A dynamic parking allocation method based on parking occupancy rate prediction in a cloud control system

Abstract

The present invention relates to a dynamic parking allocation method based on parking occupancy prediction under a cloud control system, and belongs to the field of smart parking technology. The method proposes a basic framework of an allocation method based on parking occupancy prediction, establishes a SARIMA prediction model for parking occupancy according to the characteristics of the parking lot occupancy time series, and then dynamically allocates the model according to the dynamic characteristics of the parking problem. For the established dynamic allocation model, a hybrid heuristic algorithm of an improved variable neighborhood search algorithm is designed to solve the model, an initial solution is generated by a greedy algorithm, and then an improved variable neighborhood search algorithm is used to search for the optimal solution, and a simulation experiment is set up in combination with a real data set. Based on the research on parking occupancy by the cloud control platform, the present invention establishes a SARIMA prediction model for parking occupancy to predict it, and uses the cloud control platform through a dynamic parking allocation method to provide parking users with an efficient, accurate and low-cost parking solution.

Description

Technical Field

[0001] The present invention belongs to the field of smart parking technology, and specifically relates to a dynamic parking allocation method based on parking occupancy rate prediction in a cloud control system. Background Art

[0002] The intelligent connected vehicle cloud control system, also known as the vehicle-road-cloud integrated control system, consists of a cloud control platform, roadside infrastructure, communication networks, intelligent connected vehicles and related support platforms; real-time parking data can be provided through the cloud control platform.

[0003] Currently, the parking lot occupancy prediction problem is generally regarded as a prediction of a certain time series. Time series prediction methods have been used to improve the accuracy of parking lot occupancy. However, in actual scenarios, since input information such as waiting vehicle information and parking lot information changes over time, the accuracy of predicting parking lot occupancy using the above prediction methods is low.

[0004] The present invention selects a dynamic parking allocation model and establishes a dynamic parking allocation scheme based on time series prediction, thereby solving the problems faced in actual parking applications. Summary of the Invention

[0005] In light of this, the present invention aims to provide a dynamic parking allocation method based on parking occupancy rate prediction in a cloud-based control system. This method addresses the low accuracy of parking lot occupancy rate predictions in existing technologies. Based on research on parking occupancy rates conducted by a cloud-based control platform, this invention establishes a SARIMA prediction model for parking occupancy rates. Through this dynamic parking allocation method, the cloud-based control platform is leveraged to provide users with efficient, accurate, and low-cost parking solutions.

[0006] like Figure 1-6 As shown, in order to achieve the above-mentioned purpose, the present invention provides the following technical solutions:

[0007] A dynamic parking allocation method based on parking occupancy rate prediction in a cloud control system includes the following steps:

[0008] S1: Propose the basic framework of the parking occupancy rate allocation method under the architecture of the cloud control system;

[0009] S2: within the basic framework, according to the characteristics of the parking lot occupancy rate time series, the parking lot occupancy rate is predicted based on the SARIMA model to obtain a prediction result of the parking lot occupancy rate;

[0010] S3: Based on the prediction results of S2, a time series dynamic parking allocation model is constructed for the dynamic characteristics of the parking problem;

[0011] S4: For the constructed time series dynamic parking allocation model, a hybrid heuristic algorithm of an improved variable neighborhood search algorithm is designed to solve the time series dynamic parking allocation model.

[0012] Furthermore, the basic framework in S1 includes a cloud control center, an intelligent connected vehicle, an intelligent parking lot, and a parking allocation method; the cloud control center monitors the intelligent parking lot in real time through a communication network, and the intelligent connected vehicle initiates a parking request to the cloud control center through the communication network. The cloud control center processes the parking information of the parking request, starts the parking allocation method, and allocates a parking plan for the intelligent connected vehicle. After the intelligent connected vehicle completes parking according to the allocated parking plan, the intelligent parking lot updates the parking lot status to the cloud control center. After the intelligent parking lot status update is completed, the intelligent connected vehicle updates the vehicle status to the cloud control center.

[0013] Furthermore, the S2 specifically includes constructing a SARIMA model, preprocessing data, determining the model order, optimizing the model, testing the model adaptability, and predicting the model;

[0014] The SARIMA model is developed based on the autoregressive model AR and the moving average model MA;

[0015] The basic idea of ​​the autoregressive model is that the time series {X t The value of the t-th period of} is only related to the value of the previous period, so its lagged value and the linear function of the current residual are expressed as follows:

[0016]

[0017] In formula (1), p is the autoregressive order, is the undetermined autoregressive coefficient of the model, ε t is the error term;

[0018] The basic idea of ​​the moving average model is that the current value of the time series is independent of the value in the past, but is related to the fluctuation term in the past. The current value of the time series is linearly represented by the residual of the past time. The formula is expressed as follows:

[0019] X t =μ+ε t -θ1ε t-1 -θ2ε t-2 -…-θ q ε t-q (2)

[0020] In formula (2), μ is a constant term, q is the moving average order, and θ i is the undetermined moving average coefficient of the model, ε t is the error term;

