command, or an electric heating defrosting command.

[0066] Among them, the defrosting control command determines the most suitable combination of execution based on the snow depth on the ground and / or the ambient temperature.

[0067] Among them, the mechanical clearing command refers to the initial physical removal of snow from the surface of the windshield using the wipers; the hot air defrosting command refers to activating the vehicle's air conditioning system, with the blower blowing warm air onto the windows and raising the interior temperature to melt frost and snow and suppress fogging on the inside of the glass; the electric heating defrosting command refers to activating the heating wires or electric heating film integrated into the windows, using electrical energy to directly heat the glass surface, quickly raising its temperature, thereby achieving efficient defrosting or snow removal.

[0068] In some embodiments, when the snow depth on the ground is detected to be greater than a first depth threshold, the defrost control command is a mechanical clearing command; specifically, the windshield wipers are activated to clear the surface of the windshield for a short period of time (e.g., 1 minute) to remove loose snow and prevent it from freezing further at low temperatures.

[0069] In addition, vehicle windshield wipers generally have a built-in torque detection mechanism, which is existing technology, and the specific details and principles will not be elaborated here. Specifically, if the wipers cannot operate normally due to excessive resistance caused by ice or compacted snow, they will automatically stop and switch to hot air defrosting or electric heating defrosting to continue the defrosting task, thereby avoiding mechanical damage such as motor overload, gear damage, or connecting rod breakage.

[0070] In some embodiments, the first depth threshold is 3 cm.

[0071] In some embodiments, when the ambient temperature is detected to be less than a first temperature threshold but greater than or equal to a second temperature threshold, the defrost control command is a hot air defrost command; specifically, the air conditioning heating mode is turned on, the blower blows heated airflow to the windows, and simultaneously raises the temperature of the cabin, using heat convection and heat conduction to melt the frost or thin layer of snow adhering to the window surface, while suppressing condensation fog on the inside of the glass caused by temperature difference.

[0072] In some embodiments, the first temperature threshold is 0°C.

[0073] In some embodiments, the second temperature threshold is -5°C.

[0074] In some embodiments, when the ambient temperature is detected to be lower than a second temperature threshold, the defrosting control command includes a hot air defrosting command and an electric heating defrosting command; specifically, the air conditioning heating mode is turned on, and the blower blows warm air to the windows and raises the interior temperature; at the same time, the heating wires on the windows or the rear window electric heating film are activated to directly heat the glass body through electric heating, quickly raising the local surface temperature to cope with more severe coverage scenarios such as thick frost, ice layers or frozen snow formed under severe cold conditions.

[0075] In some embodiments, when the ambient temperature is detected to be less than a first temperature threshold but greater than or equal to a second temperature threshold, and the snow depth on the ground is greater than a first depth threshold, the defrosting control command includes a mechanical clearing command and a hot air defrosting command. Specifically, the wipers are first activated for a short period (e.g., 1 minute) of surface clearing to remove loose snow and prevent it from freezing further at low temperatures. If the wipers automatically stop due to excessive resistance, the mechanical clearing stage is skipped. Then, the air conditioning heating mode is turned on, and the blower blows warm air to the windows and raises the interior temperature to melt the remaining wet snow, frost, or water film formed after snow melting on the windows, ensuring clear visibility.

[0076] In some embodiments, when the ambient temperature is detected to be lower than a second temperature threshold and the snow depth on the ground is greater than a first depth threshold, the defrosting control command includes a mechanical clearing command, a hot air defrosting command, and an electric heating defrosting command. Specifically, the windshield wipers are first activated for a short period (e.g., 1 minute) of surface clearing to remove loose snow and prevent it from freezing further at low temperatures; if the wipers automatically stop due to excessive resistance, the mechanical clearing stage is skipped; then the air conditioning heating mode is turned on, with the blower blowing warm air onto the windows and raising the interior temperature to melt the frost and snow covering the windows; at the same time, the heating wires or electric heating film on the windows are activated to provide rapid local heating, synergistically accelerating the frost melting process and effectively coexisting with the complex conditions of extreme low temperatures and thick snow accumulation.

[0077] In the above embodiments, this application configures the defrosting control command to include at least one of mechanical clearing command, hot air defrosting command, or electric heating defrosting command, and determines the most suitable execution mode based on the snow depth and / or ambient temperature. This allows the defrosting operation to adaptively match the optimal clearing method, such as electric heating, hot air, mechanical clearing, or a combination thereof, according to actual environmental conditions, avoiding resource waste or low defrosting efficiency caused by a one-size-fits-all control. This mechanism effectively overcomes the problems of low energy efficiency or incomplete clearing caused by using a fixed defrosting mode, ensuring efficient defrosting with reasonable energy consumption under different frost and snow coverage conditions. Simultaneously, by linking the control command with environmental perception data, the system optimizes energy allocation while ensuring defrosting effectiveness, avoiding the abuse of high-power modes, thereby improving the overall vehicle energy efficiency management level, extending the usable range of electric vehicles in winter parking conditions, and enhancing the reliability and intelligence of the defrosting process. This mechanism significantly enhances the system's adaptability to complex winter operating conditions, effectively reducing energy consumption, extending battery range, and improving the user experience while ensuring defrosting effectiveness.

[0078] As attached Figures 4 to 10 As shown, this application also discloses a vehicle defrosting system applied to a vehicle. The defrosting system 100 includes a data acquisition module 110, which is used to acquire vehicle status data and environmental perception data when the vehicle is in a turned-off state. Specifically, the data acquisition module 110 is used to acquire status data and environmental perception data when the vehicle is in sentry mode.

