Internet-based intelligent car washing system and car washing method

Abstract

The invention relates to the technical field of intelligent car washing, and discloses an intelligent car washing system and method based on the Internet, and the system comprises a client, a cloud server and an equipment end. The cloud server comprises a data analysis unit and an instruction generation unit; the data analysis unit is used for utilizing an asymmetric game scheduling method and combining a loss aversion theory to recommend a personalized car washing time period for the user, and determining the optimal proportion of the cleaning agent according to the car stain type, the environment temperature and the cleaning agent health degree; the equipment end comprises a vehicle recognition sensing unit and a cleaning execution unit. According to the invention, the whole process intelligent upgrading of the car washing service from reservation to execution is realized, the stability of the user waiting experience and the cleaning effect is obviously improved, the operation cost and the equipment loss are reduced by dynamically optimizing the resource configuration, and the transformation of the car washing service to individuation, high efficiency and low cost is integrally promoted.

Description

Technical Field

[0001] This invention relates to the field of smart car wash technology, and more specifically, to an internet-based smart car wash system and method. Background Technology

[0002] With the continuous growth of car ownership in my country, car wash services have become a high-frequency, essential service in the automotive aftermarket. Traditional car wash models rely on manual operation or basic automated equipment, and the contradiction between service efficiency, user experience, and operating costs is becoming increasingly prominent: on the one hand, users' demands for the timeliness and convenience of car washes are constantly increasing, especially during peak hours when their sensitivity to waiting time is significantly increased; on the other hand, the industry faces pain points such as rising labor costs, high consumption of cleaning agents, and complex equipment maintenance, and urgently needs to upgrade its services through technologies such as the Internet and artificial intelligence.

[0003] Currently, the car wash industry mainly relies on traditional manual service models or basic automated equipment, and faces the following technological bottlenecks: 1) Inefficient Waiting Time Management: Existing car wash systems lack precise time-slot scheduling mechanisms, often resulting in long queues or significant discrepancies between expected and actual waiting times, particularly during peak hours, which can easily lead to user dissatisfaction. Traditional systems only estimate waiting times based on static queue data, failing to incorporate user sensitivity to delays (loss aversion in this example) for differentiated management, leading to inconsistent user experiences.

[0004] 2) Detergent formulation relies on experience: Detergent concentration and additives are usually formulated in fixed ratios, without fully considering dynamic factors such as vehicle stain type (mud, oil, and mixed stains in this example), ambient temperature (low temperature in this example leads to decreased detergent activity), and detergent storage condition (decreased health in this example). This results in unstable cleaning effects or cost waste and equipment corrosion risks due to overuse of detergent.

[0005] No effective solutions have yet been proposed to address the problems in the relevant technologies. Summary of the Invention

[0006] In view of the problems in related technologies, this invention proposes an Internet-based smart car wash system and car wash method to overcome the aforementioned technical problems existing in the existing related technologies.

[0007] Therefore, the specific technical solution adopted by the present invention in this embodiment is as follows: According to one aspect of the present invention, an Internet-based smart car wash system is provided, comprising a client, a cloud server, and a device. The client application is used to book car wash services, view car wash progress in real time and manage orders, and recommend nearby services to users based on their car wash waiting time. Cloud servers include data analysis units and instruction generation units; The data analysis unit is used to recommend personalized car wash times to users using asymmetric game scheduling method combined with loss aversion theory, and to determine the optimal ratio of cleaning agent based on the type of vehicle stains, ambient temperature and cleaning agent health. The instruction generation unit integrates personalized car wash time slots and optimal detergent ratios to determine personalized car wash solutions suitable for different users and generate personalized car wash instructions. The equipment includes a vehicle identification and sensing unit and a cleaning execution unit; The vehicle recognition and sensing unit is used to visually identify the user's license plate number, vehicle type, and type of dirt on the vehicle body, and to detect the vehicle's location in real time. The cleaning execution unit is used to dynamically adjust the trajectory of the robotic arm based on the vehicle recognition results and execute the car wash service in combination with personalized car wash instructions.

[0008] Furthermore, the data analysis unit includes a personalized car wash time recommendation module and a cleaning agent optimal ratio determination module; Among them, the personalized car wash time recommendation module is used to recommend personalized car wash times to users by using asymmetric game scheduling method and loss aversion theory, combined with expected waiting time. The optimal detergent ratio determination module uses reinforcement learning technology to determine the best detergent ratio by combining the type of vehicle stains, ambient temperature, and detergent health status.

[0009] Furthermore, the personalized car wash time recommendation module, utilizing asymmetric game scheduling and loss aversion theory, combined with expected waiting time, recommends personalized car wash times to users, including: Collect users' historical reservation data, order cancellation triggering factors, waiting time feedback and vehicle attribute data. Based on clustering algorithms and combined with loss aversion coefficient, classify users into low tolerance, medium tolerance and high tolerance labels, and generate waiting aversion curves for each type of user. Based on the current number of vehicles queuing at each car wash point in the target area, the equipment's processing capacity per unit time, the historical failure probability of that period, and the standard deviation of waiting time, calculate the risk index for each period, determine the risk level, and combine it with the waiting aversion curve to determine the user waiting cost coefficient for each period. Determine user and system strategies, construct a payoff matrix based on loss aversion, and use Nash equilibrium iterative calculation to match the optimal recommendation strategy for users with different tolerance levels; based on the optimal recommendation strategy, recommend personalized car wash solutions for different users, including target time periods and expected waiting and compensation. The user strategy includes accepting the recommended time slot, sticking to the original appointment time slot, or giving up the appointment. The system strategy is to recommend low-risk periods, recommend medium-risk periods, or maintain the original periods; The payoff matrix based on loss aversion includes: User benefits = (expected waiting time saved × user time value) + referral rewards - (actual timeout time × user aversion coefficient), and user time value is determined by historical consumption levels; System revenue = (original risk reduction value × equipment operation and maintenance cost) - referral reward cost - potential complaint handling cost.