[0021] The autoregressive moving average model (ARMA) is a combination of the AR model and the MA model. The basic idea is that the current sequence value of the time series is a linear combination of its previous value and the residual sequence of the current value and the previous value. The formula is as follows:

[0022]

[0023] Formula (3), p is the autoregressive order, q is the moving average order, is the undetermined autoregressive coefficient of the model, θ i is the undetermined moving average coefficient of the model, ε t is the error term;

[0024] The autoregressive difference moving average model ARIMA converts the non-stationary time series into a stationary time series after differentiation, and then fits it through the ARMA model;

[0025] The non-stationary time series x t The new stationary sequence after d-order difference is expressed as follows:

[0026]

[0027] In formula (4), is the difference operator, and B is the backshift operator;

[0028] Then for the stationary series w t Establish an ARMA(p,q) model, then the original sequence x t It can be expressed as an ARIMA (p, d, q) model, mathematically expressed as:

[0029]

[0030] In formula (5), θ(B)=1-θ1B-...-θ q B q , d is the number of differences performed, E(ε t ) is the error term ε t The mean of Var(ε t ) Error term ε t The variance of E(ε t ε s ) is the mean of the product of the error terms at any two different moments;

[0031] The basic idea of ​​the seasonal autoregressive difference moving average model SARIMA is to perform seasonal difference on the time series based on the ARIMA model to eliminate its seasonality and transform it into a weak stationary series.

[0032] The model is denoted as SARIMA(p, d, q)(P, D, Q) S , its general form is:

[0033]

[0034] In formula (6), p, P, q, Q represent the maximum lag order of seasonal and non-seasonal autoregressive moving average operators, D and d represent the number of non-seasonal and seasonal differences, and S represents the length of the season.

[0035] After the model is determined, the data is preprocessed and the ACF diagram of the autocorrelation and the PACF diagram of the partial autocorrelation of the original occupancy sequence are drawn. The data is then differenced and seasonally differenced and a white noise test is performed to ensure that the stationary series is not a white noise series. The white noise test method generally uses the Ljung-Box test, also known as the LB test, and its basic method is as follows:

[0036] Assumptions:

[0037]

[0038] Among them, ρ k is the autocorrelation coefficient of the k-order delay, m is the maximum delay order, and the original hypothesis H0 indicates that the sequence is independent and identically distributed white noise;

[0039] Test statistic:

[0040]

[0041] The LB(Q) statistic approximately follows the χ2 with m degrees of freedom. 2 distributed;

[0042] After data preprocessing, the model order is determined and optimized. The parameters d and D in the SARIMA model are both set to 1, and the value of S is selected according to the data period. The autoregressive order p, P and the moving average order q, Q of the model are selected according to the size of the BIC value.

[0043] BIC=ln(k)·n-2 ln(L) (9)

[0044] Where k is the number of model parameters, n is the number of samples, and L is the likelihood function; the model order with the smallest BIC value is the optimal;

[0045] The root mean square error (RMSE), explained variance (Explained variance), and mean absolute percentage error (MAPE) are selected as evaluation indicators of the model prediction results. The specific formulas are as follows:

[0046]

[0047]

[0048]

[0049] In formula (10), y represents the actual value. Represents the predicted value; Var in formula (11) represents the variance.

[0050] Furthermore, the time series dynamic parking allocation model constructed in S3 is a generalized cost-minimizing dynamic allocation model that takes into account the driver's driving time and walking time as well as the parking lot occupancy rate, specifically:

[0051] The basic idea of ​​the dynamic parking allocation model is to discretize the h decision moments in the dynamic allocation period H into h time steps, where slices = {t1, ..., t h} represents the decision time range, where t k (k=1, ..., h) represents the discretized time step of the decision moment; at each decision moment k, the set of all vehicles is defined as X(k), the attributes of all vehicles in the system are defined as I(k), and the set of parking lot attributes is defined as J(k);

[0052] X(K)={W(K), R(k), A(k)} (13)

[0053] Among them, W(k) is the set of vehicles to be assigned, R(k) is the set of waiting vehicles, and A(k) is the set of vehicles that have completed parking;

[0054] I(K) = {V(K), L(k), D(k)} (14) where y(K) is the vehicle location, L(K) is the driver’s preference type, and D(K) is the destination;

[0055] J(K)={O(k), Q(k), θ(k), E(k)} (15)

[0056] Among them, O(k) is the location coordinate of the parking lot, Q(k) is the total capacity of the parking lot, θ(k) is the parking occupancy rate, and E(k) is the number of available parking spaces in the parking lot;

[0057] The distance from vehicle i to parking lot j is represented by D ij (k)=||O j (k)-V i (k)|| represents the driving time of vehicle i to parking lot j, and t′ ij Indicates that t′ ij =f(D ij (k), v d ), v d represents the driving speed of the vehicle, f is the function of the solution time, and the time when the vehicle arrives at the parking lot is tk +t′ ij The distance from parking lot j to the destination of vehicle i is D. ji (k)=||D i (k)-O j (k)||, then the walking time t′ of the driver of vehicle i from parking lot j to the destination wji =f(D ji (k), v w ), where v w Indicates the driver's walking speed;