[0079] In some embodiments, the data acquisition module 110 includes a high-definition camera, an environmental sensor, and an ultrasonic radar.

[0080] Specifically, high-definition cameras are used to collect image data of the window area and the surroundings of the vehicle, providing basic input for the visual recognition of the frost and snow detection module 120; environmental sensors are used to monitor external environmental parameters in real time, including ambient temperature, ambient humidity, air pressure outside the vehicle, and wind speed outside the vehicle. These parameters reflect the current meteorological conditions and play a key role in judging the possibility of frost and snow formation and assessing defrosting energy consumption; ultrasonic radar is used to calculate the snow depth on the ground by analyzing the echo signal, thereby providing a quantitative basis for the system to judge the severity of external snow accumulation.

[0081] In some embodiments, vehicle status data includes vehicle ID, vehicle location information, window area, battery state of charge (SOC), parking status, door status, and ignition switch status. The vehicle ID is a unique identifier for the vehicle and is pre-installed in the vehicle control system before leaving the factory. The vehicle location information indicates the vehicle's geographical location and is acquired and cached by an onboard positioning system (such as GPS or BeiDou) before the engine is turned off. The window area refers to the total area of ​​all vehicle windows or the area of ​​key viewing areas; this parameter is preset and stored in the control system at the time of vehicle departure. The battery state of charge (SOC), parking status, door status, and ignition switch status are collected in real time by the corresponding onboard control units. This data serves two purposes: firstly, it allows the sentry system to determine whether the vehicle is in a parked and off state; secondly, the data acquisition module 110 acquires this data via the CAN bus to provide decision-making basis for the operation of the defrosting system 100.

[0082] In some embodiments, the environmental perception data includes image data of the window area, ambient temperature, ambient humidity, outside air pressure, outside wind speed, and snow depth on the ground. The image data is collected by a high-definition camera inside the vehicle; ambient temperature, ambient humidity, and outside wind speed are acquired by external environmental sensors or other corresponding sensors; and the snow depth on the ground is measured using ultrasonic radar based on the Time-of-Flight (ToF) principle. This environmental perception data is used collectively to determine whether the windows are covered by frost or snow, providing crucial input for generating subsequent defrosting strategies.

[0083] Furthermore, the vehicle-mounted environmental sensor, through its built-in temperature-sensitive element, humidity-sensitive element, pressure-sensitive element, and wind speed-sensitive element, can continuously acquire four parameters in the external environment: ambient temperature, ambient humidity, outside air pressure, and outside wind speed. The acquired parameters are then output in real time as electrical signals for subsequent processing or recording.

[0084] Furthermore, the depth of snow accumulation on the ground is detected by an ultrasonic radar installed on the vehicle body. When the ultrasonic radar is working, it emits sound wave pulses towards the ground. The sound waves are reflected after encountering the snow surface and are received by the radar. The system calculates the distance from the radar to the snow surface by measuring the round-trip time of the signal.

[0085] Specifically, the distance from the ultrasonic radar to the ground. The calculation formula is:

[0086] Where ToF is the time required for a sound wave to travel to and from the destination (in seconds), and c is the speed of sound in the air (approximately 340 m / s at room temperature). Its value varies with the ambient temperature and can be dynamically corrected by the empirical formula c = 331 + 0.6T (where T is the temperature in Celsius). The denominator "2" in the formula is used to convert the round-trip path into a one-way distance.

[0087] Based on this, the ground snow depth The calculation formula is:

[0088] Ground_height is the radar installation height, which refers to the fixed distance from the radar to the ground in snowless conditions. This parameter is determined by calibration before the vehicle leaves the factory and is built into the vehicle's control system as a constant. In addition, the radar emits signals in a direction perpendicular to the ground to measure distance.

[0089] To improve measurement stability, the data acquisition module 110 continuously samples at a frequency of 5 times per second and averages the obtained data, effectively suppressing fluctuations caused by uneven snow surface, wind disturbance, or sensor noise, thereby obtaining a more reliable snow thickness estimate and providing accurate input for the subsequent generation of defrosting strategies.

[0090] In some embodiments, for energy-saving purposes, the ultrasonic radar is kept off by default during vehicle sentry mode operation; the system only activates the ultrasonic radar when the camera detects frost or snow covering the windows to detect the snow thickness on the ground in real time, thereby avoiding unnecessary continuous power consumption, while ensuring that critical environmental information is obtained when needed to support defrosting strategy decisions.

[0091] In some embodiments, the defrosting system 100 includes a vehicle networking module (TBOX), which acts as an in-vehicle wireless communication terminal and is responsible for uploading multi-source data collected by the vehicle to the cloud platform in real time.

[0092] Specifically, the TBOX, acting as an in-vehicle communication terminal, uploads vehicle status data and environmental perception data collected by the data acquisition module 110 to the cloud at a fixed frequency (e.g., once every 10 seconds). This data serves two purposes: firstly, it supports subsequent decision-making by the defrosting system 100 in the cloud or on the vehicle, including trigger judgment, strategy generation, and execution scheduling; secondly, it serves as training samples to continuously optimize the AI ​​models deployed in the cloud (such as vehicle usage probability prediction models), enabling online iteration and adaptive improvement of system capabilities. Through this mechanism, the vehicle can not only efficiently perform defrosting operations locally but also, relying on swarm intelligence, continuously improve the accuracy of predicting user travel behavior and changes in the external environment, as well as its control response capabilities.