[0010] Furthermore, user historical reservation data, order cancellation triggers, waiting time feedback, and vehicle attribute data are collected. Based on clustering algorithms and loss aversion coefficients, users are categorized into low-tolerance, moderate-tolerance, and high-tolerance types, and waiting aversion curves are generated for each user type, including: Collect users' historical reservation data, order cancellation triggering factors, waiting time feedback, and vehicle attribute data within a preset time period, and perform data preprocessing and structure transformation; Based on the loss aversion theory, the user loss aversion coefficient is calculated by combining the loss value of the intensity of negative reaction and the gain value of the intensity of positive reaction. Based on the user's loss aversion coefficient, average timeout duration, and frequency of timeout occurrence, a user feature matrix is ​​constructed, and the features are standardized to eliminate differences in units. Clustering algorithms are used to cluster the standardized user feature matrix, and users are categorized into low-tolerance, medium-tolerance, and high-tolerance labels based on the feature mean of each cluster in the clustering results. Based on the total number of negative feedbacks for each user type under different timeout durations and the total number of samples under that timeout duration, the probability of negative feedback is determined. A smooth curve is obtained through curve fitting, key thresholds are marked, and a waiting aversion curve for each user type is generated. The horizontal axis of the waiting aversion curve is the timeout duration, and the vertical axis is the probability of negative feedback. The formula for calculating the user loss aversion coefficient is as follows: In the formula, The user loss aversion coefficient This is the average loss value. This represents the average return. C The cancellation rate for overdue orders. D To wait and see how much the satisfaction score will drop. T L The total timeout duration. H To improve the positive feedback rate of advance orders, R To wait for the increase in satisfaction score, T G The total duration is calculated in advance.

[0011] Furthermore, based on the mean characteristics of each cluster in the clustering results, users are categorized into low-tolerance, medium-tolerance, and high-tolerance labels, including: When the loss aversion coefficient is greater than or equal to the first loss aversion coefficient threshold, the average timeout duration is greater than or equal to the first average timeout threshold, and the frequency of timeout occurrence is greater than or equal to the first occurrence frequency threshold, the user of this cluster is marked as low tolerance. When the loss aversion coefficient is greater than or equal to the second loss aversion coefficient threshold and less than the first loss aversion coefficient threshold, the average timeout duration is greater than or equal to the second average timeout threshold and less than the first average timeout threshold, and the frequency of timeout occurrence is greater than or equal to the second occurrence frequency threshold and less than the first occurrence frequency threshold, then the user of this cluster is marked as a normal tolerance type. Users in this cluster are marked as high-tolerance if the loss aversion coefficient is less than the third loss aversion coefficient threshold, the average timeout duration is less than the third average timeout threshold, and the frequency of timeout occurrence is less than the third occurrence frequency threshold.

[0012] Furthermore, based on the current number of vehicles queuing at each car wash location within the target area, the equipment's processing capacity per unit time, the historical failure probability for that period, and the standard deviation of waiting time, the risk index for each period is calculated to determine the risk level. Combined with the waiting aversion curve, the user waiting cost coefficient for each period is determined, including: The system obtains the current number of vehicles in queue, the processing capacity of the equipment per unit time, the historical failure probability and the standard deviation of waiting time for each car wash point within the target area, and performs standardization processing to eliminate differences in dimensions. Multiple linear regression analysis was used to determine the weight of each parameter on risk, and the risk index for each time period was calculated by combining the real-time number of vehicles in queue, the processing capacity of the equipment per unit time, the historical failure probability of the time period and the standard deviation of the waiting time. Based on the risk index and combined with the preset risk threshold, the risk level of each time period is determined, and the average aversion intensity and user waiting cost coefficient under different risk levels are determined according to the waiting aversion curve of each type of user. The formula for calculating the risk index is as follows: In the formula, Ris is the risk index. V This represents the current number of vehicles in the queue. U This refers to the processing capacity of the equipment per unit time. F This represents the historical failure probability for that period. S The standard deviation of the waiting time. The weights are respectively the influence of queuing pressure, historical failure probability during this period, and standard deviation of waiting time.

[0013] Furthermore, the module for determining the optimal detergent ratio utilizes reinforcement learning technology, combining factors such as vehicle stain type, ambient temperature, and detergent health, to determine the optimal detergent ratio, including: Define a state space, and convert the vehicle stain type, ambient temperature and cleaning agent health metric into corresponding values ​​to form a state vector containing the three. Define the action space, discretize the detergent ratio into several concentration levels, determine the type of auxiliary agent to be added, and construct an action vector that includes the concentration level and the type of auxiliary agent. Construct the reward function as follows ,in, Rew For the overall reward value, Res For stain residue rate, Q This refers to the amount of cleaning agent used. Cost For the cost of auxiliary agents, Corr To mitigate the risk of equipment corrosion, α 1. α 2. α 3 represents the weights for cleaning effectiveness, cleaning cost, and equipment corrosion risk, respectively; A reinforcement learning model is constructed and trained, and the trained reinforcement learning model is used to output the optimal ratio and auxiliary agent scheme corresponding to real-time stain type, ambient temperature and detergent health data.

[0014] Furthermore, cloud servers also include communication units, user management units, and device management units; The communication unit is used to connect the client and the device respectively, obtain real-time data from the client and the device, and send personalized car wash instructions to the device. The user management unit is used to store user vehicle information, order information, and consumption records; The equipment management unit is used to monitor the operating status of equipment, identify equipment faults, and automatically generate maintenance work orders, enabling the cloud server to remotely manage the equipment.