[0058] Mathematical model of dynamic model: at each time step t k Next, define the model as Model(k). Similar to the static model, define the 0-1 decision variable of the dynamic model as x ij , when vehicle i in the waiting vehicle set W(k) is assigned to parking lot j, x ij =1; the specific expression of the dynamic allocation model is as follows:

[0059]

[0060] t′ ij =f(D ij (k), v d ) (17)

[0061] t′ wji =f(D ji (k), v w ) (18)

[0062] St:

[0063]

[0064]

[0065] x ij ∈{0, 1}, (i∈W(k), j∈P) (21)

[0066]

[0067] The objective function (16) requires that the total parking cost of all vehicles to be assigned be minimized, where w θ The weight factors of each cost in the generalized cost; the constraint function (17) ensures that there is a parking lot P i The parking space in is assigned to vehicle i∈V k , that is, to ensure that each car is assigned a parking space; the constraint function (20) ensures that the number of vehicles allocated to parking lot j does not exceed parking lot Pi The number of available parking spaces e ij , where P is the parking lot set, which means there are a total of P parking lots. Similarly, in order to avoid the situation where there is no feasible solution to the model, the setting of virtual parking lots is introduced, and the virtual parking lot is also included in the parking lot set P. i In the example, the parking allocation result at each decision time k is represented by Model(k) (k=1, ..., h).

[0068] Furthermore, the S4 designs a hybrid heuristic algorithm based on an improved variable neighborhood search algorithm to solve the constructed time series dynamic parking allocation model, specifically:

[0069] Solving the time-driven dynamic allocation model: First, the dynamic allocation problem needs to be decomposed into a series of static allocation problems. At each decision point, the parking lot state and the set of vehicles to be allocated are updated. The update formula for the set of vehicles to be allocated at decision time k is as follows:

[0070] W(k)=(W(k-1)NW(k-1) a )∪W(k) new (twenty three)

[0071] Where W(k) represents the vehicle to be assigned at time k, W(k-1) a represents the vehicle that is successfully assigned a parking space at time k-1, W(k) new represents the vehicles that requested allocation at time k and the vehicles that re-apply for allocation; the travel time matrix T′ of all vehicles in W(k) to all parking lots in the area and the walking time matrix T″ required for all drivers to reach their destinations from each parking lot, where T′=(t′ ij ), i∈W(k), j∈Pi, T″=(t′ wij ), i∈W(k), j∈P i , then solve Model(k) based on these data. After completing the parking allocation at time k, continue to solve Model(k+1) at the next time k+1, and continue the cycle until the entire dynamic parking allocation process is completed;

[0072] Algorithm implementation process: The designed heuristic algorithm uses the variable neighborhood search algorithm (VNS) as the core algorithm, uses a greedy algorithm to generate the initial solution, and constructs a hybrid heuristic algorithm to solve the parking allocation problem. The variable neighborhood search algorithm is used to solve the lowest parking cost solution. The domain search algorithm adopts a domain structure of reallocation and cyclic exchange.

[0073] The greedy algorithm generates an initial solution: for each vehicle i, the parking cost of assigning the vehicle to each parking lot j is calculated, and the parking lots are sorted by cost. After sorting, a sequence o(i, j) of parking lots for vehicle i, sorted by cost, is obtained, where j = 1, ..., m. Among them, o(i, 1) represents the index of the parking lot with the lowest parking cost for vehicle i, o(i, 2) represents the index of the parking lot with the second lowest cost, and o(i, n) represents the index of the parking lot with the nth highest cost. In the greedy algorithm, the parking lot 0(i, 1) with the lowest parking cost is found for each vehicle i. However, if the constraints are not met and the solution is infeasible, that is, when the vehicle arrives at the parking lot, there are no available parking spaces in the parking lot or the number of vehicles already assigned to the parking lot at that time is greater than the number of available parking spaces, the vehicle will be assigned to the parking lot o(i, 2) with the second lowest cost, and so on.

[0074] Neighborhood setting for redistribution: Given a feasible solution x, its corresponding form is (i, x i ), for each car i, change its assigned parking lot x i ; This redistributes the neighborhood It can be defined as repeated redistribution k times to obtain a new solution y; for the two solutions x and y before and after redistribution, only when the distribution results of k vehicles are different, that is, x i ≠y i When , the number of different assignments between the two solutions is equal to k, and for the remaining nk vehicles, keep x i =y i ;

[0075] Neighborhood setting of cyclic exchange: Given a solution x, where (i1, j1) and (i2, j2) represent the results of pairing two vehicles with parking lots; suppose that vehicles i1 and i2 exchange their parking lots in solution x, respectively, to obtain (i1, j2) and (i2, j1) in a new solution y; exchange neighborhood The kth exchange neighborhood of solution x is defined as The exchange structure that represents the solution x after k exchanges to obtain a new solution;

[0076] Redistribute local search using the Redistribute Neighborhood Given an initial feasible solution x, in the reallocation neighborhood, each car attempts to change its assigned parking space. The cardinality is n×m, and the vehicles in the solution are redistributed to parking spaces r in the neighborhood structure. v (r v<m) times to significantly reduce it;

[0077] Cyclic exchange local search uses exchange local neighborhoods In the cyclic exchange local search, one exchange will have two cars i1, i2 and two corresponding parking lots j1, j2, and there will be four time points:

[0078] t1: the time when vehicle i1 arrives at its currently assigned parking lot j1;

[0079] t2: the time when vehicle i2 arrives at its currently assigned parking lot j2;

[0080] t3: the time when vehicle i1 arrives at its currently assigned parking lot j2;

[0081] t4: The time when vehicle i2 arrives at its currently assigned parking lot j1.