[0093] In some embodiments, the defrosting system 100 includes a data preprocessing module. After data is uploaded to the cloud via the vehicle networking module, it first enters the data preprocessing module for cleaning and verification. Specifically, the system will remove abnormal observations that exceed reasonable physical limits, such as ambient temperatures below a certain level. Records of temperatures of 50°C or above 60°C, or snow depths of less than 0cm or more than 200cm. These thresholds are set based on global historical extreme climate data and can effectively identify and filter dirty data introduced by sensor malfunctions, signal interference, or communication anomalies, thereby preventing them from contaminating subsequent AI model training and ensuring the robustness and accuracy of model predictions.

[0094] In some embodiments, the data preprocessing module, expected vehicle usage time estimation module 130, vehicle usage probability prediction module 140, strategy generation module 150, and interactive notification module 160 of the defrosting system 100 are deployed in the cloud.

[0095] In some embodiments, the defrosting system 100 includes a feature engineering module, which transforms raw sensing data into structured and computable feature representations to extract key features for defrosting decisions from the raw sensing data, providing data support for the expected vehicle usage time estimation module 130, the vehicle usage probability prediction module 140, and the strategy generation module 150. The core function of the feature engineering module is to extract predictive environmental dynamic trends (such as temperature trends, snow accumulation rates, etc.) and user behavior patterns (such as historical defrosting preferences, etc.) from noisy raw observation data. Among them, the temperature trend is used to characterize the ambient temperature, and the snow accumulation rate is used to estimate the ground snow depth at the expected vehicle usage time. The ground snow depth and user behavior patterns are the input features of the vehicle usage probability prediction model, providing highly discriminative and reliable contextual information for its inference.

[0096] Specifically, the feature engineering module first calculates the temperature trend to reflect the persistence of ambient temperature changes, rather than relying on easily disturbed instantaneous readings. Specifically, the feature engineering module takes historical temperature sequences and their corresponding timestamps as input, constructs a time-indexed Pandas DataFrame (a tabular data structure in Python used for structured data processing), and applies a 30-minute rolling window to calculate the local mean. The mean of the most recent window is taken as the temperature trend to characterize the ambient temperature within that window. This approach effectively suppresses noise from short-term sensor fluctuations or meteorological disturbances, allowing subsequent models to focus more on the continuous cooling process, which is the key basis for determining whether frost is imminent.

[0097] Secondly, the feature engineering module estimates the snow accumulation rate. On one hand, this reflects the rapid accumulation trend of snow on the ground, thus indirectly assessing the snow thickness on the vehicle windows. On the other hand, this feature also provides a basis for the vehicle usage probability prediction module 140 to determine whether a user is likely to use the vehicle in the near future. Specifically, the feature engineering module calculates the rate using a discrete differential method, with the formula rate = Δh / Δt, where the height change Δh comes from the continuous ranging results of the ultrasonic radar, and the time interval Δt is the time difference between two valid observations. To improve robustness, the feature engineering module sets the minimum time interval Δt to 60 seconds to filter out noise introduced by high-frequency sampling; at the same time, zero-value protection is performed before performing the division operation to prevent division by zero errors caused by abnormal timestamps. After unit conversion (seconds → hours, divided by 3600), the obtained rate forms a physically meaningful "cm / hour" index, which is directly used to predict future snow depth, thereby supporting the cloud to trigger defrosting preparation actions in advance, such as preheating the battery or pushing notifications to users.

[0098] Furthermore, the feature engineering module quantifies user behavior patterns, i.e., users' defrost preferences, to support personalized decision-making by the defrost system 100. This module converts the "defrost on" boolean label (defrost_on) in historical records into 0 or 1 values ​​and aggregates them by vehicle ID to calculate the defrost activation rate in similar scenarios such as low temperature and snow. For example, if a user actively activates defrost 85 times in the past 100 eligible scenarios, their behavior score is 0.85. This score serves as a key personalization factor, primarily providing data support for the strategy generation module 150, enabling it to distinguish between "habitual defrost users" and "users who defrost only when necessary," and dynamically adjust the aggressiveness of the strategy accordingly: initiating defrost earlier for high-scoring users and prioritizing energy saving for low-scoring users.

[0099] In addition, in some embodiments, the score is also input as an auxiliary feature to the vehicle usage probability prediction module 140 to reflect the user's sensitivity to vehicle status or travel preparation habits, thereby improving the accuracy of predicting the user's recent vehicle usage intentions.

[0100] In some embodiments, the defrosting system 100 includes a frost and snow detection module 120 for detecting whether the vehicle windows are covered by frost and snow based on environmental perception data. The frost and snow detection module 120 mainly relies on visual analysis of image data collected by an onboard high-definition camera.

[0101] Specifically, when the vehicle is off, a high-definition camera continuously collects image data (i.e., image frames) of the window area, especially the windshield. The collected images are compressed in H.265 (High Efficiency Video Coding) format and then transmitted to the frost and snow detection module 120. The frost and snow detection module 120 uses the YOLOv8 object detection model (You Only Look Once version 8, a deep learning model widely used in real-time object detection) combined with semantic segmentation technology (a pixel-level image understanding technology) to perform multimodal visual information fusion analysis. Through staged feature extraction and structured information fusion, it achieves high-precision identification of the window coverage status and generates key input parameters for subsequent defrosting decisions.

[0102] During the reasoning process, the frost and snow detection module 120 determines the type of covering on the car window by extracting optical and structural features from the image: when a large white area is identified and accompanied by granular or non-uniform texture features, it is determined to be snow; when a continuous thin layer structure with partial light transmission, specular reflection and blurred edges is detected, it is determined to be frost.

[0103] Specifically, in the first stage, this application uses YOLOv8 for target-level detection. The YOLOv8 deep learning model is used to perform a global scan on the input H.265 compressed image frame to identify and locate possible snow or frost areas. The output is a structured detection result.

