Key account vehicle maintenance monitoring system

Abstract

The application belongs to the field of customer management, and discloses a key customer vehicle maintenance monitoring system, which comprises a prediction module, the prediction module comprises a weight updating unit, a prediction unit and a probability calculation unit; the weight updating unit is used for calculating the weighted weight of each model used for prediction respectively, comprising: acquiring the parking probability of each model used for prediction in the last N times of prediction respectively, N being an adaptive parameter; calculating the weighted weight based on the parking probability of the last N times of prediction and the set probability threshold; the prediction unit is used for predicting the parking probability of the key customer using each model used for prediction respectively, and obtaining the parking probability; the probability calculation unit is used for weighting all the parking probabilities based on the weighted weight, and obtaining the weighted parking probability. The application can further improve the accuracy of the predicted result.

Description

Technical Field

[0001] This invention relates to the field of customer management, and more particularly to a vehicle maintenance monitoring system for key customers. Background Technology

[0002] To provide a better parking experience for key customers, existing technologies determine whether to reserve parking spaces in advance by predicting the probability of these customers parking within a specified future time period. To improve the stability of the prediction results, existing methods typically consider the parking probabilities predicted by multiple models and weight them to arrive at a final predicted parking probability. However, the weights in this weighting process are usually the same, which leads to models with lower accuracy in the short term receiving excessive weights, resulting in insufficient accuracy in the final prediction. Summary of the Invention

[0003] The purpose of this invention is to disclose a key customer vehicle maintenance monitoring system to solve the technical problems pointed out in the background art.

[0004] To achieve the above objectives, the present invention adopts the following technical solution:

[0005] This invention provides a key customer vehicle maintenance monitoring system, including a prediction module, which includes a weight update unit, a prediction unit, and a probability calculation unit.

[0006] The weight update unit is used to calculate the weighted weights for each model used for prediction, including:

[0007] Obtain the parking probability of each model used for prediction in the most recent N predictions, where N is the adaptive parameter;

[0008] The weighted average is calculated based on the parking probabilities predicted in the most recent N predictions and the set probability threshold.

[0009] The prediction unit is used to predict the parking probability of key customers using each prediction model to obtain the parking probability.

[0010] The probability calculation unit is used to weight all parking probabilities based on weighted weights to obtain the weighted parking probability.

[0011] Preferably, the process for determining the adaptive parameters is as follows:

[0012] Get the standard value M of the set adaptive parameter;

[0013] After each prediction, the adaptive parameter N to be used in the next prediction is calculated based on the weighted parking probability obtained from the most recent M predictions and the set probability threshold.

[0014] Preferably, the adaptive parameter N used in the next prediction process is calculated based on the weighted parking probabilities obtained from the most recent M predictions and the set probability threshold, including:

[0015] Get the number of weighted parking probabilities (num1) that are greater than the set probability threshold from the weighted parking probabilities obtained in the most recent M predictions;

[0016] Get the latest The number of weighted parking probabilities greater than the set probability threshold in the weighted parking probabilities obtained from the prediction, num2; Indicates rounding up;

[0017] Calculate the adaptive parameter N based on num1 and num2.

[0018] Preferably, the weighted weights are calculated based on the parking probabilities predicted in the most recent N predictions and a set probability threshold, including:

[0019] Store all models used for prediction into set U;

[0020] For each model in set U, obtain the number of parking probabilities that are greater than the set probability threshold in the most recent N predictions of parking probabilities;

[0021] Based on the set probability threshold, the parking probability of each model in U in the most recent N predictions is converted into a result sequence;

[0022] Calculate the first weight of each model in U based on the result sequence;

[0023] The second weight of each model in U is calculated based on the number of parking probabilities that are greater than the set probability threshold.

[0024] The weighted weights are calculated based on the first and second weights.

[0025] Preferably, the parking probabilities predicted by each model in U in the most recent N predictions are converted into a result sequence, including:

[0026] For model q in U, store the parking probabilities predicted by model q in the most recent N predictions in a sequence according to the prediction time from earliest to latest. ;

[0027] Sequences Each parking probability in the sequence is processed as follows to obtain the result sequence:

[0028] for The probability of the i-th parking space ,like If the probability exceeds the set probability threshold, then... Change the value to 1, otherwise... The value was changed to 0.