[0015] Furthermore, the equipment also includes an equipment status monitoring unit and a security protection unit; Among them, the equipment status monitoring unit is used to monitor the operating vibration of the robotic arm, motor temperature parameters, and liquid levels of cleaning agent and clean water in real time. The safety protection unit is used to identify unauthorized intrusions into the car wash area using infrared beam technology and trigger an emergency shutdown, while simultaneously recording the car wash process in real time.

[0016] According to another aspect of the present invention, an internet-based smart car wash method is provided, the method comprising the following steps: S1. The user sends a car wash reservation request to the cloud server through the client. S2. The cloud server receives the user's car wash reservation request and uses the asymmetric game scheduling method combined with loss aversion theory to recommend personalized car wash time slots to the user. S3 and the cloud server will send the recommended personalized car wash time slots to the client. The user selects the corresponding personalized car wash time slot and completes the car wash service reservation. S4. The cloud server sends the reservation data to the device. The device receives the reservation data and, when the user drives the vehicle to the device, identifies the vehicle based on the reservation data and uploads the identification information to the cloud server. The S5 cloud server receives the recognition results and compares the vehicle information and car wash time slot with the reservation requirements. When the comparison results meet the reservation requirements, it determines the best ratio of cleaning agent based on the type of vehicle stains, ambient temperature and cleaning agent health, and generates personalized car wash instructions to send to the device. S6. The device receives personalized car wash instructions and dynamically adjusts the trajectory of the robotic arm based on the vehicle recognition results, while simultaneously executing the car wash service in conjunction with the personalized car wash instructions.

[0017] The beneficial effects of Ming are: 1) This invention achieves a fully intelligent upgrade of the car wash service from reservation to execution by integrating intelligent algorithms such as asymmetric game scheduling, loss aversion theory, and reinforcement learning through the collaborative linkage of the client, cloud server, and device. This not only significantly improves the user waiting experience and the stability of the cleaning effect, but also reduces operating costs and equipment wear and tear by dynamically optimizing resource allocation, thus promoting the overall transformation of car wash services towards personalization, efficiency, and low cost.

[0018] 2) This invention integrates asymmetric game scheduling, loss aversion theory, and reinforcement learning technology to achieve intelligent matching and dynamic optimization of user needs and car wash resources. This not only balances user waiting experience and equipment load through precise personalized time slot recommendations, but also ensures the stability and economy of cleaning effect through adaptive adjustment of detergent ratio. The system achieves synergistic optimization in dimensions such as user satisfaction, cleaning quality stability, operating costs, and equipment efficiency, providing an effective guarantee for the intelligent upgrade of smart car wash systems. Attached Figure Description

[0019] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0020] Figure 1This is a schematic diagram of the principle of an Internet-based smart car wash system according to an embodiment of the present invention. Detailed Implementation

[0021] To further illustrate the various embodiments, the present invention provides accompanying drawings, which are part of the disclosure of the present invention. These drawings are mainly used to illustrate the embodiments and can be used in conjunction with the relevant descriptions in the specification to explain the operating principles of the embodiments. With reference to these drawings, those skilled in the art should be able to understand other possible implementation methods and the advantages of the present invention. The components in the drawings are not drawn to scale, and similar component symbols are generally used to represent similar components.

[0022] According to an embodiment of the present invention, an Internet-based smart car wash system and car wash method are provided.

[0023] The present invention will now be further described in conjunction with the accompanying drawings and specific embodiments, such as... Figure 1 As shown, the Internet-based smart car wash system according to an embodiment of the present invention includes a client (i.e., for the terminal), a cloud server, and a device (i.e., car wash equipment).

[0024] The client application is used to book car wash services, view car wash progress in real time, and manage orders (order management includes historical record queries, invoice applications, complaint feedback, etc.). It also recommends nearby services to users based on their car wash waiting time.

[0025] A cloud server includes a data analysis unit, an instruction generation unit, a communication unit, a user management unit, and a device management unit.

[0026] The data analysis unit uses asymmetric game scheduling method combined with loss aversion theory to recommend personalized car wash times for users, and determines the optimal ratio of cleaning agents based on the type of vehicle stains, ambient temperature, and cleaning agent health.

[0027] Specifically, the data analysis unit includes a personalized car wash time recommendation module and a cleaning agent optimal ratio determination module.

[0028] The personalized car wash time recommendation module utilizes asymmetric game scheduling and loss aversion theory, combined with expected waiting time, to recommend personalized car wash times to users; specifically, it includes: 1) Collect users' historical reservation data, order cancellation trigger factors, waiting time feedback and vehicle attribute data. Based on the clustering algorithm and combined with the loss aversion coefficient, classify users into low tolerance, medium tolerance and high tolerance types, and generate waiting aversion curves for each type of user.

[0029] Specifically, the system collects users' historical reservation data, order cancellation triggers, waiting time feedback, and vehicle attribute data. Based on clustering algorithms and loss aversion coefficients, users are categorized into low-tolerance, moderate-tolerance, and high-tolerance types. A waiting aversion curve is generated for each user type, including: 11) Collect users' historical reservation data, order cancellation triggering factors, waiting time feedback and vehicle attribute data within a preset time period, and perform data preprocessing and structure transformation.

[0030] In this embodiment, four types of raw data from users over the past 12 months are collected: historical appointment data includes appointment time slots, accurate to 30-minute intervals, actual arrival time, time interval between appointment and arrival, and actual time spent completing the car wash. Cancellation triggers include recording users' choices when canceling via system-preset options such as excessively long waiting times, unforeseen circumstances, and price reasons. Orders with excessively long waiting times are marked as wait-sensitive cancellations. Waiting time feedback includes textual evaluations related to waiting in the user's APP rating (1-5 stars) after completing the car wash, such as "too long," "faster than expected," and a separate satisfaction rating for waiting time (1-10 points). Vehicle attribute data includes high-end / mid-range / low-end vehicle brands, sedan / SUV / commercial vehicle models, and usage frequency based on average daily trips.