[0104] In the second stage, this application uses semantic segmentation technology for pixel-level semantic analysis. Based on the candidate regions provided by YOLOv8, it further invokes the semantic segmentation network to classify the image pixel by pixel. This stage not only identifies the type of covering but also accurately classifies its spatial distribution, supporting refined modeling of complex scenes such as "partial transparency" and "partial occlusion". For example, large white areas with granular or non-uniform texture features are identified as "snow"; continuous thin-layer structures with partial translucency, specular reflection, and blurred edges are identified as "frost".

[0105] In the third stage, this application performs information fusion and structured output, spatially aligning and fusing the object-level detection results of YOLOv8 with the pixel-level classification results of semantic segmentation to generate three types of structured output information: (1) Coverage type label (e.g., "snow" means snow, "frost" means frost) is used to specify the current coverage status of the car window; (2) Confidence, which is the degree to which the module believes the judgment (e.g., 0.92), is used to evaluate the reliability of the identification when making subsequent decisions; (3) Location information, represented by bounding boxes (bbox for short, in the format of [x, y, w, h]), shows the coordinates and size of the frost-covered area in the image, which can be used to locate the window area to be processed.

[0106] The above output results together constitute the basis for the frost and snow detection module 120 to determine the state of the vehicle window. Based on the covering type label, confidence level and location information, the frost and snow detection module 120 determines whether the vehicle window is effectively covered by frost and snow: for example, when the confidence level is higher than the preset confidence level threshold (this threshold can be calibrated based on historical statistical data or actual false alarm rate to balance sensitivity and reliability), it is considered to be effectively covered.

[0107] If the frost and snow detection module 120 outputs a determination result of effective coverage, the defrosting system 100 will trigger subsequent processing procedures accordingly, including calculating the user's expected vehicle usage time, predicting the probability of vehicle usage at that time, generating a corresponding defrosting strategy, and controlling the actuator to complete the window clearing operation, thereby realizing closed-loop control from visual perception to intelligent decision-making to automatic execution.

[0108] In some embodiments, the frost and snow detection module 120 is deployed on an in-vehicle edge computing node, in which case the image data is not uploaded to the cloud to save network resources. The in-vehicle edge computing node refers to an embedded computing unit (such as an in-vehicle domain controller or dedicated AI inference chip based on ARM or NPU architecture) integrated inside the vehicle, capable of real-time analysis and decision-making on sensor data without relying on a cloud server.

[0109] The frost and snow detection module 120 is built on YOLOv8 and semantic segmentation, and converted into a lightweight model suitable for embedded platforms using TensorFlow Lite (a lightweight machine learning inference framework), performing local inference on the vehicle's edge computing node. Through this localized architecture, the system can efficiently perform frost and snow recognition tasks without relying on an external network. Even when the vehicle is off, in environments with weak or no network, it can still detect frost and snow coverage on the windows with low latency and high stability, ensuring timely triggering and reliable operation of the defrosting function. This design significantly improves the practicality, response speed, and system robustness of the entire solution in complex winter scenarios.

[0110] In some embodiments, the frost and snow detection module 120 is deployed on a cloud platform. The vehicle uploads image data collected by the camera to the cloud platform via an in-vehicle communication terminal (T-Box), where the frost and snow detection module 120 analyzes the images to determine whether the windows are covered by frost and snow.

[0111] In some embodiments, before outputting the final judgment result, the frost and snow detection module 120 also performs a secondary verification of the model's preliminary identification result in combination with the ambient temperature and humidity, so as to eliminate misjudgments caused by changes in light, water stains, dust or other non-frost and snow factors, thereby effectively avoiding accidental triggering of defrosting operations.

[0112] In some embodiments, the defrosting system 100 includes an expected vehicle usage time calculation module 130, which calculates the expected vehicle usage time in the future when it detects that the vehicle window is covered by frost and snow, so as to support advance planning of defrosting operations.

[0113] In some embodiments, the expected vehicle usage time estimation module 130 includes a vehicle start time prediction model, which is implemented based on a Long Short-Term Memory (LSTM) network to model and predict user vehicle usage behavior. LSTM models have significant advantages in capturing periodicity and trends in time series (such as historical vehicle usage time, weekdays, and weather changes). They are particularly suitable for handling periodic patterns such as "departing at 7:00–8:00 AM on weekdays and returning home at 5:00–6:00 PM."

[0114] In some embodiments, the input to the vehicle start-up time prediction model may include, in addition to historical vehicle start-up time data, coarse-grained meteorological information for the next few hours to 24 hours obtained from weather forecasts, including temperature ranges (e.g., -5°C to 0°C) and precipitation probabilities. This type of meteorological information is mainly used to reflect the general weather trend in the future period, helping to determine the user's tendency to adjust their car usage plans due to weather changes. By integrating this meteorological information, the model can dynamically adjust its inference of expected car usage time, making it closer to the user's actual travel behavior under real-world environmental conditions.

[0115] The input to the vehicle start-up time prediction model is a feature vector at a time step t. This includes the user's actual vehicle start times from several past instances, as well as the temperature range and precipitation probability for a future period obtained from weather forecasts. The model output is the predicted next start time. This refers to the expected vehicle usage time. Furthermore, to optimize model parameters, the system uses Mean Squared Error (MSE) as the loss function, defined as:

[0116] in, This represents the total loss value, used to measure the overall predictive performance of the model; The predicted start time of the model at time step t. Where is the corresponding actual startup time, and T is the total number of time steps for the training samples.