[0082] Furthermore, the performance evaluation in S5 specifically includes: generating parking demands according to historical data and certain rules based on changes in parking occupancy rates, conducting dynamic parking allocation experimental research, and evaluating the effectiveness of the model based on evaluation indicators of model effectiveness;

[0083] Experimental settings: The vehicle's driving distance and the driver's walking distance are calculated using Manhattan distance. The Manhattan distance calculation formula is as follows:

[0084] c=|x1-x2|+|y1-y2|=|Δx|+|Δy| (24)

[0085] The locations of the vehicle and its destination are expressed in latitude and longitude. When calculating mileage, the latitude and longitude are converted into distance using the following formula:

[0086]

[0087]

[0088] In formulas (25) and (26), lon1 and lat1 represent the longitude and latitude of the vehicle’s destination, lon2 and lat2 represent the longitude and latitude of the current vehicle, and R e is the radius of the Earth.

[0089] An individual optimal parking space allocation method and an overall allocation method were set up for experimental comparison;

[0090] The individual optimal parking space allocation method is specifically as follows:

[0091] S61: The driving time a driver spends driving to the parking lot closest to his destination without system allocation;

[0092] S62: When the driver arrives and finds that there are no vacancies in the parking lot described in S61, he / she tries the next nearest parking lot, consuming additional time;

[0093] S63: If there are no vacancies in the parking lot in S62, then continue to the third closest parking lot, and so on;

[0094] The overall allocation method is specifically as follows: the system allocates the current vehicle only according to the parking lot occupancy information at the current moment when the parking request is initiated.

[0095] Furthermore, the evaluation indicators of the effectiveness of the model include average driving distance, average walking distance and number of parking failures.

[0096] The beneficial effects of the present invention are:

[0097] First, the present invention proposes a parking allocation method based on parking occupancy rate prediction, which can effectively improve the user's parking success rate, reduce the repeated parking searches of vehicles, and improve parking efficiency.

[0098] Second, the parking allocation method based on prediction proposed by the present invention is in STD Des The smaller the value, that is, the more concentrated the user destinations are, the more effective the allocation of parking resources can be.

[0099] Other advantages, objects, and features of the present invention will be described in part in the following description and, in part, will be apparent to those skilled in the art upon examination of the following description or may be learned from practice of the present invention. The objects and other advantages of the present invention may be realized and obtained through the following description. BRIEF DESCRIPTION OF THE DRAWINGS

[0100] In order to make the purpose, technical solutions and advantages of the present invention more clear, the present invention will be described in detail below with reference to the accompanying drawings, in which:

[0101] Figure 1 This is a time flow chart of the parking allocation method under the cloud control system;

[0102] Figure 2 Modeling flow chart for SARIMA model;

[0103] Figure 3 This is a parking lot occupancy rate change trend chart;

[0104] Figure 4 The autocorrelation and autopartial correlation diagram of the parking lot occupancy rate series before difference;

[0105] Figure 5 The autocorrelation and autopartial correlation diagram of the parking lot occupancy rate series after difference;

[0106] Figure 6 This is a framework diagram of the parking allocation method based on parking occupancy rate prediction in the cloud control system;

[0107] Figure 7 Comparison of prediction results between ARIMA and SARIMA models;

[0108] Figure 8 Solve the flow chart for the dynamic allocation model;

[0109] Figure 9 To improve the VNS algorithm flow chart;

[0110] Figure 10 The effect diagrams of different allocation methods under the conditions of sufficient and insufficient parking spaces; Figure 10 (a) Average driving distances for different allocation methods under the conditions of sufficient and insufficient parking spaces; Figure 10 (b) Graphs of average walking distances for drivers under different allocation methods with and without sufficient parking spaces; Figure 10 (c) A graph showing the number of people who failed to park in one parking lot under different allocation methods with sufficient and insufficient parking spaces;

[0111] Figure 11 The effect diagram of different allocation methods under different request time periods; Figure 11 (a) is the average driving distance diagram of different allocation methods in each time period; Figure 11 (b) is the average walking distance of drivers in different time periods under different allocation methods; Figure 11 (c) is a graph showing the number of people who failed to park in each time period under different allocation methods;

[0112] Figure 12 The effect diagram of different allocation methods under different destination densities; Figure 12 (a) is the average driving distance of each method under different destination densities; Figure 12 (b) is the average walking distance of drivers under different destination densities for each method; Figure 12 (c) is a graph showing the number of people who failed to dock in one stop for each method under different destination densities. DETAILED DESCRIPTION

[0113] The following describes the embodiments of the present invention by means of specific examples, and those skilled in the art can easily understand other advantages and effects of the present invention from the contents disclosed in this specification. The present invention can also be implemented or applied through other different specific embodiments, and the details in this specification can also be modified or changed in various ways based on different viewpoints and applications without departing from the spirit of the present invention. It should be noted that the illustrations provided in the following embodiments are only schematic illustrations of the basic concept of the present invention, and the following embodiments and features in the embodiments can be combined with each other without conflict.

