A smart parking management system based on 5G networking

Abstract

The present invention discloses a smart parking management system based on 5G networking, comprising: a current state acquisition unit, used to acquire the parking space state of vacant parking spaces at the current moment and the off-site environmental state outside the parking lot; a dual-feature variable unit, used to use the parking space state of vacant parking spaces at the current moment and the off-site environmental state outside the parking lot as dual-feature variables; a vehicle label acquisition unit, used to acquire the type label of the vehicle to be parked; a recommendation unit, used to input the type label of the vehicle to be parked and the dual-feature variables into a preset parking management model to obtain a parking space recommendation set for the vehicle to be parked; wherein the recommendation item of the parking space recommendation set is a binary array including a parking space matching degree and a parking space number; the present invention performs weighted optimization of the dual-objective tasks of the occupancy state and the occupancy time of the parking space, and the system can simultaneously consider whether the parking space is occupied and the occupancy time, so as to recommend a highly matching parking space for drivers with different parking needs.

Description

Technical Field

[0001] The present invention relates to the field of parking management, and specifically to a smart parking management system based on 5G networking. Background Art

[0002] With the acceleration of urbanization, the number of vehicles in cities has increased rapidly, and the problem of parking difficulties has become increasingly prominent. Existing parking management systems usually rely on traditional fixed rule recommendations, such as simple empty parking space detection or static prediction based on historical data. The Chinese patent document with patent publication number CN117196917A discloses a campus smart parking management system based on digital twins, the Internet of Things and 5G technologies, which can improve the management efficiency of parking lots, reduce the operating costs of parking lots, and provide car owners with a more convenient and comfortable parking experience.

[0003] However, most of the conventional parking management systems in the prior art make recommendations based on the current parking space status, and cannot predict the future parking space occupancy status or duration, resulting in the driver arriving at the parking space and finding that the parking space is already occupied, affecting parking efficiency. In addition, in the prior art, the parking space management system using 5G networking technology only speeds up the data transmission efficiency, and it still performs a simple sorting based on whether the parking space is free, ignoring the impact of multi-dimensional factors on the parking space recommendation sorting, resulting in the recommended parking space matching degree not being able to achieve the "smart" effect, that is, the recommended parking space is not enough to meet the driver's parking needs. Summary of the invention

[0004] In view of the shortcomings of the prior art, the present invention provides a smart parking management system based on 5G networking, which solves the technical problems raised in the background technology through the smart parking management system.

[0005] To achieve the above objectives, the present invention is implemented through the following technical solutions:

[0006] A smart parking management system based on 5G networking, applied to a parking management server, the management system comprising:

[0007] A current status acquisition unit, used to acquire the parking status of available parking spaces at the current moment and the off-site environment status outside the parking lot;

[0008] A dual-feature variable unit is used to use the parking space status of the vacant parking spaces at the current moment and the off-site environment status outside the parking lot as dual-feature variables;

[0009] A vehicle label acquisition unit, used to acquire a type label of a vehicle to be parked;

[0010] The recommendation unit is used to input the type label of the vehicle to be parked and the dual feature variable into a preset parking management model to obtain a parking space recommendation set for the vehicle to be parked; wherein the recommendation item of the parking space recommendation set is a binary array including a parking space matching degree and a parking space number.

[0011] In some embodiments, the method for constructing the parking management model includes:

[0012] S1, obtain the historical state combination matrix of the parking lot in time series;

[0013] S2, training the LSTM model using the historical state combination matrix after data standardization;

[0014] S3, using the trained LSTM model that reaches the convergence condition as the parking management model.

[0015] In some embodiments, step S1 includes:

[0016] S1-1, collecting the parking space status of all parking spaces in the parking lot in time series and defining it as a parking space status matrix;

[0017] The expression of the parking space state matrix is:

[0018]

[0019] Among them, P t-W+1:t represents the parking space state matrix from t-W+1 to t time node, whose dimension is W×M×N; W represents the length of the time node, M represents the parking space number, and the parking space number is a continuous sequence; N represents the feature label of each parking space; the feature label of each parking space includes: parking space occupancy status, entry time, and exit time; Indicates the state of a parking space with parking space number M and feature label N at time point t;

[0020] S1-2, collecting the parking behaviors of the vehicles parked in the parking lot in a time series and defining them as a parking behavior matrix;

[0021] The expression of the parking behavior matrix is:

[0022]

[0023] Among them, F t-W+1:t represents the parking behavior matrix from t-W+1 to time node t, with the dimension of W×K×P;

[0024] K represents the type label of the parked vehicle, and P represents the historical behavior label of each vehicle; the historical behavior label of the vehicle includes: parking time, entry time, and exit time; represents the parking behavior at time point t, when the vehicle type label is K and the vehicle's historical behavior label is P;

[0025] S1-3, collecting the off-site environmental status outside the parking lot and defining it as an off-site environmental status matrix;