[0117] It should be noted that the vehicle start-up time prediction model is trained using supervised learning. Specifically, it first obtains the user's vehicle usage records and corresponding environmental data from the cloud over several consecutive days. The usage records include the actual start time of the vehicle each day (e.g., 07:30), and the environmental data includes the temperature range and precipitation probability for that day. When constructing training samples, a continuous historical period (e.g., the last 3 to 7 days) is selected as the input window. The daily start-up times within this window, along with the temperature range and precipitation probability of the target day (the day to be predicted), together form the model's input feature sequence. The model's prediction target is the user's actual start-up time on that target day. During training, the model continuously adjusts its internal parameters to make its predicted start-up time as close as possible to the actual start-up time on the target day. The training uses mean squared error as the optimization objective, iteratively updating the model weights through backpropagation and standard optimization algorithms (such as Adam) until the prediction error converges to an acceptable range.

[0118] The vehicle start-up time prediction model uses the user's start times over several past days, along with the temperature range and precipitation probability for the target day, as input. This is because a user's actual car usage time is influenced by both their own travel habits and weather changes. Historical start times reflect stable, periodic behavior (such as consistently leaving home on weekday mornings), while weather information for the target day indicates potential deviations (e.g., a low temperature range or a high probability of rain might delay the user's departure). By considering both factors simultaneously, the model can more accurately predict the user's actual start time on the target day.

[0119] In some embodiments, the defrosting system 100 includes a vehicle usage probability prediction module 140 for predicting the probability of a user using the vehicle during the expected vehicle usage time based on the expected vehicle usage time.

[0120] In some embodiments, the vehicle usage probability prediction module 140 includes a vehicle usage probability prediction model, which is implemented based on an XGBoost regression model and is used to quantitatively predict the probability that a user will actually use the vehicle at the expected usage time. The XGBoost regression model is better at handling nonlinear, high-dimensional feature combinations (e.g., "low temperature + strong wind + Monday morning" may correspond to a higher commuting probability), and only needs to evaluate the probability for a single candidate time point output by the LSTM model, avoiding traversal scanning of the entire day and significantly improving computational efficiency.

[0121] Specifically, when the system detects that the car windows are covered with frost and snow, it first uses a vehicle start-up time prediction model (LSTM model) to calculate the most likely time when the user will use the car in the future (e.g., 07:30) based on the user's historical vehicle start-up time. Then, combining the current environmental conditions (such as weather, temperature, snow depth, etc.) and user behavior characteristics at that time, it calls a vehicle usage probability prediction model (XGBoost model) to calculate the probability of using the car at that time. If this probability is greater than a preset threshold, the system confirms that the time is a valid expected usage time and triggers the subsequent defrosting strategy generation process.

[0122] By combining the advantages of LSTM models in temporal behavior modeling with the ability of XGBoost models in complex feature discrimination, this application achieves collaborative prediction of expected vehicle usage time and its confidence level, which significantly improves prediction accuracy and system resource utilization efficiency while ensuring timely response.

[0123] In some embodiments, the input to the vehicle usage probability prediction model is a feature vector x, which includes contextual features such as expected usage time (hour of day), day of week (day of week), temperature obtained from the weather forecast based on the expected usage time, ground snow depth at the expected usage time, last drive duration (last drive duration), and user behavior pattern (user type). The ground snow depth at the expected usage time is calculated based on the snow accumulation rate. First, the feature engineering module calculates the current snow accumulation rate, and then, combined with the predicted expected usage time, determines the time interval from the current moment to that expected usage time. Subsequently, the current ground snow thickness is added to the product of the snow accumulation rate and the time interval to obtain the estimated ground snow depth at the expected usage time. Furthermore, the user behavior pattern reflects the user's defrosting preferences and is quantitatively represented by a behavior score calculated by the feature engineering module.

[0124] The XGBoost model first calculates a real-valued output through its internal decision tree ensemble mechanism. Then through the Sigmoid function Mapping this value to the interval (0, 1) yields the final probability of vehicle use:

[0125] Here, P(use) represents the probability that a car owner will actually use the vehicle under the environmental conditions and user behavior patterns corresponding to the expected usage time. This probability is based on a comprehensive judgment of future weather forecast data (such as temperature), currently available environmental data (such as current snow depth and calculated snow growth rate), and historical user behavior characteristics. For example, if the expected usage time is a weekday morning (e.g., 7:30 AM), the weather forecast shows a temperature of -5°C with light snow at the expected usage time, and the calculated snow depth is 2 cm, and the behavior score is 0.85, the model will output a high probability of usage. Conversely, if the expected usage time is late at night on a weekend (e.g., 11:00 PM), or the weather forecast shows a sudden drop in temperature at the expected usage time, even if there is a regular usage habit, the model will output a low probability of usage.

[0126] The vehicle usage probability is transmitted to the strategy generation module 150 to determine whether it is necessary to perform a defrosting operation; only when the vehicle usage probability exceeds the trigger threshold will the energy-intensive active defrosting measures be triggered, thereby ensuring user experience while avoiding unnecessary energy waste.

[0127] It's important to note that the XGBoost regression model is trained using supervised learning. During training, the system retrieves a large number of completed defrosting task records from a cloud-based historical database as training samples. Each sample contains a complete set of contextual features, including the expected usage time, day of the week, temperature obtained from the weather forecast based on the expected usage time, calculated snow depth at the expected usage time, last driving duration, and user behavior rating. Simultaneously, each sample is associated with a real-world travel tag, determined by whether the user actually starts the vehicle within a preset time window before or after the expected usage time. This window can be configured to 3 or 5 minutes depending on actual needs. If the vehicle is started within this window, it is marked as 1; otherwise, it is marked as 0. By learning from a large number of such samples, the model outputs a value between 0 and 1, representing the probability of the user using the vehicle under given conditions.