[0114] Among them, the accompanying drawings are only for illustrative purposes and represent only schematic diagrams rather than actual pictures, and should not be understood as limiting the present invention. In order to better illustrate the embodiments of the present invention, some parts of the accompanying drawings may be omitted, enlarged or reduced, and do not represent the dimensions of actual products. For those skilled in the art, it is understandable that some well-known structures and their descriptions may be omitted in the accompanying drawings.

[0115] The same or similar numbers in the drawings of the embodiments of the present invention correspond to the same or similar parts; in the description of the present invention, it should be understood that if there are terms such as "upper", "lower", "left", "right", "front", "back", etc. indicating directions or positional relationships, they are based on the directions or positional relationships shown in the drawings. They are only for the convenience of describing the present invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific direction, be constructed and operate in a specific direction. Therefore, the terms describing the positional relationship in the drawings are only used for illustrative purposes and cannot be understood as limiting the present invention. For ordinary technicians in this field, the specific meanings of the above terms can be understood according to specific circumstances.

[0116] See also Figure 6-12 As shown, a dynamic allocation method based on parking occupancy rate in a cloud system is implemented.

[0117] like Figure 6 As shown, the implementation process of the present invention includes:

[0118] S1: Propose a basic framework for the allocation method based on parking occupancy rate prediction under the architecture of the cloud control system;

[0119] S2: Parking lot occupancy prediction based on SARIMA model;

[0120] S3: For the parking allocation problem, a dynamic parking allocation model is selected;

[0121] S4: Dynamic Allocation Model Solution: Introduces the solution process of the time series dynamic parking allocation model and the algorithm implementation process.

[0122] S5: Performance evaluation: The performance of the model built in S4 is evaluated based on the real dataset from the SFPark roadside parking system project.

[0123] This example uses real parking data collected from the SFPark project in Los Angeles, USA as raw data to explain parking occupancy rate prediction, and uses the prediction results to verify the effectiveness of the dynamic parking allocation model.

[0124] (1) Determine that the parameters d and D in the SARIMA model are both 1. The autocorrelation plot shows that the sequence period is 18, so the value of S is 18. Then, by minimizing the BIC value, the optimal model is determined to be SARIMA (1,1,7) × (1,1,1). 18 .

[0125] (2) The Ljung-Box test is used to test the residual sequence of the model for white noise. The p-values ​​are much larger than 0.05 at the 1st, 6th, 12th, and 18th orders. It can be considered that the residual sequence is a white noise sequence, which can be tested by SARIMA (1,1,7)×(1,1,1) 18 The model is fitted on the training set and predictions are made on the test set.

[0126] The prediction results show that the SARIMA model is better than the ARIMA model in predicting periodic time series such as parking occupancy rate. The RMSE of the prediction results is reduced by about 22.5%; the explained variance is also closer to 1, increasing from 0.76 to 0.86, indicating that the dispersion of the predicted value and the true value is closer; the MAPE is also reduced from 14.10% to 9.66%, and the prediction error is smaller. Figure 7 It can also be seen that the predicted value of the SARIMA model is closer to the true value curve and the prediction effect is better.

[0127] (3) Figure 8 As shown in the figure, the SARIMA model prediction value is used in combination with the dynamic parking allocation model to provide users with the lowest cost parking plan and solve the model flow chart; Figure 9 As shown in the figure, the flow chart of the improved VNS algorithm in the dynamic parking allocation model.

[0128] (4) The vehicle speed is set to 40 km / h and the walking speed is set to 6 km / h. Three groups of comparative experiments are set up to verify the effectiveness of the parking allocation method proposed in this patent, namely: a. comparison of parking allocation under sufficient and tight parking spaces; b. comparison of parking allocation under different time periods; c. comparison of parking allocation under different destination densities.

[0129] a. The simulation start time is set to 17:00, and the standard deviation STD of the user destination is generated Des=3000, considering the two cases of sufficient parking spaces and parking shortage, the number of vehicles set is 100 and 300 respectively. The simulation experiments of the three parking allocation methods are carried out under the two cases of sufficient parking spaces and parking shortage. The final simulation results are as follows Figure 10 、 Figure 10 (a) Figure 10 (b) Figure 10 (c) As shown; the prediction-based dynamic parking allocation method proposed in the present invention can effectively reduce the repeated parking of vehicles, reduce the driving distance of vehicles and the number of people who fail to park in one time, especially when the number of parking spaces in the system is insufficient.

[0130] b. Explore the impact of the time period when parking allocation requests are initiated on the parking allocation model. Compare the average driving distance, walking distance, and number of people who failed to park in one parking allocation method for three parking allocation methods in different request time periods. The specific start times are: 08:00, 11:00, 14:00, 17:00, and 20:00. The specific simulation results are as follows: Figure 11 、 Figure 11 (a) Figure 11 (b) Figure 11 (c) As shown; the parking allocation method based on parking occupancy rate prediction proposed in this patent can show better results in different time periods, and improve the user's parking success rate through prediction.