[0029] Preferably, it also includes a model training module;

[0030] The model training module is used to train each model in set V to obtain a model for prediction;

[0031] V is a set of untrained models capable of predicting parking probabilities.

[0032] Preferably, the models in set V include gradient boosting tree models, Naive Bayes models, LSTM neural network models, and Bayesian structure time series models.

[0033] Preferably, it also includes a training data collection module;

[0034] The training data collection module is used to acquire data for training the model.

[0035] Preferably, the data used to train the model includes:

[0036] Customer attribute data, time characteristic data, and external environment data.

[0037] Preferably, it also includes a prompting module;

[0038] The prompt module is used to provide prompts to parking lot managers based on the weighted parking probability.

[0039] Preferably, it also includes a license plate monitoring module;

[0040] The license plate monitoring module is used to obtain the license plate numbers of vehicles entering the parking lot and to determine whether the vehicle owner is a key customer based on the license plate number.

[0041] Beneficial effects:

[0042] In predicting the parking probability of key customers using multiple models, this invention no longer uses equal weighting to weight the prediction results of various models. Instead, it calculates the weighting based on the parking probabilities obtained from the most recent N predictions and a set probability threshold. This allows the weighting to change as the prediction results change. By comparing the prediction results of different models, the model with more accurate recent predictions has a greater impact on the final weighted parking probability, thereby further improving the accuracy of the prediction results. Attached Figure Description

[0043] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the embodiments will be briefly introduced below. It should be understood that the following drawings only show some embodiments of the present invention and should not be regarded as a limitation on the scope. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort.

[0044] Figure 1 This is a first schematic diagram of the key customer vehicle maintenance monitoring system of the present invention.

[0045] Figure 2 This is a schematic diagram illustrating the process of calculating the weighted weights.

[0046] Figure 3 This is a second schematic diagram of the key customer vehicle maintenance monitoring system of the present invention.

[0047] Figure 4 This is a third schematic diagram of the key customer vehicle maintenance monitoring system of the present invention.

[0048] Figure 5 This is the fourth schematic diagram of the key customer vehicle maintenance monitoring system of the present invention. Detailed Implementation

[0049] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. The components of the embodiments of the present invention described and shown in the accompanying drawings can generally be arranged and designed in various different configurations. Therefore, the following detailed description of the embodiments of the present invention provided in the accompanying drawings is not intended to limit the scope of the claimed invention, but merely to illustrate selected embodiments of the invention. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without inventive effort are within the scope of protection of the present invention.

[0050] like Figure 1 As shown in one embodiment, the present invention provides a key customer vehicle maintenance monitoring system, including a prediction module, which includes a weight update unit, a prediction unit, and a probability calculation unit.

[0051] The weight update unit is used to calculate the weighted weights for each model used for prediction, including:

[0052] Obtain the parking probability of each model used for prediction in the most recent N predictions, where N is the adaptive parameter;

[0053] The weighted average is calculated based on the parking probabilities predicted in the most recent N predictions and the set probability threshold.

[0054] The prediction unit is used to predict the parking probability of key customers using each prediction model to obtain the parking probability.

[0055] The probability calculation unit is used to weight all parking probabilities based on weighted weights to obtain the weighted parking probability.

[0056] In this invention, the predicted parking probabilities of the model used for prediction are uniformly normalized to the interval [0,1].

[0057] The probability threshold set in this invention can be 0.7. Furthermore, the probability threshold can be adjusted based on the prediction results and parking results. For example, if it is found that a lower probability threshold has a higher success rate in determining whether a key customer has parked in the previous time period (e.g., one month), the probability threshold can be lowered accordingly in the next time period.

[0058] This invention typically predicts the parking probability for each time period separately. For example, 9:00 AM to 12:00 PM is considered one time period, and 2:00 PM to 6:00 PM is another. Therefore, during the training of the prediction model, the same model needs to be trained separately for each time period to improve the model's specificity and the accuracy of the prediction results.

[0059] Preferably, the process for determining the adaptive parameters is as follows:

[0060] Get the standard value M of the set adaptive parameter;

[0061] After each prediction, the adaptive parameter N to be used in the next prediction is calculated based on the weighted parking probability obtained from the most recent M predictions and the set probability threshold.

[0062] The standard value M in this invention can be 20. That is, the adaptive parameter N is calculated by obtaining the prediction results of the most recent 20 predictions.