[0128] The model adjusts its internal parameters by optimizing the difference between the predicted probability and the true label. It employs a gradient boosting mechanism to progressively build multiple decision trees and introduces regularization constraints during training to control model complexity and prevent overfitting. Furthermore, the system uses an early stopping strategy, monitoring model performance on an independent validation set and terminating training once performance no longer improves, ensuring good generalization ability. After sufficient training, the model can reliably predict the probability of a user actually using a vehicle at a specific time by integrating environmental conditions, temporal context, and the user's historical behavior.

[0129] It should be noted that the reason why the car usage probability prediction model can effectively output the probability of a user actually using a car at a certain time based on features such as expected car usage time, day of the week, weather forecast information for the corresponding time period, calculated ground snow depth at the expected car usage time, last driving duration, and user behavior score is that the above features together constitute a key context set that affects the user's travel decision.

[0130] Among these metrics, expected usage time reflects the most likely travel window formed by users based on historical habits. The day of the week and whether it is a public holiday together reflect the regularity of users' behavior constrained by social schedules; for example, weekdays usually have higher commuting rigidity, while travel plans on holidays or adjusted workdays are more prone to change. Weather forecast information and the estimated snow depth at the expected usage time point characterize the disturbance effect of the external environment on travel intentions: low temperatures or snow cover will inhibit users' actual usage behavior. The duration of the last trip refers to the time from start to finish of the last trip, which can indirectly reflect the user's recent travel status; shorter trips (e.g., only a few minutes) usually correspond to occasional use such as temporary parking or short-distance pickup, while longer trips (e.g., commuting, shopping, or cross-regional travel) often indicate that the user is in an active travel cycle, with a relatively higher probability of using the car again on the same day or the next day. User behavior scores quantify an individual's defrosting preference and travel persistence in winter driving scenarios; high-scoring users tend to use the car as planned even under adverse weather conditions. By learning the complex nonlinear relationship between the aforementioned multidimensional factors and actual travel outcomes from a large number of historical samples, the model can comprehensively assess the likelihood of a user actually starting the vehicle under specific spatiotemporal and environmental conditions, thereby providing a high-confidence decision basis for triggering subsequent defrosting strategies.

[0131] In some embodiments, the defrosting system 100 includes a strategy generation module 150, used to generate a defrosting strategy based on environmental perception data, status data, and expected vehicle usage time when the vehicle usage probability exceeds a trigger threshold. The defrosting strategy includes a defrosting start time, an expected execution duration, and specific defrosting control instructions; the defrosting control instructions include at least one of mechanical defrosting instructions, hot air defrosting instructions, and electric heating defrosting instructions, and include corresponding power parameters.

[0132] In some embodiments, the policy generation module 150 includes a thermodynamic calculation subunit, a rule engine subunit, and a reinforcement learning subunit.

[0133] Furthermore, the thermodynamic calculation subunit is responsible for determining the defrosting start time and the expected execution duration.

[0134] In some embodiments, the lead time is dynamically determined based on the user's behavior score: the behavior score is divided into multiple preset intervals, each interval corresponding to a specific lead time, and the higher the behavior score, the larger the lead time, so as to start the defrosting operation earlier and increase the interior temperature; conversely, it reduces redundant waiting time and avoids energy waste. The mapping relationship between each preset interval and its corresponding lead time can be configured according to the actual usage scenario or regional climate conditions.

[0135] In some embodiments, the rules engine subunit is responsible for generating specific defrosting control command types. It has a built-in set of decision logic based on environmental thresholds, which can determine the most appropriate execution method based on real-time sensing data. For example, when the ground snow depth exceeds a preset threshold (e.g., 3 cm), a mechanical clearing command is activated; when the ambient temperature is below -5°C and the snow accumulation is shallow, a combination of electric heating and hot air defrosting is selected; when the ambient temperature is greater than or equal to -5°C and less than 0°C, and the ground snow depth is below the preset threshold, only hot air defrosting is activated to reduce energy consumption.

[0136] In some embodiments, the reinforcement learning subunit is used for the initial setting and continuous optimization of power parameters, and includes a power decision model. The power decision model is implemented using deep reinforcement learning technology, specifically based on algorithms such as DQN (Deep Q-Network, a core algorithm combining deep learning and reinforcement learning) or PPO (Proximal Policy Optimization, an important algorithm in the field of reinforcement learning). The power decision model can dynamically recommend the optimal power parameters based on the current environmental perception data, the vehicle battery state of charge (SOC), and the type of defrosting control command output by the rule engine.

[0137] In some embodiments, the reinforcement learning subunit is used to determine power parameters based on the vehicle's current battery state of charge (SOC). Specifically, when the battery is sufficiently charged (e.g., SOC greater than 70%), a higher power parameter is set, meaning that the air conditioner, blower, or window heating wires operate at high power to speed up defrosting. When the battery is insufficient (e.g., SOC less than 30%), the power parameter is reduced, meaning that the air conditioner, blower, or window heating wires operate at the lowest power to ensure the electrical energy reserves required for vehicle startup and driving.