[0026] The expression of the off-site environment state matrix is:

[0027]

[0028] Among them, E t-W+1:t represents the off-site environment state matrix from t-W+1 to time node t, with a dimension of W×Q;

[0029] Q represents the off-site environment label, Indicates the off-site environment state at time point t, with the off-site environment label Q; the off-site environment label includes: a weather continuous label or a traffic continuous label;

[0030] S1-4, the parking space state matrix, the parking behavior matrix and the off-site environment state matrix are combined to obtain a historical state combination matrix, the expression of which is:

[0031] C t-W+1:t = {P t-W+1:t ,F t-W+1:t ,E t-W+1:t};

[0032] Among them, C t-W+1:t It represents the historical state combination matrix that combines the parking space status, parking behavior, and off-site environment status cases from t-W+1 to t time node; its dimensional expression is: W×(M×N+K×P+Q)=W×D; where D represents the dimension of the combined parking space status, parking behavior, and off-site environment status.

[0033] In some embodiments, the step S2 includes:

[0034] The historical state combination matrix after data normalization is used as the LSTM model input to train the parking management model; wherein the training goal of the parking management model is to minimize the total model loss, and the total model loss is the weighted sum loss of minimizing the parking space occupancy state loss and minimizing the parking space occupancy time loss.

[0035] In some embodiments, the historical state combination matrix after data normalization is used as the LSTM model input to train the parking management model, including:

[0036] S2-1, receives the historical state combination matrix after data standardization as the input of the LSTM model;

[0037] S2-2, the LSTM model forward propagates the historical state combination matrix to obtain the hidden state vector at the last time point t;

[0038] The expression of the hidden state vector at the last time point t is:

[0039] h t =f LSTM (C t-W+1:t ;θ);

[0040] Among them, h t represents the hidden state vector at time point t, f LSTM Represents the function of the model's forward propagation of the input, C t-W+1:t is the historical state combination matrix input from time point t-W+1 to t, and θ represents the parameters of the LSTM model;

[0041] S2-3, output the first target task value and the second target task value according to the hidden state vector at the last time point t; the expression of the first target task value is:

[0042]

[0043] in, represents the first target task value at time point t, representing the probability that the parking space will be occupied at the next time point t+1; σ is the sigmoid activation function, which is used to limit the output to [0,1]; W s The weight parameter representing the parking space occupancy status, b s Bias parameter representing the parking space occupancy status;

[0044] The expression of the second target task value is:

[0045]

[0046] in, W represents the second target task value, which represents the parking space occupancy time at the next time point t+1; t is the weight parameter of parking space occupancy time, b t is the bias parameter of the parking space occupancy time;

[0047] S2-4, determining a total loss function according to the first target task value and the second target task value;

[0048] S2-5, taking the total loss function as the target, updating the parking management model;

[0049] In some of the embodiments, determining the total loss function according to the first target task value and the second target task value includes:

[0050] S2-4-1, according to the first target task value, determine the cross entropy loss of the first target task value

[0051] The cross entropy loss function of the first target task value is:

[0052]

[0053] Among them, L status represents the cross entropy loss of the first target task value, represents the actual occupancy status of the mth parking space at historical time point t, represents the occupancy probability of the mth parking space predicted by the model;

[0054] S2-4-2, determining a mean square error loss of the second target task value according to the second target task value;

[0055] The mean square error loss function of the second target task value is:

[0056]

[0057] Among them, L time represents the mean square error loss of the second target task value, represents the actual occupancy time of the mth parking space at the historical time point t, represents the occupancy time of the mth parking space predicted by the model;

[0058] S2-4-3, determining the total loss of the model according to the cross entropy loss of the first task value and the mean square error loss of the second task value;

[0059] The total loss function of the model is:

[0060] L total =λ1L status +λ2L time ;

[0061] Among them, L total represents the total model loss, λ1 represents the weight of the cross entropy loss in the total model loss, and λ2 represents the mean square error loss.

[0062] In some embodiments, the parking management model is updated by taking the total loss function as a target, including:

[0063] S2-5-1, backpropagate the model using the chain rule to determine the gradient of the total loss function with respect to each parameter;

[0064] The gradient expression for each parameter is:

[0065]

[0066] in,

[0067] Represents the total loss function L tota l The gradient of the weight parameter relative to the parking space occupancy state;

[0068] Represents the total loss function L total The gradient of the bias parameter relative to the parking space occupancy state;

[0069] Represents the total loss function L total The gradient of the weight parameter relative to the parking space occupancy time;

[0070] Represents the total loss function L total The gradient of the bias parameter relative to the parking space occupancy time;

[0071] Indicates the partial derivative symbol, representing the total loss function L total partial derivatives with respect to the parameters;

[0072] S2-5-2, using a preset learning rate to substitute the gradient to obtain updated parameters;

[0073] The expression of the updated parameter is:

[0074]

[0075] in, represents the weight parameter of the updated parking space occupancy status, The offset parameter representing the updated parking space occupancy status, represents the weight parameter of the updated parking space occupancy time, The offset parameter representing the updated parking space occupancy time.