[0063] In the initial running phase, if the cumulative number of predictions is less than M, then the value of N is directly set to M.

[0064] Preferably, the adaptive parameter N used in the next prediction process is calculated based on the weighted parking probabilities obtained from the most recent M predictions and the set probability threshold, including:

[0065] Get the number of weighted parking probabilities (num1) that are greater than the set probability threshold from the weighted parking probabilities obtained in the most recent M predictions;

[0066] Get the latest The number of weighted parking probabilities greater than the set probability threshold in the weighted parking probabilities obtained from the prediction, num2; Indicates rounding up;

[0067] The adaptive parameter N is calculated based on num1 and num2, including:

[0068] First step, calculate the quantity num3:

[0069] num3 = num1 - num2;

[0070] The second step is to calculate the value of N:

[0071] .

[0072] In calculating the adaptive parameter N, this invention divides the weighted parking probabilities in the most recent M predictions into two intervals that are greater than a set probability threshold. +1 to M is one interval, while 1 to This is another interval. The calculation is performed by comparing the changes in the weighted parking probabilities between these two intervals. Therefore, the value of N in this invention can adaptively change with the changes in the weighted parking probability, thus showing an increasing trend in prediction accuracy. The higher the accuracy, the smaller the value of N, thereby improving the calculation efficiency of the weighted weights; conversely, the larger the value of N is, allowing the weighted weights to more accurately represent the accuracy of the model's predictions by using more historical data.

[0073] Preferably, such as Figure 2 The weighted average is calculated based on the parking probabilities predicted in the most recent N predictions and the set probability threshold, including:

[0074] The first step is to store all the models used for prediction into set U.

[0075] The second step is to obtain the number of parking probabilities that are greater than the set probability threshold for each model in set U in the most recent N predictions of parking probabilities.

[0076] The third step, based on the set probability threshold, is to convert the parking probabilities predicted by each model in U in the most recent N predictions into a result sequence, including:

[0077] For model q in U, store the parking probabilities predicted by model q in the most recent N predictions in a sequence according to the prediction time from earliest to latest. ;

[0078] Sequences Each parking probability in the sequence is processed as follows to obtain the result sequence:

[0079] for The probability of the i-th parking space ,like If the probability exceeds the set probability threshold, then... Change the value to 1, otherwise... The value was changed to 0.

[0080] For example, if the parking probabilities predicted in the most recent 5 predictions are 0.5, 0.8, 0.9, 0.6, and 0.3, then the sequence... =[0.5, 0.8, 0.9, 0.6, 0.3]. When the probability threshold is set to 0.7, the result sequence is [0, 1, 1, 0, 0].

[0081] By transforming the sequence of parking probabilities into a result sequence, compared to directly using parking probabilities to calculate the first weight, the differences in prediction results between different models are amplified. This allows for a more effective representation of the differences in prediction results between various models in subsequent calculations. Therefore, under the same prediction success rate, the weighting of models with stronger prediction independence can be effectively increased, thus improving the effectiveness of the final weighted parking probability. When the prediction success rates of different models meet the usage requirements (e.g., prediction success rate greater than 0.8), the lower the correlation between the parking probability sequence of one model and the corresponding parking probability sequences of other models, the stronger the independence of the prediction results.

[0082] The fourth step is to calculate the first weight of each model in U based on the result sequence, including:

[0083] S1, For model q in U, calculate the difference coefficient between the prediction results of q and each model in U other than model q based on the result sequence:

[0084] ;

[0085] The coefficient of difference between the prediction results of model q and model r; For the judgment value, if The k-th element and If the k-th element is different, then It is 1 if it is true, otherwise it is 0. Let be the time interval between the generation time of the parking probability corresponding to the k-th element and the current time. This is the weight control factor for the k-th element;

[0086] The acquisition process is as follows:

[0087] The standard deviation of the parking probability corresponding to the kD-th to k-th elements in the result sequence is calculated; D is a pre-set integer, for example, D can be 5.

[0088] When k is less than D, the result of kD is set to k, which avoids kD being negative due to insufficient number of elements in the prior period.

[0089] The standard deviation is standardized and mapped to the interval [0,1] to obtain the standardized standard deviation. ;

[0090] Standard deviation calculation based on standardization :

[0091] ;

[0092] and These are the maximum and minimum values ​​of the weight control factors corresponding to the first k elements in the result sequence, respectively.