[0138] In some embodiments, the reinforcement learning subunit simultaneously outputs the corresponding expected clearing speed while determining the defrosting power parameters; the expected clearing speed characterizes the area of ​​frost and snow that the system expects to clear per unit time. During the execution of the defrosting task, the reinforcement learning subunit evaluates key indicators such as the actual cleared area, power usage, and execution time at regular intervals (e.g., 30 seconds, 60 seconds, or other durations). Specifically, the reinforcement learning subunit first multiplies the expected clearing speed by the current execution time to obtain the expected clearing area, and then compares the actual cleared area with the expected clearing area. If the actual value reaches more than 80% of the expected value, the current power configuration is considered valid; otherwise, it is determined that the power is insufficient and needs to be adjusted. Simultaneously, this subunit maintains a continuous compliance counter. When the same combination of "defrosting control command - power parameters" meets the effect requirements three times consecutively, it is marked as an "effective strategy," a unique identifier (e.g., STRAT_20251205_001) is generated and stored in the strategy library for direct use in subsequent identical or similar scenarios. The expected clearing area... If the target is not met in a single execution, the reinforcement learning subunit will trigger the power adjustment mechanism, which will appropriately increase the power parameters while keeping the defrosting control command unchanged, and send the updated configuration to the vehicle; the vehicle will then re-execute the defrosting task with the new power and enter the next feedback loop.

[0139] In some embodiments, the reinforcement learning subunit performs offline training daily based on historical execution records, updating its power decision model. Training data includes environmental awareness information for each task (e.g., temperature, snow depth), defrosting control command type, actual power parameters used, whether the strategy is marked as effective, and the corresponding execution effect and expected clearing speed; wherein, the expected clearing speed is obtained based on the ratio of actual cleared area to execution time. By continuously accumulating the "defrosting control command—power parameter—execution effect" triple as training samples, the model can learn, under specific environmental conditions, which power level can achieve a faster clearing speed and energy efficiency balance for a given defrosting control command. Based on this, the reinforcement learning subunit can automatically recommend more reasonable power configurations in subsequent decisions, continuously improving the synergy between energy utilization efficiency and system response performance while ensuring defrosting quality.

[0140] In some embodiments, the data acquisition module 110 includes an infrared transmission sensor, an infrared thermal imaging sensor (such as a miniature uncooled thermal imaging module), and a current and voltage sensing unit (such as a Hall effect current sensor and a high-precision voltage detection chip) integrated in the defrosting circuit.

[0141] Furthermore, the data acquisition module 110 continuously monitors key performance indicators during the defrosting process and samples them every 5 seconds: the infrared transmission sensor measures the infrared transmittance of the window glass to quantify its transparency, thus facilitating the data acquisition module 110 to calculate the actual area of ​​frost and snow removal; the infrared thermal imaging sensor acquires the temperature distribution on the glass surface to assess the uniformity of defrosting and the thermal response state; and the current and voltage sensing unit collects the electrical parameters in the defrosting heating circuit in real time and calculates the system power consumption. The collected multi-source sensor data is aggregated by an onboard vehicle networking module (such as TBOX) and uploaded to the cloud platform, providing timely and reliable feedback for subsequent defrosting performance evaluation, dynamic optimization of power parameters, and identification and solidification of effective strategies.

[0142] In some embodiments, the defrosting system 100 includes an execution control module 170, used to control the vehicle to perform the operation of clearing frost and snow from the windows based on a defrosting strategy. Specifically, the execution control module 170 is used to convert the defrosting strategy determined by the strategy generation module 150 into specific vehicle control commands when execution conditions are met, and then send them to the vehicle terminal. After the vehicle terminal performs security verification on the commands, it transmits them to the body control unit or thermal management electronic control unit, thereby driving actuators such as electric heating, hot air dampers, or mechanical defrosting to complete the operation of clearing frost and snow from the windows.

[0143] In some embodiments, the defrosting system 100 further includes an interactive notification module 160, which is used to push a prompt message containing the defrosting strategy to the vehicle owner terminal after generating the defrosting strategy, and to receive feedback instructions from the user. The execution control module 170 is configured to: perform corresponding operations according to feedback instructions when a feedback instruction is received; and automatically perform defrosting operations according to the defrosting strategy when no feedback instruction is received and the preset waiting time has exceeded.

[0144] Specifically, the system pushes a defrosting reminder to the user's mobile application via the network. The message includes a title "Vehicle Frost Reminder", explanatory text (such as "Frost has been detected on the windshield. It is recommended to turn on automatic defrosting"), and strategy information such as the recommended defrosting mode, estimated duration, and power level.

[0145] Users will see a pop-up message on their terminal: "Frost detected. Do you want to start defrosting immediately?", and can choose "Start immediately," "Start in 10 minutes," or "Do not start." If the user chooses "Start immediately," the execution control module 170 will immediately initiate the defrosting operation; if the user chooses "Start in 10 minutes," the system will create a scheduled task to execute automatically at the specified time; if the user chooses "Do not start," the execution control module 170 will interrupt the defrosting operation, the current strategy will be archived, and it will not be recommended again in the short term under similar conditions. To ensure operational security, only the vehicle owner's account has direct control permissions; family members' accounts must be authorized by the vehicle owner to execute relevant commands.

[0146] If no feedback is received from the user within the preset waiting time (e.g., 1 minute), the execution control module 170 will automatically perform the defrosting operation according to the original defrosting strategy. This default behavior is set based on the vehicle usage probability prediction result to ensure that the user's travel is not affected.

[0147] When the user confirms execution or the system automatically triggers the defrost operation after a timeout, the execution control module 170 sends a control message to the vehicle terminal containing defrost control instructions, estimated execution duration, and power parameters. This message is transmitted through a secure communication channel, employing an encryption protocol and establishing a connection based on two-way authentication to ensure that only authorized devices can receive it. A digital signature is appended to the message before transmission. Before parsing, the vehicle terminal (i.e., the vehicle networking module, TBOX) must complete three verifications: verifying the validity of the digital signature, confirming that the source of the instruction is consistent with the current vehicle binding, and checking whether the timestamp is within the allowed time window (typically ±500 milliseconds).