[0031] Data preprocessing and structured transformation include: filling missing waiting feedback data with the average feedback of similar orders for the same user; supplementing missing vehicle attributes by associating license plate recognition with publicly available data from the vehicle management office; performing NLP sentiment analysis on text reviews to convert them into waiting satisfaction scores, ranging from -5 to +5, with negative reviews being negative and positive reviews being positive; extracting peak / off-peak labels from the appointment time slots, in this embodiment 7:00-9:00 and 17:00-19:00 are defined as peak times, and calculating the matching degree between the appointment time slot and historical high-frequency appointment time slots.

[0032] 12) Based on the loss aversion theory, calculate the user loss aversion coefficient by combining the loss value of the intensity of negative reaction and the gain value of the intensity of positive reaction; In this embodiment, the loss aversion coefficient for a single user is calculated based on the loss aversion theory: the intensity of a user's negative reaction to a 1-minute delay is set as the loss value (L), and the intensity of a positive reaction to a 1-minute early arrival is set as the gain value (G); the formula for calculating the user's loss aversion coefficient is as follows: In the formula, The user loss aversion coefficient The higher the value, the greater the user's aversion to waiting timeouts. This is the average loss value. This represents the average return. C The cancellation rate for overdue orders.D To wait and see how much the satisfaction score will drop. T L The total timeout duration. H To improve the positive feedback rate of advance orders, R To wait for the increase in satisfaction score, T G The total duration is calculated in advance.

[0033] 13) Construct a user feature matrix based on the user's loss aversion coefficient, average timeout duration, and frequency of timeout occurrence, and standardize the features to eliminate dimensional differences.

[0034] In this embodiment, the user's loss aversion coefficient, average timeout duration, and frequency of timeout durations exceeding 5 minutes are used as core features to construct a three-dimensional feature matrix. The features are then standardized, such as by Z-score standardization, to eliminate dimensional differences and ensure clustering effectiveness.

[0035] 14) Use clustering algorithms to cluster the standardized user feature matrix, and use the feature mean of each cluster in the clustering results to classify users into low-tolerance, normal-tolerance, and high-tolerance labels.

[0036] In this embodiment, firstly, the K-means clustering algorithm (preset K=3) is used to group users: using the feature matrix as input, iterative calculations are performed to maximize the feature similarity of users within each cluster and the difference between clusters. During the clustering process, the silhouette coefficient is used to verify the effectiveness of the clustering. A silhouette coefficient > 0.6 indicates good clustering; if the coefficient is too low, the feature variables are adjusted (e.g., increasing the tolerance for timeouts exceeding 10 minutes) and the clustering is repeated. Then, user labels are assigned based on the clustering results: the mean features of the three clusters are calculated, and user labels are assigned as low-tolerance, medium-tolerance, and high-tolerance based on the mean features of each cluster in the clustering results. When the loss aversion coefficient is greater than or equal to the first loss aversion coefficient threshold (3.0 in this embodiment), the average timeout duration is greater than or equal to the first average timeout threshold (10 minutes in this embodiment), and the frequency of timeout occurrence is greater than or equal to the first occurrence frequency threshold (30% in this embodiment), then the user of this cluster is marked as low tolerance. When the loss aversion coefficient is greater than or equal to the second loss aversion coefficient threshold (1.5 in this embodiment) and less than the first loss aversion coefficient threshold, the average timeout duration is greater than or equal to the second average timeout threshold (5 negative minutes in this embodiment) and less than the first average timeout threshold, and the frequency of timeout occurrence is greater than or equal to the second occurrence frequency threshold (10% in this embodiment) and less than the first occurrence frequency threshold, then the user of this cluster is marked as a normal tolerance type. When the loss aversion coefficient is less than the third loss aversion coefficient threshold (1.0 in this embodiment), the average timeout duration is less than the third average timeout threshold (3 minutes in this embodiment), and the frequency of timeout occurrence is less than the third occurrence frequency threshold (5% in this embodiment), then the user of this cluster is marked as high tolerance.

[0037] 15) Determine the probability of negative feedback based on the total number of negative feedbacks for each user under different timeout durations and the total number of samples under that timeout duration. Obtain a smooth curve through curve fitting, mark key thresholds, and generate a waiting aversion curve for each user type. The horizontal axis of the waiting aversion curve is the timeout duration, and the vertical axis is the probability of negative feedback.

[0038] In this embodiment, a waiting aversion curve is generated for each user type: the horizontal axis represents the timeout duration, 0-25 minutes, with 1-minute intervals; the vertical axis represents the probability of negative feedback, 0-100%. For each user type, the total number of negative feedback instances under different timeout durations is counted. Negative feedback includes order cancellation, complaints, and 1-2 star ratings. Dividing this by the total number of samples under that duration yields the corresponding probability value. A smooth curve is obtained through curve fitting (logistic regression in this embodiment), and key thresholds are marked. For example, the probability of negative feedback for low-tolerance users surges to 80% after 5 minutes of timeout, for average-tolerance users it reaches 60% after 10 minutes, and for high-tolerance users it reaches 50% after 15 minutes. Simultaneously, new samples are randomly selected for each user type, i.e., data not involved in clustering. The actual timeout duration is substituted into the curve, and the error rate between the predicted negative feedback probability and the actual feedback is calculated. If the average error rate is <10%, the curve is valid; otherwise, the curve is refitted after supplementing sample data until the error reaches the target.