[0076] S2-5-3, use the updated parameters to perform the next round of forward propagation and backward propagation of the LSTM model until the total loss function converges, and use the converged LSTM model as the parking management model.

[0077] In some embodiments, the step of constructing the parking space recommendation set includes:

[0078] Receiving a type tag of a vehicle to be parked;

[0079] According to the type label of the vehicle to be parked, matching the historical behavior label in the parking management model;

[0080] Based on the matched historical behavior labels and dual feature variables, the parking management model outputs the predicted occupancy status of the vacant parking spaces. And the predicted value of occupancy time

[0081] The predicted value of occupancy status and the predicted value of occupancy duration are weighted inversely to determine the matching degree of available parking spaces;

[0082] The weighted inverse expression is:

[0083]

[0084] Among them, S m is the matching degree of the mth vacant parking space, α and β are weighted coefficients, satisfying α+β=1;

[0085] The vacant parking spaces are sorted in ascending order of matching degree, and the number of vacant parking spaces whose matching degree falls within the threshold range [S min ,S max ] as the index to match the corresponding parking space number;

[0086] The expression for ascending sorting is: S min ≤S m ≤S max ,

[0087] Among them, the matching degree satisfies the expression: S m =S1,S2,…,S M ;

[0088] Combine the vacant parking space matching degree and the matched parking space number into a recommendation item, and output the combined recommendation set;

[0089] The expression of the recommendation set is:

[0090] R={(S m ,p m )∣S min ≤S m ≤S max};

[0091] Among them, R represents the recommendation set, p m Indicates the parking space number of the mth parking space.

[0092] The present invention provides a smart parking management system based on 5G networking. By performing weighted optimization of the dual-objective tasks of the occupancy status and occupancy duration of parking spaces, the system can simultaneously consider whether the parking space is occupied and the occupancy duration, thereby being able to recommend highly matched parking spaces for drivers with different parking needs. BRIEF DESCRIPTION OF THE DRAWINGS

[0093] Figure 1This is a structural block diagram of a smart parking management system based on 5G networking in the present invention;

[0094] Figure 2 This is a step diagram of a model building method of a smart parking management system based on 5G networking in the present invention;

[0095] Figure 3 This is a model training diagram of a smart parking management system based on 5G networking in the present invention; DETAILED DESCRIPTION

[0096] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0097] First, the prior art and related concepts involved in the embodiments of the present invention are described:

[0098] Fully connected layer: The fully connected layer is the basic building block in the neural network. Each neuron is connected to all neurons in the previous layer to form a dense network of connections. Its function is to perform a weighted summation on the output of the previous layer, and then perform a nonlinear transformation through the activation function to obtain the input of the next layer.

[0099] Sigmoid activation function: It is a commonly used nonlinear activation function, and its output value is between 0 and 1. It is often used in binary classification tasks and can compress any real value into the interval [0,1]. Its function expression is: x is regarded as the original input data, and σ is the compressed data.

[0100] LSTM model: The LSTM (Long Short-Term Memory) model is a special recurrent neural network (RNN) used to process and predict time series data. It can better capture the relationship between long-term and short-term information when processing long sequence data, thereby improving prediction and analysis results.

[0101] refer to Figures 1 to 3 In an embodiment of the present invention, a smart parking management system based on 5G networking is provided.

[0102] Embodiment 1 discloses a smart parking management system based on 5G networking, which is applied to a parking management server and includes:

[0103] A current status acquisition unit, used to acquire the parking status of available parking spaces at the current moment and the off-site environment status outside the parking lot;

[0104] A dual-feature variable unit is used to use the parking space status of the vacant parking spaces at the current moment and the off-site environment status outside the parking lot as dual-feature variables;

[0105] A vehicle label acquisition unit, used to acquire a type label of a vehicle to be parked;

[0106] The recommendation unit is used to input the type label of the vehicle to be parked and the dual feature variable into a preset parking management model to obtain a parking space recommendation set for the vehicle to be parked; wherein the recommendation item of the parking space recommendation set is a binary array including a parking space matching degree and a parking space number.