[0131] c. Analyze the differences between parking allocation methods with different user destination distribution densities. In this experiment, this paper uses three STD Des Values: 1000, 3000, and 5000 generate three densities. STD Des The larger the value, the smaller the density of user destinations; on the contrary, the larger the density, the more user destinations are concentrated near the center point. The specific simulation results are as follows: Figure 12 、 Figure 12 (a) Figure 12 (b) Figure 12 (c) shows that the parking allocation method based on prediction proposed by the present invention is in STD Des The smaller the value, that is, the more concentrated the user destinations are, the more effective the allocation of parking resources can be.

[0132] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not limiting. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present invention can be modified or replaced by equivalents without departing from the purpose and scope of the technical solutions, which should all be included in the scope of the claims of the present invention.

Claims

1. A dynamic parking allocation method based on parking occupancy rate prediction in a cloud-controlled system, characterized by: The method comprises the following steps: S1: Propose the basic framework of the parking occupancy rate allocation method under the architecture of the cloud control system; S2: within the basic framework, according to the characteristics of the parking lot occupancy rate time series, the parking lot occupancy rate is predicted based on the SARIMA model to obtain a prediction result of the parking lot occupancy rate; S3: Based on the prediction results of S2, a time series dynamic parking allocation model is constructed for the dynamic characteristics of the parking problem; S4: for the constructed time series dynamic parking allocation model, designing a hybrid heuristic algorithm of an improved variable neighborhood search algorithm to solve the time series dynamic parking allocation model; S5: Evaluate the performance of the model constructed in S4; The S4 solves the time series dynamic parking allocation model by designing a hybrid heuristic algorithm based on an improved variable neighborhood search algorithm. Specifically, the algorithm is as follows: Solving the time-driven dynamic allocation model: First, the dynamic allocation problem needs to be decomposed into a series of static allocation problems. At each decision point, the parking lot state and the set of vehicles to be allocated are updated. The update formula for the set of vehicles to be allocated at decision time k is as follows: W(k)=(W(k-1)\W(k-1) a )∪W(k) new (1) Where W(k) represents the vehicle to be assigned at time k, W(k-1) a represents the vehicle that is successfully assigned a parking space at time k-1, W(k) new represents the vehicles that requested allocation at time k and the vehicles that re-apply for allocation; the travel time matrix T′ of all vehicles in W(k) to all parking lots in the area and the walking time matrix T″ required for all drivers to reach their destinations from each parking lot, where T′=(t′ ij ),i∈W(k),j∈P i , T″=(t′ wij ),i∈W(k),j∈P i , then solve Model(k) based on these data. After completing the parking allocation at time k, continue to solve Model(k+1) at the next time k+1, and continue the cycle until the entire dynamic parking allocation process is completed; Algorithm implementation process: The designed heuristic algorithm uses the variable neighborhood search algorithm (VNS) as the core algorithm, uses a greedy algorithm to generate the initial solution, and constructs a hybrid heuristic algorithm to solve the parking allocation problem. The variable neighborhood search algorithm is used to find the lowest parking cost solution. The variable neighborhood search algorithm adopts a neighborhood structure of reallocation and cyclic exchange. The greedy algorithm generates an initial solution: for each vehicle i, the parking cost of assigning the vehicle to each parking lot j is calculated, and the parking lots are sorted by cost. After sorting, a sequence o(i,j), j=1,…,m, of parking lots for vehicle i is obtained, sorted by cost. Among them, o(i,1) represents the index of the parking lot with the lowest parking cost for vehicle i, o(i,2) represents the index of the parking lot with the second lowest cost, and o(i,n) represents the index of the parking lot with the nth highest cost. In the greedy algorithm, the parking lot o(i,1) with the lowest parking cost is found for each vehicle i. However, if the constraints are not met and the solution is infeasible, that is, there are no available parking spaces in the parking lot when the vehicle arrives or the number of vehicles already assigned to the parking lot at that time is greater than the number of available parking spaces, the vehicle will be assigned to the parking lot o(i,2) with the second lowest cost, and so on. Reassigned neighborhood setting: Given a feasible solution x, its corresponding form is (i,x i ), for each car i, change its assigned parking lot x i ; This redistributes the neighborhood It is defined as repeated redistribution k times to obtain a new solution y; for the two solutions x and y before and after redistribution, only when the allocation results of k vehicles are different, that is, x i ≠y i When , the number of different assignments between the two solutions is equal to k, and for the remaining nk vehicles, keep x i =y i ; Neighborhood setting of cyclic exchange: Given a solution x, where (i1, j1) and (i2, j2) represent the results of pairing two vehicles with parking lots; suppose that vehicles i1 and i2 exchange their parking lots in solution x, respectively, to obtain (i1, j2) and (i2, j1) in a new solution y; exchange neighborhood The kth exchange neighborhood of solution x is defined as The exchange structure that represents the solution x after k exchanges to obtain a new solution; Redistribute local search using the Redistribute Neighborhood Given an initial feasible solution x, in the reallocation neighborhood, each car attempts to change its assigned parking space. The cardinality is n×m, and the vehicles in the solution are redistributed to parking spaces r in the neighborhood structure. v This can be significantly reduced by the following method, r v <m; Cyclic exchange local search uses exchange local neighborhoods In the cyclic exchange local search, one exchange will have two cars i1, i2 and two corresponding parking lots j1, j2, and there will be four time points: t1: the time when vehicle i1 arrives at its currently assigned parking lot j1; t2: the time when vehicle i2 arrives at its currently assigned parking lot j2; t3: the time when vehicle i1 arrives at its currently assigned parking lot j2; t4: The time when vehicle i2 arrives at its currently assigned parking lot j1.