[0148] TBOX continuously monitors this secure channel. Upon receiving and verifying an instruction, it parses and forwards it to the vehicle control system. The vehicle control system then activates the electric heating elements, hot air dampers, or mechanical defrosting devices to perform the defrosting operation. This entire process ensures the user's right to know and operate while also prioritizing automated response capabilities and information security, achieving efficient and reliable defrosting management through human-machine collaboration.

[0149] In the above embodiments, this application adds an interactive notification module 160 to the vehicle defrosting system 100. This module proactively pushes a notification message containing strategy details to the vehicle owner's terminal after generating a defrosting strategy and receives user feedback instructions, thereby introducing a final confirmation mechanism for user intent on top of automated execution. This design retains the efficiency of the system's intelligent prediction and automatic execution while granting users the right to know and intervene in the defrosting operation, significantly improving the humanization of the service. Simultaneously, by setting a preset waiting time and defaulting to execute the strategy after the timeout, defrosting is ensured to be completed promptly even in scenarios where the user has no operation but the probability of travel is high. Furthermore, if the user explicitly cancels or postpones execution, the defrosting operation can be suspended or delayed, effectively preventing energy waste due to prediction errors. Thus, while considering user experience, operational flexibility, and energy efficiency optimization, the reliability, adaptability, and user trust of the defrosting system 100 are further enhanced.

[0150] In addition, this application also provides a vehicle that integrates the above-mentioned defrosting system 100, which can realize intelligent recognition, strategy generation and automatic removal of frost and snow on the windows based on environmental perception, vehicle usage prediction and user interaction, thereby improving the safety and convenience of winter travel.

[0151] Although embodiments of this application have been shown and described above, it is understood that the above embodiments are exemplary and should not be construed as limiting this application. Those skilled in the art can make changes, modifications, substitutions and variations to the above embodiments within the scope of this application.

Claims

1. A vehicle defrosting method, characterized in that, The defrosting method includes: When the vehicle is turned off, acquire vehicle status data and environmental perception data; Based on the environmental perception data, detect whether the car windows are covered by frost or snow; When the car window is detected to be covered by frost and snow, the user's expected car usage time is estimated, and the probability of the user using the car during the expected car usage time is predicted. When the probability of vehicle use is greater than the trigger threshold, a defrosting strategy is obtained based on the environmental perception data, the status data, and the expected vehicle use time. Based on the aforementioned defrosting strategy, the vehicle is controlled to perform the operation of clearing frost and snow from the windows.

2. The vehicle defrosting method according to claim 1, characterized in that, Methods for estimating a user's expected car usage time include: Obtain the user's historical vehicle start time data; The historical vehicle start time data is input into a pre-trained vehicle start time prediction model; The time output by the vehicle start-up time prediction model is used as the user's expected vehicle usage time.

3. The vehicle defrosting method according to claim 1, characterized in that, The status data includes vehicle location information, and the environmental perception data includes snow depth on the ground; the method for predicting the probability of a user using the vehicle during the expected usage time includes: Get date data; Based on the vehicle location information, obtain weather forecast data corresponding to the expected vehicle usage time; The expected vehicle usage time, the date data, the weather forecast data, and the ground snow depth are input into the pre-trained vehicle usage probability prediction model; The probability value output by the vehicle usage probability prediction model is determined as the probability of a user using the vehicle during the expected usage time.

4. A vehicle defrosting method according to any one of claims 1 to 3, characterized in that, The trigger threshold is calculated as follows: multiply the amplification factor by the meteorological risk level and add one, then multiply the resulting value by the base threshold. The result is the trigger threshold.

5. A vehicle defrosting method according to claim 3, characterized in that, The environmental perception data includes ambient temperature and wind speed outside the vehicle, and the status data includes the area of ​​the vehicle windows; The defrosting strategy includes an estimated execution time, which is determined based on the ambient temperature, the outside wind speed, the target defrosting temperature, and the window area.

6. A vehicle defrosting method according to claim 5, characterized in that, The defrosting strategy includes a defrosting start time, which is determined based on the expected vehicle usage time and the estimated execution duration.

7. A vehicle defrosting method according to claim 5, characterized in that, The defrosting strategy includes defrosting control instructions, which are determined based on the snow depth on the ground and / or the ambient temperature.

8. A vehicle defrosting system, characterized in that, For use in vehicles, the defrosting system (100) includes: The data acquisition module (110) is used to acquire vehicle status data and environmental perception data when the vehicle is in a turned-off state. The frost and snow detection module (120) is used to detect whether the vehicle window is covered by frost and snow based on the environmental perception data. The expected vehicle usage time calculation module (130) is used to calculate the user's expected vehicle usage time when the vehicle window is detected to be covered by frost and snow. The vehicle usage probability prediction module (140) is used to predict the probability of a user using a vehicle during the expected vehicle usage time based on the expected vehicle usage time. The strategy generation module (150) is used to generate a defrosting strategy based on the environmental perception data, the state data and the expected vehicle usage time when the vehicle usage probability is greater than the trigger threshold. The execution control module (170) is used to control the vehicle to perform the operation of clearing frost and snow from the windows based on the defrosting strategy.

9. A vehicle defrosting system according to claim 8, characterized in that, The defrosting system (100) also includes: The interactive notification module (160) is used to push a prompt message containing the defrosting strategy to the vehicle owner terminal after the defrosting strategy is generated, and to receive feedback instructions from the user. The execution control module (170) is configured to: when receiving the feedback instruction, perform a corresponding operation according to the feedback instruction; when not receiving the feedback instruction and exceeding a preset waiting time, automatically perform a defrosting operation according to the defrosting strategy.

10. A vehicle, characterized in that, The vehicle includes the defrosting system (100) as described in claim 9.

Images