[0107] Embodiment 2: The technical solution of Embodiment 2 is different from that of Embodiment 1 in that: a method for constructing the parking management model in Embodiment 1 is disclosed, and the method for constructing the parking management model includes:

[0108] Step S1, obtaining the historical state combination matrix of the parking lot in time series, which specifically includes:

[0109] S1-1, collecting the parking space status of all parking spaces in the parking lot in time series and defining it as a parking space status matrix;

[0110] The expression of the parking space state matrix is:

[0111]

[0112] Among them, P t-W+1:t represents the parking space status matrix from t-W+1 to time node t, and its dimension is W×M×N; W represents the length of the time node (the past W time points), M represents the parking space number, and the parking space number is a continuous sequence; N represents the feature label of each parking space; the feature label of the parking space includes: parking space occupancy status, entry time, and exit time;

[0113] Indicates the state of a parking space with parking space number M and feature label N at time point t;

[0114] For example, It can be said that at midnight, the parking space numbered 0012 has a characteristic label of "occupied", so a vector value that can describe the parking space status is constructed with these three features; then, the corresponding vector values ​​are collected in the time series from t-W+1 to t, and a matrix that can describe the parking space status in the time series is constructed with the corresponding vector values; the characteristics of each vector in the matrix include the parking space occupancy status, the entry time and exit time of the occupied vehicle.

[0115] S1-2, collecting the parking behaviors of the vehicles parked in the parking lot in a time series and defining them as a parking behavior matrix;

[0116] The expression of parking behavior matrix is:

[0117]

[0118] Among them, F t-W+1:t represents the parking behavior matrix from t-W+1 to time node t, with the dimension of W×K×P;

[0119] K represents the type label of the parked vehicle, and P represents the historical behavior label of each vehicle; the historical behavior label of the vehicle includes: parking time, entry time, and exit time

[0120] represents the parking behavior at time point t, when the vehicle type label is K and the vehicle's historical behavior label is P;

[0121] For example, It can represent that at 12 noon, the vehicle type label is 093391, and the historical behavior label of the vehicle is "entry time at 10:30 am", so as to construct a vector value that can describe the parking behavior with these three features; then, in the time series to t, the corresponding vector values ​​are collected, and a matrix that can describe the parking behavior in the time series is constructed with the corresponding vector values; the characteristics of each vector in the matrix include the parking time of the vehicle (calculated by the number of parking times in the unit time, and the unit time can be in months or weeks), the entry time and the exit time when the vehicle is parked.

[0122] S1-3, collecting the off-site environmental status outside the parking lot and defining it as an off-site environmental status matrix;

[0123] The expression of the off-site environmental state matrix is:

[0124]

[0125] Among them, E t-W+1:t represents the off-site environment state matrix from t-W+1 to time node t, with a dimension of W×Q;

[0126] Q represents the off-site environment label, It indicates the off-site environment state at time point t with off-site environment label Q; the off-site environment label includes: weather continuity label or traffic continuity label.

[0127] Specifically, the weather continuous label can be based on whether it is raining outside the parking lot as a reference. For example, the weather such as clear, cloudy, overcast, light rain, moderate rain, and heavy rain can be constructed as weather continuous labels according to levels 1 to 6; similarly, smooth, slow, congested, and extremely congested can also be used as traffic continuous labels, etc.

[0128] For example, It can be said that at 3 pm, the feature labels outside the parking lot are sunny and extremely congested. In other words, the off-site environment label can be a feature label of a category, or double labels can be used to enrich the off-site environment state, thereby constructing a richer off-site environment state matrix.

[0129] S1-4, merging the parking space state matrix, the parking behavior matrix and the off-site environment state matrix to obtain a historical state combination matrix;

[0130] The expression of the historical state combination matrix is:

[0131] C t-W+1:t = {P t-W+1:t ,F t-W+1:t ,E t-W+1:t};

[0132] Among them, C t-W+1:t It represents the historical state combination matrix that concatenates the parking space status, parking behavior, and off-site environment status cases from time node t-W+1 to time node t. Its dimension expression is: W×(M×N+K×P+Q)=W×D; where D represents the combined dimension of parking space status, parking behavior, and off-site environment status.

[0133] The construction method also includes:

[0134] Step S2, using the historical state combination matrix after data standardization to train the LSTM model, specifically includes:

[0135] The historical state combination matrix after data standardization is used as the input of the LSTM model to train the parking management model. The training goal of the parking management model is to minimize the total loss of the model, which is the weighted sum loss of minimizing the parking space occupancy state loss and minimizing the parking space occupancy time loss.

[0136] Specifically, the role of data normalization in this embodiment is to ensure that different features (such as parking space status, vehicle behavior, external environment, etc.) are in the same numerical range or scale, so as to prevent certain features from having too much impact on the model training process due to their large numerical range.

[0137] Specifically, different data features in a parking management system may have different dimensions. For example, whether a parking space is occupied is binary data (0 or 1), while the occupancy time may be continuous data expressed in hours or minutes. Without standardization, features with larger values ​​(such as occupancy time) may dominate the model's learning process, causing the model to ignore features with smaller values ​​but equally important (such as parking space occupancy status). Through data standardization, the numerical range of each feature can be adjusted to the same scale (such as converted to the interval of 0 to 1), so that the LSTM model can treat each feature equally during training, improving the stability of the model and the training effect. This allows the model to more accurately learn and predict complex time series data in parking management.