[0039] 2) Based on the current number of vehicles queuing at each car wash point in the target area, the processing capacity of the equipment per unit time, the historical failure probability of the period and the standard deviation of the waiting time, calculate the risk index for each period, determine the risk level, and combine it with the waiting aversion curve to determine the user waiting cost coefficient for each period.

[0040] Based on the current number of vehicles queuing at each car wash location within the target area, the equipment's processing capacity per unit time, the historical failure probability for that time period, and the standard deviation of waiting time, the risk index for each time period is calculated to determine the risk level. Furthermore, the user waiting cost coefficient for each time period is determined by combining this with the waiting aversion curve, including: 21) Obtain the current number of vehicles in the queue, the processing capacity of the equipment per unit time, the historical failure probability (value 0-1) and the standard deviation of waiting time for each car wash point in the target area. The standard deviation of waiting time reflects the fluctuation of waiting time and is standardized to eliminate the difference in dimensions.

[0041] 22) Use multiple linear regression analysis to determine the weight of each parameter on risk, and combine it with the real-time number of vehicles in queue, the processing capacity of equipment per unit time, the historical failure probability of the period and the standard deviation of waiting time to calculate the risk index for each period.

[0042] In this embodiment, historical data is analyzed using multiple linear regression to determine the weight of each parameter on the risk: This reflects the contribution of queuing pressure to risk, typically ranging from 0.4 to 0.6. Queuing time is the most directly perceived factor for users. Reflects the risk of equipment failure, with a value ranging from 0.2 to 0.3. This reflects the impact of uncertainty in waiting time, with a value ranging from 0.2 to 0.3. The formula for calculating the risk index is as follows: In the formula, Ris is the risk index. V This represents the current number of vehicles in the queue. U This refers to the processing capacity of the equipment per unit time. F This represents the historical failure probability for that period. S The standard deviation of the waiting time. The weights are respectively the influence of queuing pressure, historical failure probability during this period, and standard deviation of waiting time.

[0043] 23) Based on the risk index and combined with the preset risk threshold, the risk level of each time period is determined, and the average aversion intensity and user waiting cost coefficient under different risk levels are determined according to the waiting aversion curve of each type of user.

[0044] In this embodiment, firstly, based on the historical risk index distribution and the corresponding negative user feedback rate, three threshold levels are set: high risk: Ris > 0.7, corresponding to a historical negative feedback rate > 40%; medium risk: 0.3 ≤ Ris ≤ 0.7, corresponding to a negative feedback rate of 15%-40%; low risk: Ris < 0.3, corresponding to a negative feedback rate < 15%. Then, the waiting aversion curves of low-tolerance, medium-tolerance, and high-tolerance users are extracted. The horizontal axis represents the timeout duration, and the vertical axis represents the probability of negative feedback. The average aversion intensity under different risk levels is calculated: In high-risk periods, the average timeout duration and volatility of users are both high, corresponding to the high-probability negative feedback interval in the aversion curve. The cost coefficient is set to 1.5, indicating that the psychological cost of waiting for users in this period is 1.5 times the benchmark. In medium-risk periods, the corresponding interval in the aversion curve is the medium probability interval, and the cost coefficient is set to 1.2. In low-risk periods, the corresponding interval in the aversion curve is the low probability interval, and the cost coefficient is set to 1.0. Finally, the risk level and the matching user waiting cost coefficient for each period are output and stored separately for each user type. For example, the cost coefficient for low-tolerance users can be increased by 0.2 in high-risk periods.

[0045] 3) Determine user and system strategies, construct a payoff matrix based on loss aversion, and use Nash equilibrium iterative calculation to match the optimal recommendation strategy for users with different tolerance levels; based on the optimal recommendation strategy, recommend personalized car wash solutions for different users, including target time periods and expected waiting and compensation.

[0046] The user strategy includes accepting the recommended time slot, sticking to the original appointment time slot, or giving up the appointment. The system strategy is to recommend low-risk periods, recommend medium-risk periods, or maintain the original periods; The payoff matrix based on loss aversion includes: User benefits = (expected waiting time saved × user time value) + referral rewards - (actual timeout time × user aversion coefficient), and user time value is determined by historical consumption levels; System revenue = (original risk reduction value × equipment operation and maintenance cost) - referral reward cost - potential complaint handling cost; Furthermore, by iteratively calculating the Nash equilibrium, the optimal recommendation strategy for each user type is determined as follows: for low-tolerance users, recommend off-peak hours, promising a conservative expected waiting time (i.e., the actual longest possible time) and compensation for any delays; for high-tolerance users, recommend peak-edge hours, informing them of aggressive expected waiting times and compensation for delays with points; for average-tolerance users, recommend mid-peak hours, balancing waiting time and discounts. The recommended content must include the core time period, flexible expectations, and the rationale for the decision. The client provides three options: one-click confirmation, viewing the rationale, and refusing to retain the original time period.

[0047] The expected waiting time and compensation include: for users with low tolerance, the expected waiting time is the 95% confidence interval of the time period, and the compensation is a 10 yuan coupon for exceeding the time limit; for users with average tolerance, the expected waiting time is the 75% confidence interval of the time period, and the compensation is 3 points for exceeding the time limit by 1 minute; for users with high tolerance, the expected waiting time is the 50% confidence interval of the time period, and the compensation is a car wash discount coupon for completing the task on time. In addition, this embodiment includes daily collection of recommendation performance data: recommendation acceptance rate for each user type, deviation between actual and expected waiting time (positive deviation for overtime / negative deviation for early arrival), and user satisfaction ratings for waiting in post-event evaluations. Based on this data, the user aversion coefficient is updated every 3 days (in this embodiment, a tolerant user accepts two consecutive 5-minute delays without complaint, so their aversion coefficient is lowered), the weight of the time-period risk index is optimized weekly (in this embodiment, it is found that equipment failure in a certain area has a greater impact on waiting time, so the corresponding weight is increased), and A / B testing is conducted monthly (in this embodiment, the effects of different compensation schemes are tested on the same type of user), iterating the recommendation script and compensation methods. The ultimate goal is to increase the recommendation acceptance rate to over 60% and maintain user waiting satisfaction ratings above 4.5 / 5.