2. The dynamic parking allocation method based on parking occupancy rate prediction in a cloud control system according to claim 1 is characterized by: The basic framework of S1 includes a cloud control center, an intelligent connected vehicle, an intelligent parking lot, and a parking allocation method; the cloud control center monitors the intelligent parking lot in real time through a communication network, the intelligent connected vehicle initiates a parking request to the cloud control center through the communication network, the cloud control center processes the parking information of the parking request, starts the parking allocation method, and allocates a parking plan for the intelligent connected vehicle. After the intelligent connected vehicle completes parking according to the allocated parking plan, the intelligent parking lot updates the parking lot status to the cloud control center. After the intelligent parking lot status update is completed, the intelligent connected vehicle updates the vehicle status to the cloud control center.

3. The dynamic parking allocation method based on parking occupancy rate prediction in a cloud control system according to claim 1, characterized in that: Said S2 specifically includes building a SARIMA model, preprocessing data, determining the model order, optimizing the model, testing the model adaptability and making model predictions; The SARIMA model is developed based on the autoregressive model AR and the moving average model MA; The basic idea of ​​the autoregressive model is that the time series {X t The value of the t-th period of} is only related to the value of the previous period, so its lagged value and the linear function of the current residual are expressed as follows: In formula (2), p is the autoregressive order, is the undetermined autoregressive coefficient of the model, ε t is the error term; The basic idea of ​​the moving average model is that the current value of the time series is independent of the value in the past, but is related to the fluctuation term in the past. The current value of the time series is linearly represented by the residual of the past time. The formula is expressed as follows: X t =μ+e t -θ1ε t-1 -θ2ε t-2 -…-θ q e t-q (3) In formula (3), μ is a constant term, q is the moving average order, and θ i is the undetermined moving average coefficient of the model, ε t is the error term; The autoregressive moving average model (ARMA) is a combination of the AR model and the MA model. The basic idea is that the current sequence value of the time series is a linear combination of its previous value and the residual sequence of the current value and the previous value. The formula is as follows: Formula (4), p is the autoregressive order, q is the moving average order, is the undetermined autoregressive coefficient of the model, θ i is the undetermined moving average coefficient of the model, ε t is the error term; The autoregressive difference moving average model ARIMA converts the non-stationary time series into a stationary time series after differentiation, and then fits it through the ARMA model; The non-stationary time series x t The new stationary sequence after d-order difference is expressed as follows: In formula (5), is the difference operator, and B is the backshift operator; Then for the stationary series w t Establish an ARMA(p,q) model, then the original sequence x t It is expressed as an ARIMA (p, d, q) model, and its mathematical expression is: In formula (6), θ(B)=1-θ1B-…-θ q B q , d is the number of differences performed, E(ε t ) is the error term ε t The mean of Var(ε t ) Error term ε t The variance of E(ε t ε s ) is the mean of the product of the error terms at any two different moments; The basic idea of ​​the seasonal autoregressive difference moving average model SARIMA is to perform seasonal difference on the time series based on the ARIMA model to eliminate its seasonality and transform it into a weak stationary series. The model is denoted as SARIMA(p,d,q)(P,D,Q) S , its general form is: In formula (7), p, P, q, Q represent the maximum lag order of seasonal and non-seasonal autoregressive moving average operators, D and d represent the number of non-seasonal and seasonal differences, and S represents the length of the season. After the model is determined, the data is preprocessed to draw the ACF plot of the autocorrelation and the PACF plot of the partial autocorrelation of the original occupancy sequence. The data is then differenced and seasonally differenced and a white noise test is performed to ensure that the stationary series is not a white noise series. The white noise test method generally uses the Ljung-Box test, also known as the LB test, and its basic method is as follows: Assumptions: Among them, ρ k is the autocorrelation coefficient of the k-order delay, m is the maximum delay order, and the original hypothesis H0 indicates that the sequence is independent and identically distributed white noise; Test statistic: The LB(Q) statistic approximately follows the χ2 with m degrees of freedom. 2 distributed; After data preprocessing, the model order is determined and optimized. The parameters d and D in the SARIMA model are both set to 1, and the value of S is selected according to the data period. The autoregressive order p, P and the moving average order q, Q of the model are selected according to the size of the BIC value. BIC=ln(k)·n-2ln(L) (10) Where k is the number of model parameters, n is the number of samples, and L is the likelihood function; the model order with the smallest BIC value is the optimal; The root mean square error (RMSE), explained variance (Explained variance), and mean absolute percentage error (MAPE) are selected as evaluation indicators of the model prediction results. The specific formulas are as follows: In formula (11), y represents the actual value. Represents the predicted value; Var in formula (12) represents the variance.