[0138] Furthermore, the historical state combination matrix after data standardization is used as the input of the LSTM model to train the parking management model, including:

[0139] S2-1, receives the historical state combination matrix after data standardization as the input of the LSTM model;

[0140] The input data is the historical state combination matrix of parking space status, parking behavior and off-site environment status from time point t-W+1 to t.

[0141] S2-2, the LSTM model forward propagates the historical state combination matrix to obtain the hidden state vector at the last time point t;

[0142] The LSTM model receives the historical state combination matrix from time point t-W+1 to t, gradually processes the data at these time points, generates the hidden state vector at each time point, and finally outputs the hidden state vector h at time point t. t .

[0143] The expression of the hidden state vector at the last time point t is:

[0144] h t =f LSTM (C t-W+1:t ;θ);

[0145] Among them, h t represents the hidden state vector at time point t, f LSTM Represents the function of the model's forward propagation of the input, C t-W+1:t is the historical state combination matrix input from time point t-W+1 to t, and θ represents the parameters of the LSTM model;

[0146] Specifically, the LSTM layer of the LSTM model processes time series data through its internal memory units and gating mechanisms (input gate, forget gate, and output gate) to generate a hidden state vector h tis a comprehensive summary of all time steps before time point t. The hidden state vector h t It can be regarded as a summary of the entire time series features in the historical state combination matrix, which compresses all the information of the past W time points. Furthermore, LSTM can capture the dependencies in time series data, so the hidden state vector h t It can contain the dependency information of the previous time step and provide a basis for future predictions.

[0147] S2-3, output the first target task value and the second target task value according to the hidden state vector at the last time point t;

[0148] The expression of the first target task value is:

[0149]

[0150] in, represents the first target task value at time point t, representing the probability that the parking space will be occupied at the next time point t+1; σ is the sigmoid activation function, which is used to limit the output to [0,1]; W s The weight parameter representing the parking space occupancy status, b s Bias parameter representing the parking space occupancy status;

[0151] The expression of the second target task value is:

[0152]

[0153] in, W represents the second target task value, which represents the parking space occupancy time at the next time point t+1; t is the weight parameter of parking space occupancy time, b t is the bias parameter of the parking space occupancy time;

[0154] Specifically, the hidden state vector h output by the LSTM model is t Through a fully connected layer with a sigmoid activation function, the probability of the parking space occupancy state at the next time point t+1 is output, and through another fully connected layer without a sigmoid activation function, the parking space occupancy time is predicted.

[0155] S2-4, determining a total loss function according to the first target task value and the second target task value;

[0156] Among them, S2-4 further includes:

[0157] S2-4-1, according to the first target task value, determine the cross entropy loss of the first target task value

[0158] The cross entropy loss function of the first target task value is:

[0159]

[0160] Among them, L status represents the cross entropy loss of the first target task value, represents the actual occupancy status of the mth parking space at historical time point t, represents the occupancy probability of the mth parking space predicted by the model;

[0161] S2-4-2, determining a mean square error loss of the second target task value according to the second target task value;

[0162] The mean square error loss function of the second target task value is:

[0163]

[0164] Among them, L time represents the mean square error loss of the second target task value, represents the actual occupancy time of the mth parking space at the historical time point t, represents the occupancy time of the mth parking space predicted by the model;

[0165] S2-4-3, determining the total loss of the model according to the cross entropy loss of the first task value and the mean square error loss of the second task value;

[0166] The total loss function of the model is:

[0167] L total =λ1L status +λ2L time ;

[0168] Among them, L total represents the total model loss, λ1 represents the weight of the cross entropy loss in the total model loss, and λ2 represents the mean square error loss.

[0169] S2-5, taking the total loss function as the target, the parking management model is updated.

[0170] Furthermore, S2-5 also includes:

[0171] S2-5-1, backpropagate the model using the chain rule to determine the gradient of the total loss function with respect to each parameter;

[0172] The gradient of each parameter represents the rate of change of the parameter, that is, the total loss function L total The rate of change as the parameters change. During backpropagation, the gradient of each parameter is calculated layer by layer.

[0173] The gradient expression for each parameter is:

[0174]

[0175] in,

[0176] Represents the total loss function L total The gradient of the weight parameter relative to the parking space occupancy state;

[0177] Represents the total loss function L total The gradient of the bias parameter relative to the parking space occupancy state;

[0178] Represents the total loss function L total The gradient of the weight parameter relative to the parking space occupancy time;

[0179] Represents the total loss function L total The gradient of the bias parameter relative to the parking space occupancy time;

[0180] Indicates the partial derivative symbol, representing the total loss function L total Partial derivatives with respect to parameters;

[0181] S2-5-2, use the preset learning rate to substitute the gradient to obtain the updated parameters;

[0182] The expression of the updated parameter is:

[0183]

[0184] in, represents the weight parameter of the updated parking space occupancy status, The offset parameter representing the updated parking space occupancy status, represents the weight parameter of the updated parking space occupancy time, The offset parameter representing the updated parking space occupancy time.