[0048] The optimal detergent ratio determination module utilizes reinforcement learning technology, combined with factors such as vehicle stain type, ambient temperature, and detergent health, to determine the optimal detergent ratio; specifically, it includes: 1) Define a state space, and convert the vehicle stain type, ambient temperature and cleaning agent health into corresponding values ​​to form a state vector containing the three.

[0049] In this embodiment, a state space is defined, and vehicle stain types are quantified into values ​​ranging from 0 to 1 according to their cleaning difficulty. In this embodiment, mud and sand are 0.2, oil stains are 0.5, bird droppings are 0.8, tar spots are 0.9, and mixed stains are 1.0. The higher the value, the higher the cleaning difficulty. The ambient temperature is standardized into a range of 0 to 1. In this embodiment, -10℃ is 0.1, 25℃ is 0.5, and 40℃ is 0.9, reflecting the effect of temperature on the activity of the cleaning agent. The health of the cleaning agent is quantified into a health index ranging from 0 to 1 through detection. In this embodiment, newly opened is 1.0, stored for 30 days and oxidized is 0.3, and insufficient effective ingredients are 0.1, forming a state vector containing the three factors.

[0050] 2) Define the action space, discretize the detergent ratio into several concentration levels, determine the type of auxiliary agent to be added, and construct an action vector that includes the concentration level and the type of auxiliary agent.

[0051] In this embodiment, an action space is defined, and the detergent ratio is discretized into 10 concentration levels, with 1:20 to 1:110 corresponding to 0.1 to 1.0. The addition of auxiliary agents is set to 0 (no addition), 1 (chelating agent), and 2 (stabilizing agent), forming an action vector that includes concentration level and auxiliary agent type.

[0052] 3) Construct the reward function as follows ,in, Rew For the overall reward value, Res For stain residue rate, Q This refers to the amount of cleaning agent used. Cost For the cost of auxiliary agents, Corr To mitigate the risk of equipment corrosion, α 1. α 2. α 3 represents the weights of cleaning effect, cleaning cost, and equipment corrosion risk, respectively. In this embodiment, α 1. α 2. α 3. Take values ​​of 0.6, 0.3, and 0.1 respectively.

[0053] 4) Construct and train a reinforcement learning model, and use the trained reinforcement learning model to output the optimal ratio and auxiliary agent scheme corresponding to real-time stain type, ambient temperature and detergent health data.

[0054] In this embodiment, firstly, a simulation environment is constructed using historical cleaning data to simulate the cleaning effects corresponding to actions under different states. Later, real equipment is connected to collect real-time status data. After an action is executed, the actual stain residue rate, consumable consumption, and equipment status are obtained as environmental feedback. The state transition process, residual stain type, and detergent health decay after cleaning are recorded. Then, a deep Q-network model is trained. The network structure is designed with an input layer of 3 neurons (corresponding to the state vector), 2 hidden layers (64+32 neurons), and an output layer of 30 neurons (10 concentration levels × 3 types of auxiliary agents). Training is performed by randomly sampling samples through experience replay. The target network is updated every 1000 steps. Initially, an ε-greedy strategy (ε=0.9) is used for exploration, gradually decreasing to 0.1 until the average reward on the validation set converges. Finally, the trained model is deployed to a cloud server, receiving real-time data on stain type, ambient temperature, and detergent health from the device. The action with the highest Q-value is output as the optimal ratio and auxiliary agent scheme. After each cleaning, the actual feedback data is sent back to the experience pool. The model is fine-tuned daily with new data to adapt to new scenarios.

[0055] The instruction generation unit is used to integrate personalized car wash time slots and the optimal ratio of cleaning agents, determine personalized car wash solutions suitable for different users, and generate personalized car wash instructions.

[0056] The communication unit is used to connect the client and the device respectively, obtain real-time data from the client and the device, and send personalized car wash instructions to the device.

[0057] The user management unit is used to store user vehicle information, order information, and consumption records.

[0058] The equipment management unit is used to monitor the operating status of equipment, identify equipment faults, and automatically generate maintenance work orders, enabling the cloud server to remotely manage the equipment.

[0059] The equipment includes a vehicle identification and sensing unit, a cleaning execution unit, an equipment status monitoring unit, and a safety protection unit.

[0060] The vehicle recognition and sensing unit is used to visually identify the user's license plate number, vehicle type, and type of dirt on the vehicle body, and to detect the vehicle's location in real time.

[0061] The cleaning execution unit is used to dynamically adjust the trajectory of the robotic arm based on the vehicle recognition results and execute the car wash service in combination with personalized car wash instructions.

[0062] The equipment status monitoring unit is used to monitor the operating vibration of the robotic arm, motor temperature parameters, and the liquid levels of cleaning agent and clean water in real time.

[0063] The safety protection unit is used to identify unauthorized intrusions into the car wash area using infrared beam technology and trigger an emergency shutdown, while simultaneously recording the car wash process in real time.