4. The method for dynamic parking allocation based on parking occupancy rate prediction in a cloud control system according to claim 1, characterized in that: The time series dynamic parking allocation model constructed by S3 is a generalized cost-minimizing dynamic allocation model that takes into account the driver's driving time and walking time as well as the parking lot occupancy rate. Specifically, The basic idea of ​​the dynamic parking allocation model is to discretize the h decision moments in the dynamic allocation period H into h time steps, H = {t1,…,t h } represents the decision time range, where t k , k=1,…,h represents the discretized time step of the decision moment; at each decision moment k, the set of all vehicles is defined as X(K), the attributes of all vehicles in the system are defined as I(K), and the parking lot attribute set is defined as J(K); X(K)={W(k),R(k),A(k)} (14) Among them, W(k) is the set of vehicles to be assigned, R(k) is the set of waiting vehicles, and A(k) is the set of vehicles that have completed parking; I(K)={V(k),L(k),D(k)} (15) Where V(k) is the vehicle location, L(k) is the driver’s preference type, and D(k) is the destination; J(K)={O(k),Q(k),θ(k),E(k)} (16) Among them, O(k) is the location coordinate of the parking lot, Q(k) is the total capacity of the parking lot, θ(k) is the parking occupancy rate, and E(k) is the number of available parking spaces in the parking lot; The distance from vehicle i to parking lot j is represented by D ij (k)=||o j (K)-V i (K)|| represents the driving time of vehicle i to parking lot j, and t′ ij Indicates that t′ ij =f(D ij (k),v d ), v d represents the driving speed of the vehicle, f is the function of the solution time, and the time when the vehicle arrives at the parking lot is t k +t′ ij The distance from parking lot j to the destination of vehicle i is D. ji (k)=||D i (k)-O j (k)||, then the walking time t′ of the driver of vehicle i from parking lot j to the destination wji =f(D ji (k),v w ), where v w Indicates the driver's walking speed; Mathematical model of dynamic model: at each time step t k Next, define the model as Model(k). Similar to the static model, define the 0-1 decision variable of the dynamic model as x ij , when vehicle i in the waiting vehicle set W(k) is assigned to parking lot j, x ij =1; the specific expression of the dynamic allocation model is as follows: t′ ij =f(D ij (k),v d ) (18) t′ wji =f(D ji (k),v w ) (19) st: x ij ∈{0,1},i∈W(k),j∈P(22) The objective function (17) requires that the total parking cost of all vehicles to be assigned be minimized, where w θ The weight factors of each cost in the generalized cost; the constraint function (18) ensures that there is a parking lot P i The parking space in is assigned to vehicle i∈V k , that is, to ensure that each car is assigned a parking space; the constraint function (21) ensures that the number of vehicles allocated to parking lot j does not exceed parking lot P i The number of available parking spaces ij , where P is the parking lot set, which means there are a total of P parking lots. Similarly, in order to avoid the situation where there is no feasible solution to the model, the setting of virtual parking lots is introduced, and the virtual parking lot is also included in the parking lot set P. i In the example, the parking allocation result at each decision time k is represented by Model(k), k=1,……,h.

5. The dynamic parking allocation method based on parking occupancy rate prediction in a cloud control system according to claim 1 is characterized by: The performance evaluation in S5 specifically includes: generating parking demands according to historical data and certain rules based on the changes in parking occupancy rates, conducting dynamic parking allocation experiments, and evaluating the effectiveness of the model based on the evaluation indicators of the model effectiveness; Experimental settings: The vehicle's driving distance and the driver's walking distance are calculated using Manhattan distance. The Manhattan distance calculation formula is as follows: c=|x1-x2|+|y1-y2|=|Δx|+|Δy| (24) The locations of the vehicle and its destination are expressed in latitude and longitude. When calculating mileage, the latitude and longitude are converted into distance using the following formula: In formulas (25) and (26), lon1 and lat1 represent the longitude and latitude of the vehicle’s destination, lon2 and lat2 represent the longitude and latitude of the current vehicle, and R e is the radius of the Earth; An individual optimal parking space allocation method and an overall allocation method were set up for experimental comparison; The individual optimal parking space allocation method is specifically as follows: S61: The driving time a driver spends driving to the parking lot closest to his destination without system allocation; S62: When the driver arrives and finds that there are no vacancies in the parking lot described in S61, he / she will try the next nearest parking lot, consuming additional time; S63: If there are no vacancies in the parking lot in S62, then continue to the third closest parking lot, and so on; The overall allocation method is specifically as follows: the system allocates the current vehicle only according to the parking lot occupancy information at the current moment when the parking request is initiated.

6. The dynamic parking allocation method based on parking occupancy rate prediction in a cloud control system according to claim 5 is characterized by: The evaluation indicators of the model effectiveness include average driving distance, average walking distance and number of parking failures.

Images