[0185] S2-5-3, use the updated parameters to perform the next round of forward propagation and back propagation of the LSTM model until the total loss function converges, and use the converged LSTM model as the parking management model.

[0186] The construction method also includes:

[0187] Step S3, using the trained LSTM model that has reached the convergence condition as the parking management model.

[0188] In this embodiment, the convergence condition can be a common convergence condition, such as the total loss function no longer decreases significantly or reaches a preset number of iterations, etc.

[0189] Through the above steps, a parking management model for outputting a parking space recommendation set for vehicles to be parked can be obtained, and a parking space recommendation item including a parking space matching degree and a parking space number can be obtained through the parking space recommendation set.

[0190] Based on Example 2, the steps of constructing a parking space recommendation set include:

[0191] Receiving a type tag of a vehicle to be parked;

[0192] According to the type label of the vehicle to be parked, the historical behavior label is matched in the parking management model;

[0193] Based on the matched historical behavior labels and dual feature variables, the parking management model outputs the predicted occupancy status of the vacant parking spaces. And the predicted value of occupancy time

[0194] Of course, the occupancy state prediction value and occupancy time prediction value are based on the prediction values ​​of the corresponding time nodes of several time steps after the driver's current time node; and these several time steps are generally the empirical time between the moment the driver enters the parking lot and arrives at the parking space to be parked. This time can be based on the average time to find a parking space in the parking lot or preset by the parking lot manager according to the size of the parking lot.

[0195] The predicted value of occupancy status and the predicted value of occupancy duration are weighted inversely to determine the matching degree of available parking spaces;

[0196] The weighted inverse expression is:

[0197]

[0198] Among them, S m is the matching degree of the mth vacant parking space, α and β are weighted coefficients, satisfying α+β=1;

[0199] Sort the available parking spaces in ascending order and select the spaces with matching degrees within the threshold range [S min ,S max ] as the index to match the corresponding parking space number;

[0200] The expression for ascending sorting is: S min ≤S m ≤S max ,

[0201] Among them, the matching degree satisfies the expression: S m =S1,S2,…,S M ;

[0202] Combine the vacant parking space matching degree and the matched parking space number into a recommendation item, and output the combined recommendation set;

[0203] The expression of the recommendation set is:

[0204] R={(S m ,p m )∣S min ≤S m ≤S max};

[0205] Among them, R represents the recommendation set, p m Indicates the parking space number of the mth parking space.

[0206] Specifically, the matching degree S m The smaller the value, the higher the predicted value of occupancy status. Or the predicted value of the duration The larger the matching degree S is, the higher the probability that the parking space is occupied or the longer it is occupied, and the less suitable it is for parking. m The bigger it is, the better it is for parking.

[0207] The above-mentioned embodiment is a smart parking management system based on 5G networking. From the application perspective of the management system, the system can dynamically adjust the parking space recommendation set based on real-time data by obtaining the status of vacant parking spaces and the environmental status outside the parking lot (such as weather and traffic conditions) in real time, and recommend real-time high-matching parking spaces to drivers, thereby reducing the time it takes for vehicles to find parking spaces.

[0208] On the other hand, the system recommends highly matching parking spaces based on the historical parking behaviors and real-time environmental data of different types of vehicles (such as passenger cars, large vehicles, disabled vehicles, etc.), thereby optimizing the parking experience of different types of vehicles and reducing invalid recommendations.

[0209] On the other hand, the system predicts the future occupancy status (occupancy probability, occupancy duration) of parking spaces based on the LSTM model, and can inform drivers in advance that certain parking spaces will soon be vacant or remain idle, thereby improving the flexibility of recommendations.

[0210] On the other hand, the system generates a parking space matching degree by performing a weighted inverse calculation on the predicted values ​​of parking space occupancy status and duration, and recommends highly matching parking spaces based on the matching degree ranking, ensuring that parking spaces with higher matching degrees are more suitable for parking, thereby improving the accuracy of recommendations.

[0211] In the above-mentioned embodiment, a smart parking management system based on 5G networking, from the perspective of model construction, by constructing a historical state combination matrix (including data in multiple dimensions such as parking space status, parking behavior, and off-site environment), the system can integrate the historical behavior and current status of the parking lot, thereby providing the model with richer training data.

[0212] Using the LSTM model to train the historical state combination matrix can effectively capture the time dependency of different states in the parking lot. The time series modeling method enables the system to predict the future occupancy of parking spaces, thereby more accurately recommending vacant parking spaces, which is particularly suitable for scenarios where parking space occupancy has time-series dependency.

[0213] During the model training process, by performing weighted optimization of the dual-objective tasks of the parking space occupancy status and occupancy duration, the model can simultaneously consider whether the parking space is occupied and the occupancy duration, and can recommend highly matching parking spaces for drivers with different parking needs.