[0064] According to another embodiment of the present invention, an Internet-based smart car wash method includes the following steps: S1. The user sends a car wash reservation request to the cloud server through the client. S2. The cloud server receives the user's car wash reservation request and uses the asymmetric game scheduling method combined with loss aversion theory to recommend personalized car wash time slots to the user. S3 and the cloud server will send the recommended personalized car wash time slots to the client. The user selects the corresponding personalized car wash time slot and completes the car wash service reservation. S4. The cloud server sends the reservation data to the device. The device receives the reservation data and, when the user drives the vehicle to the device, identifies the vehicle based on the reservation data and uploads the identification information to the cloud server. The S5 cloud server receives the recognition results and compares the vehicle information and car wash time slot with the reservation requirements. When the comparison results meet the reservation requirements, it determines the best ratio of cleaning agent based on the type of vehicle stains, ambient temperature and cleaning agent health, and generates personalized car wash instructions to send to the device. S6. The device receives personalized car wash instructions and dynamically adjusts the trajectory of the robotic arm based on the vehicle recognition results, while simultaneously executing the car wash service in conjunction with the personalized car wash instructions.

[0065] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. An internet-based smart car wash system, characterized in that, This includes the client, cloud server, and device. The client application is used to book car wash services, view car wash progress in real time and manage orders, and recommend nearby services to users based on their car wash waiting time. The cloud server includes a data analysis unit and an instruction generation unit; The data analysis unit is used to recommend personalized car wash times to users using asymmetric game scheduling method combined with loss aversion theory, and to determine the optimal ratio of cleaning agent based on vehicle stain type, ambient temperature and cleaning agent health. The instruction generation unit is used to integrate personalized car wash time slots and optimal detergent ratios to determine personalized car wash plans suitable for different users and generate personalized car wash instructions; The device includes a vehicle identification and sensing unit and a cleaning execution unit; The vehicle recognition and sensing unit is used to visually identify the user's license plate number, vehicle type, and type of dirt on the vehicle body, and to detect the vehicle's location in real time. The cleaning execution unit is used to dynamically adjust the trajectory of the robotic arm based on the vehicle recognition results and execute the car wash service in conjunction with personalized car wash instructions.

2. The Internet-based smart car wash system according to claim 1, characterized in that, The data analysis unit includes a personalized car wash time recommendation module and a cleaning agent optimal ratio determination module; The personalized car wash time recommendation module is used to recommend personalized car wash times to users by using asymmetric game scheduling method and loss aversion theory, combined with expected waiting time. The cleaning agent optimal ratio determination module is used to determine the optimal ratio of cleaning agent by using reinforcement learning technology, combined with vehicle stain type, ambient temperature and cleaning agent health status.

3. The Internet-based smart car wash system according to claim 2, characterized in that, The personalized car wash time recommendation module, when recommending personalized car wash times to users using asymmetric game scheduling and loss aversion theory, combined with expected waiting time, includes: Collect users' historical reservation data, order cancellation triggering factors, waiting time feedback and vehicle attribute data. Based on clustering algorithms and combined with loss aversion coefficient, classify users into low tolerance, medium tolerance and high tolerance labels, and generate waiting aversion curves for each type of user. Based on the current number of vehicles queuing at each car wash point in the target area, the equipment's processing capacity per unit time, the historical failure probability of that period, and the standard deviation of waiting time, calculate the risk index for each period, determine the risk level, and combine it with the waiting aversion curve to determine the user waiting cost coefficient for each period. Determine user and system strategies, construct a payoff matrix based on loss aversion, and use Nash equilibrium iterative calculation to match the optimal recommendation strategy for users with different tolerance levels; based on the optimal recommendation strategy, recommend personalized car wash solutions for different users, including target time periods and expected waiting and compensation. The user strategy includes accepting the recommended time slot, sticking to the original appointment time slot, or giving up the appointment. The system strategy is to recommend low-risk periods, recommend medium-risk periods, or maintain the original periods; The payoff matrix based on loss aversion includes: User benefits = (expected waiting time saved × user time value) + referral rewards - (actual timeout time × user aversion coefficient), and user time value is determined by historical consumption levels; System revenue = (original risk reduction value × equipment operation and maintenance cost) - referral reward cost - potential complaint handling cost.

4. The Internet-based smart car wash system according to claim 3, characterized in that, The collected user historical reservation data, order cancellation triggering factors, waiting time feedback, and vehicle attribute data are used to classify users into low-tolerance, moderate-tolerance, and high-tolerance categories based on a clustering algorithm and loss aversion coefficient. A waiting aversion curve is generated for each user category, including: Collect users' historical reservation data, order cancellation triggering factors, waiting time feedback, and vehicle attribute data within a preset time period, and perform data preprocessing and structure transformation; Based on the loss aversion theory, the user loss aversion coefficient is calculated by combining the loss value of the intensity of negative reaction and the gain value of the intensity of positive reaction. Based on the user's loss aversion coefficient, average timeout duration, and frequency of timeout occurrence, a user feature matrix is ​​constructed, and the features are standardized to eliminate differences in units. Clustering algorithms are used to cluster the standardized user feature matrix, and users are categorized into low-tolerance, medium-tolerance, and high-tolerance labels based on the feature mean of each cluster in the clustering results. Based on the total number of negative feedbacks for each user type under different timeout durations and the total number of samples under that timeout duration, the probability of negative feedback is determined. A smooth curve is obtained through curve fitting, key thresholds are marked, and a waiting aversion curve for each user type is generated. The horizontal axis of the waiting aversion curve is the timeout duration, and the vertical axis is the probability of negative feedback. The formula for calculating the user loss aversion coefficient is as follows: In the formula, The user loss aversion coefficient This is the average loss value. This represents the average return. C The cancellation rate for overdue orders. D To wait and see how much the satisfaction score will drop. T L The total timeout duration. H To improve the positive feedback rate of advance orders, R To wait for the increase in satisfaction score, T G The total duration is calculated in advance.