[0093] S2, calculate the first weight of model q using the following formula:

[0094] ;

[0095] Uq represents the set of all models in U except for model q; This is the first weight of model q.

[0096] The first weight in this invention is calculated based on the difference coefficient. Therefore, the larger the sum of the difference coefficients between model q and other models, the larger the first weight of model q, indicating a stronger independence between the prediction results of model q and the prediction results of other models. Furthermore, in calculating the difference coefficient, this invention also sets corresponding influence coefficients for each judgment value according to the position of the elements in the model in the result sequence. Therefore, if the generation time of the parking probability corresponding to the element in the result sequence is further away from the current time, the greater the fluctuation of the parking probability corresponding to the most recent D elements, the smaller the influence of the judgment value corresponding to the element on the first weight. This can improve the response sensitivity of the weighted weight of the present invention to changes in the prediction accuracy of the model.

[0097] When the recent parking probability exhibits high volatility (high variance), it may indicate unstable predictions or the presence of noise. In this case, the rate of change of the influence coefficient should be slowed down, allowing more stable, older data to have relatively higher weights to smooth out the noise. Conversely, when the data exhibits low volatility (stable), the rate of change of the influence coefficient can be accelerated, emphasizing recent trends. This allows for a more timely assessment of the impact of models with improved prediction accuracy on the final weighted result, thereby enhancing the accuracy of the final prediction.

[0098] The parking probability corresponding to each element in the resulting sequence can be determined based on the element's position from the sequence. Obtained from [the source].

[0099] For example, for model q, This is the sequence obtained by sorting the parking probabilities predicted by model q in the most recent N predictions according to the prediction time from earliest to latest.

[0100] The parking probability corresponding to the k-th element in the result sequence is then... The k-th element in.

[0101] Fifth, calculate the second weight of each model in U based on the number of parking probabilities greater than the set probability threshold, including:

[0102] For model q, the number of parking probabilities predicted by model q in the most recent N predictions that are greater than the set probability threshold is represented as: ;

[0103] The second weight of model q is calculated using the following formula:

[0104] ;

[0105] This is the second weight of model q.

[0106] Step 6: Calculate the weighted weights based on the first and second weights, including:

[0107] Calculate the weighted weights using the following formula:

[0108] ;

[0109] These are the weighted weights for model q.

[0110] The weighting method of this invention considers not only the independence of the model's prediction results but also the accuracy of the model's prediction results. Therefore, the higher the accuracy and the stronger the independence of the model's prediction results, the greater its weighting. This effectively improves the final prediction accuracy after integrating the prediction results of each model, and also enhances the stability of the final prediction results.

[0111] Preferably, the weighted parking probability is obtained by weighting all parking probabilities based on weighted averages, including:

[0112] ;

[0113] To weight the probability of parking, Let q be the parking probability output by model q in the latest prediction.

[0114] Preferably, such as Figure 3 The present invention also includes a model training module;

[0115] The model training module is used to train each model in set V to obtain a model for prediction;

[0116] V is a set of untrained models capable of predicting parking probabilities.

[0117] During training, the training data can be divided into different independent sets (e.g., training set and test set), and the model can be trained and the prediction accuracy of the model can be evaluated separately based on the different sets.

[0118] Preferably, the models in set V include gradient boosting tree models, Naive Bayes models, LSTM neural network models, and Bayesian structure time series models.

[0119] Preferably, such as Figure 4 The present invention also includes a training data collection module;

[0120] The training data collection module is used to acquire data for training the model.

[0121] Specifically, the training data collection module can obtain data from the parking management system's database, as well as from the network.

[0122] Preferably, the data used to train the model includes:

[0123] Customer attribute data, time characteristic data, and external environment data.

[0124] Customer attribute data includes membership level, historical parking frequency, age, etc. Time characteristic data includes the day of the week the parking occurred and the time period within that day. External environment data includes the weather type and the number of large-scale events near the parking lot when key customers park.

[0125] Preferably, such as Figure 5 The present invention also includes a prompting module;

[0126] The prompt module is used to provide prompts to parking lot managers based on the weighted parking probability.

[0127] Specifically, when the weighted parking probability is greater than the set probability threshold, the license plate number of the key customer's vehicle and the name of the key customer are sent to the equipment used by the parking lot manager, so that the parking lot manager can reserve parking spaces for the key customer in advance.