[0214] Through the gradient descent and back propagation mechanism of the LSTM model, the system continuously optimizes the model parameters during the training process, allowing the loss function to gradually converge, thereby ensuring that the model's predictive performance can be continuously improved.

[0215] The above embodiments can be implemented in whole or in part by software, hardware, firmware or other arbitrary combinations. When implemented using software, the above embodiments can be implemented in whole or in part in the form of a computer program product. The computer program product includes one or more computer instructions or computer programs. When the computer instructions or computer programs are loaded or executed on a computer, the process or function described in the embodiment of the present application is generated in whole or in part. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable devices. The computer instructions can be stored in a computer-readable storage medium, or transmitted from one computer-readable storage medium to another computer-readable storage medium, for example, the computer instructions can be transmitted from a website site, computer, server or data center by wired (e.g., infrared, wireless, microwave, etc.) mode to another website site, computer, server or data center.

[0216] The computer-readable storage medium may be any available medium that can be accessed by a computer or a data storage device such as a server or a data center that includes one or more available media. The available medium may be a magnetic medium (e.g., a floppy disk, a hard disk, a tape), an optical medium (e.g., a DVD), or a semiconductor medium. The semiconductor medium may be a solid-state hard disk.

[0217] The above description is only a specific implementation manner of the present application, but the protection scope of the present application is not limited thereto. Any technician familiar with the technical field can easily think of changes or substitutions within the technical scope disclosed in the present application, which should be included in the protection scope of the present application.

Claims

1. A smart parking management system based on 5G networking, applied to a parking management server, characterized in that: The management system comprises: A current status acquisition unit, used to acquire the parking status of available parking spaces at the current moment and the off-site environment status outside the parking lot; A dual-feature variable unit is used to use the parking space status of the vacant parking spaces at the current moment and the off-site environment status outside the parking lot as dual-feature variables; A vehicle label acquisition unit, used to acquire a type label of a vehicle to be parked; A recommendation unit, used for inputting the type label of the vehicle to be parked and the dual feature variable into a preset parking management model to obtain a parking space recommendation set for the vehicle to be parked; wherein the recommendation item of the parking space recommendation set is a binary array including a parking space matching degree and a parking space number; The method for constructing the parking management model includes: S1, obtain the historical state combination matrix of the parking lot in time series; S2, training the LSTM model using the historical state combination matrix after data standardization; S3, using the trained LSTM model that reaches the convergence condition as the parking management model; Get the historical status combination matrix of the parking lot in time series, including: S1-1, collecting the parking space status of all parking spaces in the parking lot in time series and defining it as a parking space status matrix; The expression of the parking space state matrix is: Among them, P t-W+1:t represents the parking space state matrix from t-W+1 to t time node, whose dimension is W×M×N; W represents the length of the time node, M represents the parking space number, and the parking space number is a continuous sequence; N represents the feature label of each parking space; the feature label of each parking space includes: parking space occupancy status, entry time, and exit time; Indicates the state of a parking space with parking space number M and feature label N at time point t; S1-2, collecting the parking behaviors of the vehicles parked in the parking lot in a time series and defining them as a parking behavior matrix; The expression of the parking behavior matrix is: Among them, F t-W+1:t represents the parking behavior matrix from t-W+1 to time node t, with the dimension of W×K×P; K represents the type label of the parked vehicle, and P represents the historical behavior label of each vehicle; the historical behavior label of the vehicle includes: parking time, entry time, and exit time; represents the parking behavior at time point t, when the vehicle type label is K and the vehicle's historical behavior label is P; S1-3, collecting the off-site environmental status outside the parking lot and defining it as an off-site environmental status matrix; The expression of the off-site environment state matrix is: Among them, E t-W+1:T represents the off-site environment state matrix from t-W+1 to time node t, with a dimension of W×Q; Q represents the off-site environment label, Indicates the off-site environment state at time point t, with the off-site environment label Q; the off-site environment label includes: a weather continuous label or a traffic continuous label; S1-4, the parking space state matrix, the parking behavior matrix and the off-site environment state matrix are combined to obtain a historical state combination matrix, the expression of which is: C t-W+1:t =P t-W+1:t ,F t-W+1:t ,E t-W+1:t ; Among them, C t-W+1:t It represents the historical state combination matrix that combines the parking space status, parking behavior, and off-site environment status cases from t-W+1 to t time node; its dimensional expression is: W×(M×N+K×P+Q)=W×D; where D represents the dimension of the combined parking space status, parking behavior, and off-site environment status.

2. According to claim 1, a smart parking management system based on 5G networking is characterized in that: The step S2 comprises: The historical state combination matrix after data normalization is used as the LSTM model input to train the parking management model; wherein the training goal of the parking management model is to minimize the total model loss, and the total model loss is the weighted sum loss of minimizing the parking space occupancy state loss and minimizing the parking space occupancy time loss.