5. The Internet-based smart car wash system according to claim 4, characterized in that, Based on the mean characteristics of each cluster in the clustering results, users are categorized into low-tolerance, moderate-tolerance, and high-tolerance labels, including: When the loss aversion coefficient is greater than or equal to the first loss aversion coefficient threshold, the average timeout duration is greater than or equal to the first average timeout threshold, and the frequency of timeout occurrence is greater than or equal to the first occurrence frequency threshold, the user of this cluster is marked as low tolerance. When the loss aversion coefficient is greater than or equal to the second loss aversion coefficient threshold and less than the first loss aversion coefficient threshold, the average timeout duration is greater than or equal to the second average timeout threshold and less than the first average timeout threshold, and the frequency of timeout occurrence is greater than or equal to the second occurrence frequency threshold and less than the first occurrence frequency threshold, then the user of this cluster is marked as a normal tolerance type. Users in this cluster are marked as high-tolerance if the loss aversion coefficient is less than the third loss aversion coefficient threshold, the average timeout duration is less than the third average timeout threshold, and the frequency of timeout occurrence is less than the third occurrence frequency threshold.

6. The Internet-based smart car wash system according to claim 2, characterized in that, The process involves calculating the risk index for each time period based on the current number of vehicles queuing at each car wash location within the target area, the equipment's processing capacity per unit time, the historical failure probability for that time period, and the standard deviation of waiting time. This determines the risk level, and, combined with the waiting aversion curve, determines the user waiting cost coefficient for each time period, including: The system obtains the current number of vehicles in queue, the processing capacity of the equipment per unit time, the historical failure probability and the standard deviation of waiting time for each car wash point within the target area, and performs standardization processing to eliminate differences in dimensions. Multiple linear regression analysis was used to determine the weight of each parameter on risk, and the risk index for each time period was calculated by combining the real-time number of vehicles in queue, the processing capacity of the equipment per unit time, the historical failure probability of the time period and the standard deviation of the waiting time. Based on the risk index and combined with the preset risk threshold, the risk level of each time period is determined, and the average aversion intensity and user waiting cost coefficient under different risk levels are determined according to the waiting aversion curve of each type of user. The formula for calculating the risk index is as follows: In the formula, Ris is the risk index. V This represents the current number of vehicles in the queue. U This refers to the processing capacity of the equipment per unit time. F This represents the historical failure probability for that period. S The standard deviation of the waiting time. The weights are respectively the influence of queuing pressure, historical failure probability during this period, and standard deviation of waiting time.

7. The Internet-based smart car wash system according to claim 2, characterized in that, The optimal detergent ratio determination module, when using reinforcement learning technology and combining vehicle stain type, ambient temperature, and detergent health status to determine the optimal detergent ratio, includes: Define a state space, and convert the vehicle stain type, ambient temperature and cleaning agent health metric into corresponding values ​​to form a state vector containing the three. Define the action space, discretize the detergent ratio into several concentration levels, determine the type of auxiliary agent to be added, and construct an action vector that includes the concentration level and the type of auxiliary agent. Construct the reward function as follows ,in, Rew For the overall reward value, Res For stain residue rate, Q This refers to the amount of cleaning agent used. Cost For the cost of auxiliary agents, Corr To mitigate the risk of equipment corrosion, α 1. α 2. α 3 represents the weights for cleaning effectiveness, cleaning cost, and equipment corrosion risk, respectively; A reinforcement learning model is constructed and trained, and the trained reinforcement learning model is used to output the optimal ratio and auxiliary agent scheme corresponding to real-time stain type, ambient temperature and detergent health data.

8. The Internet-based smart car wash system according to claim 1, characterized in that, The cloud server also includes a communication unit, a user management unit, and a device management unit; The communication unit is used to connect to the client and the device respectively, obtain real-time data from the client and the device, and send personalized car wash instructions to the device. The user management unit is used to store user vehicle information, order information, and consumption records; The equipment management unit is used to monitor the operating status of the equipment, identify equipment faults, and automatically generate maintenance work orders, enabling the cloud server to remotely manage the equipment.

9. The Internet-based smart car wash system according to claim 1, characterized in that, The device also includes a device status monitoring unit and a security protection unit; The equipment status monitoring unit is used to monitor the operating vibration of the robotic arm, motor temperature parameters, and liquid levels of cleaning agent and clean water in real time. The security protection unit is used to identify unauthorized intrusions into the car wash area using infrared beam technology and trigger an emergency shutdown, while simultaneously recording the car wash process in real time.

10. A smart car wash method based on the Internet, implemented using the smart car wash system based on the Internet as described in any one of claims 1-9, characterized in that, The method includes the following steps: S1. The user sends a car wash reservation request to the cloud server through the client. S2. The cloud server receives the user's car wash reservation request and uses the asymmetric game scheduling method combined with loss aversion theory to recommend personalized car wash time slots to the user. S3 and the cloud server will send the recommended personalized car wash time slots to the client. The user selects the corresponding personalized car wash time slot and completes the car wash service reservation. S4. The cloud server sends the reservation data to the device. The device receives the reservation data and, when the user drives the vehicle to the device, identifies the vehicle based on the reservation data and uploads the identification information to the cloud server. The S5 cloud server receives the recognition results and compares the vehicle information and car wash time slot with the reservation requirements. When the comparison results meet the reservation requirements, it determines the best ratio of cleaning agent based on the type of vehicle stains, ambient temperature and cleaning agent health, and generates personalized car wash instructions to send to the device. S6. The device receives personalized car wash instructions and dynamically adjusts the trajectory of the robotic arm based on the vehicle recognition results, while simultaneously executing the car wash service in conjunction with the personalized car wash instructions.

Images