[0128] Preferably, it also includes a license plate monitoring module;

[0129] The license plate monitoring module is used to obtain the license plate numbers of vehicles entering the parking lot and to determine whether the vehicle owner is a key customer based on the license plate number.

[0130] The notification module also alerts parking lot staff when the license plate monitoring module determines the vehicle owner is a key customer based on the license plate number. This allows parking lot staff to arrange parking services for key customers in advance based on their parking habits.

[0131] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A key customer vehicle maintenance monitoring system, characterized in that, It includes a prediction module, which comprises a weight update unit, a prediction unit, and a probability calculation unit; The weight update unit is used to calculate the weighted weights for each model used for prediction, including: Obtain the parking probability of each model used for prediction in the most recent N predictions, where N is the adaptive parameter; The weighted average is calculated based on the parking probabilities predicted in the most recent N predictions and the set probability threshold. The prediction unit is used to predict the parking probability of key customers using each prediction model to obtain the parking probability. The probability calculation unit is used to weight all parking probabilities based on weighted weights to obtain the weighted parking probability; The weighted average is calculated based on the parking probabilities predicted in the most recent N predictions and the set probability threshold, including: Store all models used for prediction into set U; For each model in set U, obtain the number of parking probabilities that are greater than the set probability threshold in the most recent N predictions of parking probabilities; Based on the set probability threshold, the parking probability of each model in U in the most recent N predictions is converted into a result sequence; Calculate the first weight of each model in U based on the result sequence, including: S1, For model q in U, calculate the difference coefficient between the prediction results of q and each model in U other than model q based on the result sequence: ； The coefficient of difference between the prediction results of model q and model r; For the judgment value, if The k-th element and If the k-th element is different, then It is 1 if it is true, otherwise it is 0; Let be the time interval between the generation time of the parking probability corresponding to the k-th element and the current time. This is the weight control factor for the k-th element; S2, calculate the first weight of model q using the following formula: ； Uq represents the set of all models in U except for model q; The first weight of model q; The second weight of each model in U is calculated based on the number of parking probabilities that are greater than the set probability threshold. The weighted weights are calculated based on the first and second weights. Convert the parking probabilities predicted by each model in U in the most recent N predictions into a result sequence, including: For model q in U, store the parking probabilities predicted by model q in the most recent N predictions in sequence from earliest to latest, according to the prediction time. ; Sequences Each parking probability in the sequence is processed as follows to obtain the result sequence: for The probability of the i-th parking space ,like If the probability exceeds the set probability threshold, then... Change the value to 1, otherwise... The value was changed to 0.

2. The key customer vehicle maintenance monitoring system according to claim 1, characterized in that, The process of determining the adaptive parameters is as follows: Get the standard value M of the set adaptive parameter; After each prediction, the adaptive parameter N to be used in the next prediction is calculated based on the weighted parking probability obtained from the most recent M predictions and the set probability threshold.

3. The key customer vehicle maintenance monitoring system according to claim 2, characterized in that, The adaptive parameter N used in the next prediction is calculated based on the weighted parking probabilities obtained from the most recent M predictions and the set probability threshold, including: Get the number of weighted parking probabilities (num1) that are greater than the set probability threshold from the weighted parking probabilities obtained in the most recent M predictions; Get the latest The number of weighted parking probabilities greater than the set probability threshold in the weighted parking probabilities obtained from the prediction, num2; Indicates rounding up; Calculate the adaptive parameter N based on num1 and num2.

4. The key customer vehicle maintenance monitoring system according to claim 1, characterized in that, It also includes a model training module; The model training module is used to train each model in set V to obtain a model for prediction; V represents the set of untrained models capable of predicting parking probabilities.

5. The key customer vehicle maintenance monitoring system according to claim 4, characterized in that, The models in set V include gradient boosting tree models, Naive Bayes models, LSTM neural network models, and Bayesian structure time series models.

6. The key customer vehicle maintenance monitoring system according to claim 1, characterized in that, It also includes a training data collection module; The training data collection module is used to acquire data for training the model.

7. The key customer vehicle maintenance monitoring system according to claim 6, characterized in that, The data used to train the model includes: Customer attribute data, time characteristic data, and external environment data.

8. The key customer vehicle maintenance monitoring system according to claim 1, characterized in that, It also includes a prompt module; The prompt module is used to provide prompts to parking lot managers based on the weighted parking probability.

Images