3. According to claim 2, a smart parking management system based on 5G networking is characterized in that: The historical state combination matrix after data standardization is used as the LSTM model input to train the parking management model, including: S2-1, receives the historical state combination matrix after data standardization as the input of the LSTM model; S2-2, the LSTM model forward propagates the historical state combination matrix to obtain the hidden state vector at the last time point t; The expression of the hidden state vector at the last time point t is: h t =f LSTM (C t-W+1:t ;θ); Among them, h t represents the hidden state vector at time point t, f LSTM Represents the function of the model's forward propagation of the input, C t-W+1:t is the historical state combination matrix input from time point t-W+1 to t, and θ represents the parameters of the LSTM model; S2-3, output the first target task value and the second target task value according to the hidden state vector at the last time point t; The expression of the first target task value is: in, represents the first target task value at time point t, representing the probability that the parking space will be occupied at the next time point t+1; σ is the sigmoid activation function, which is used to limit the output to [0,1]; W s The weight parameter representing the parking space occupancy status, b s Bias parameter representing the parking space occupancy status; The expression of the second target task value is: in, W represents the second target task value, which represents the parking space occupancy time at the next time point t+1; t is the weight parameter of parking space occupancy time, b t is the bias parameter of the parking space occupancy time; S2-4, determining a total loss function according to the first target task value and the second target task value; S2-5, taking the total loss function as the target, updating the parking management model.

4. According to claim 3, a smart parking management system based on 5G networking is characterized in that: According to the first objective task value and the second objective task value, a total loss function is determined, including: S2-4-1, determining a cross entropy loss of the first target task value according to the first target task value; The calculation formula of the cross entropy loss of the first target task value is: Among them, L status represents the cross entropy loss of the first target task value, represents the actual occupancy status of the mth parking space at historical time point t, represents the occupancy probability of the mth parking space predicted by the model; S2-4-2, determining a mean square error loss of the second target task value according to the second target task value; The calculation formula of the mean square error loss of the second target task value is: Among them, L time represents the mean square error loss of the second target task value, represents the actual occupancy time of the mth parking space at the historical time point t, represents the occupancy time of the mth parking space predicted by the model; S2-4-3, determining the total loss of the model according to the cross entropy loss of the first task value and the mean square error loss of the second task value; The total loss of the model is calculated as: L total =λ1L status +λ2L time ; Among them, L total represents the total model loss, λ1 represents the weight of the cross entropy loss in the total model loss, and λ2 represents the mean square error loss.

5. According to claim 4, a smart parking management system based on 5G networking is characterized in that: Taking the total loss function as the target, the parking management model is updated, including: S2-5-1, backpropagate the model using the chain rule to determine the gradient of the total loss function with respect to each parameter; The gradient expression for each parameter is: in, Represents the total loss function L total The gradient of the weight parameter relative to the parking space occupancy state; Represents the total loss function L total The gradient of the bias parameter relative to the parking space occupancy state; Represents the total loss function L total The gradient of the weight parameter relative to the parking space occupancy time; Represents the total loss function L total The gradient of the bias parameter relative to the parking space occupancy time; Indicates the partial derivative symbol, representing the total loss function L total partial derivatives with respect to the parameters; S2-5-2, using a preset learning rate to substitute the gradient to obtain updated parameters; The expression of the updated parameter is: in, represents the weight parameter of the updated parking space occupancy status, The offset parameter representing the updated parking space occupancy status, represents the weight parameter of the updated parking space occupancy time, The offset parameter representing the updated parking space occupancy time; S2-5-3, use the updated parameters to perform the next round of forward propagation and backward propagation of the LSTM model until the total loss function converges, and use the converged LSTM model as the parking management model.

6. According to claim 5, a smart parking management system based on 5G networking is characterized in that: The steps of constructing the parking space recommendation set include: Receiving a type tag of a vehicle to be parked; According to the type label of the vehicle to be parked, matching the historical behavior label in the parking management model; Based on the matched historical behavior labels and dual feature variables, the parking management model outputs the predicted occupancy status of the vacant parking spaces. And the predicted value of occupancy time The predicted value of occupancy status and the predicted value of occupancy duration are weighted inversely to determine the matching degree of available parking spaces; The calculation formula of the weighted inverse ratio is: Among them, S m is the matching degree of the mth vacant parking space, α and β are weighted coefficients, satisfying α+β=1; The vacant parking spaces are sorted in ascending order of matching degree, and the number of vacant parking spaces whose matching degree falls within the threshold range [S min ,S max ] as the index to match the corresponding parking space number; The expression for ascending sorting is: S min ≤S m ≤S max , Among them, the matching degree satisfies the expression: S m =S1,S2,…,S M ; Combine the vacant parking space matching degree and the matched parking space number into a recommendation item, and output the combined recommendation set; The expression of the recommendation set is: R={(S m ,p m )∣S min ≤S m ≤S max }; Among them, R represents the recommendation set, p m Indicates the parking space number of the mth parking space.

Images