WO2019119319A1 - 停车场泊位占用率预测方法、装置、设备及存储介质 - Google Patents

停车场泊位占用率预测方法、装置、设备及存储介质 Download PDF

Info

Publication number
WO2019119319A1
WO2019119319A1 PCT/CN2017/117570 CN2017117570W WO2019119319A1 WO 2019119319 A1 WO2019119319 A1 WO 2019119319A1 CN 2017117570 W CN2017117570 W CN 2017117570W WO 2019119319 A1 WO2019119319 A1 WO 2019119319A1
Authority
WO
WIPO (PCT)
Prior art keywords
parking lot
prediction
neural network
time point
wavelet neural
Prior art date
Application number
PCT/CN2017/117570
Other languages
English (en)
French (fr)
Inventor
彭磊
李慧云
房祥彦
Original Assignee
中国科学院深圳先进技术研究院
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by 中国科学院深圳先进技术研究院 filed Critical 中国科学院深圳先进技术研究院
Priority to PCT/CN2017/117570 priority Critical patent/WO2019119319A1/zh
Publication of WO2019119319A1 publication Critical patent/WO2019119319A1/zh

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q10/00—Administration; Management
    • G06Q10/04—Forecasting or optimisation specially adapted for administrative or management purposes, e.g. linear programming or "cutting stock problem"
    • G—PHYSICS
    • G08—SIGNALLING
    • G08G—TRAFFIC CONTROL SYSTEMS
    • G08G1/00—Traffic control systems for road vehicles
    • G08G1/14—Traffic control systems for road vehicles indicating individual free spaces in parking areas

Definitions

  • the invention belongs to the technical field of information, and in particular relates to a parking lot berth occupancy prediction method, device, device and storage medium.
  • PGIS Parking Guidance and Information
  • the field's idle berth prediction which is essentially a time series based prediction problem, on which autoregressive integral moving average models (ARIMA), wavelet neural networks (WNN) or long- and short-term memory networks (LSTM) can Achieve more accurate short-term predictions.
  • ARIMA autoregressive integral moving average models
  • WNN wavelet neural networks
  • LSTM long- and short-term memory networks
  • ARIMA autoregressive integral moving average models
  • WNN wavelet neural networks
  • LSTM long- and short-term memory networks
  • the method of analysis and prediction based on the historical data of the parking lot needs to be combined with the medium and long-term prediction technology.
  • the current method for medium and long-term prediction is mainly the largest Lyapunov exponential method, the largest Li
  • the essence of the Yapnov exponential method is the detection of chaos.
  • the prediction period continues to increase, there will be a large deviation, and the method needs to reconstruct the phase space for each prediction, and the computational complexity is high. .
  • the object of the present invention is to provide a parking lot berth occupancy prediction method, device, device and storage medium, which aims to solve the problem that the calculation complexity of the medium and long-term prediction of the parking lot idle berth in the prior art is high and prone to large deviation.
  • the present invention provides a parking lot berth occupancy prediction method, the method comprising the following steps:
  • the hybrid prediction model Predicting the number of free berths of the parking lot at the preset time point by using a pre-trained hybrid prediction model and the number of free berths at the last recorded time point, the hybrid prediction model adopting a preset wavelet neural network
  • the network and the preset non-stationary stochastic process are combined with training;
  • the berth occupancy rate of the parking lot at the preset time point is obtained and output.
  • the present invention provides a parking lot berth occupancy prediction device, the device comprising:
  • a berth number obtaining unit configured to acquire, from the historical data of the parking lot, the number of free berths of the parking lot at the last recorded time point when receiving the berth prediction request of the parking lot at a preset time point;
  • a berth prediction unit configured to predict, by using a pre-trained hybrid prediction model and the number of free berths at the last recorded time point, the number of free berths of the parking lot at the preset time point, the hybrid prediction model Obtained through a combination of a preset wavelet neural network and a preset non-stationary stochastic process;
  • An occupancy rate output unit configured to predict, according to the prediction, that the parking lot is at the preset time point The number of free berths is obtained, and the berth occupancy rate of the parking lot at the preset time point is obtained and output.
  • the present invention also provides a computing device including a memory, a processor, and a computer program stored in the memory and operable on the processor, the processor implementing the computer program The steps described in the parking lot berth occupancy prediction method described above.
  • the present invention provides a computer readable storage medium storing a computer program, when the computer program is executed by a processor, implementing the parking lot berth occupancy prediction method as described above A step of.
  • the present invention When receiving the berth prediction request of the parking lot at a preset time point, the present invention obtains the number of free berths of the parking lot at the last recorded time point from the historical data of the parking lot, and passes the trained mixed prediction model and the number of the free berths.
  • the hybrid prediction model combined with wavelet neural network and non-stationary stochastic process realizes the medium and long-term prediction of parking lot berth occupancy rate, effectively reduces the computational complexity of medium and long-term prediction of parking lot berth occupancy rate, and effectively improves the parking lot.
  • the accuracy of the mid- and long-term prediction of berth occupancy increases the efficiency of berth occupancy prediction.
  • FIG. 1 is a flowchart of implementing a parking lot berth occupancy prediction method according to Embodiment 1 of the present invention
  • FIG. 2 is a flowchart of implementing a hybrid prediction model training process in a parking lot berth occupancy prediction method according to Embodiment 1 of the present invention
  • FIG. 3 is a diagram showing an example of a structure of a wavelet neural network in a parking lot berth occupancy prediction method according to Embodiment 1 of the present invention
  • FIG. 4 is a flowchart of implementing a parking lot berth occupancy prediction method according to Embodiment 2 of the present invention.
  • FIG. 5 is a schematic structural diagram of a parking lot berth occupancy prediction apparatus according to Embodiment 3 of the present invention.
  • FIG. 6 is a schematic diagram of a preferred structure of a parking lot berth occupancy prediction apparatus according to Embodiment 3 of the present invention.
  • FIG. 7 is a schematic structural diagram of a computing device according to Embodiment 4 of the present invention.
  • Embodiment 1 is a diagrammatic representation of Embodiment 1:
  • FIG. 1 is a flowchart showing an implementation process of a parking lot berth occupancy prediction method according to Embodiment 1 of the present invention. For convenience of description, only parts related to the embodiment of the present invention are shown, which are as follows:
  • step S101 when the berth prediction request of the parking lot at the preset time point is received, the number of free berths of the parking lot at the last recorded time point is obtained from the historical data of the parking lot.
  • Embodiments of the present invention are applicable to computing devices such as in-vehicle devices or parking management servers, and in particular, to parking guidance systems or platforms on such devices.
  • the parking lot can be sent at the preset time point berth prediction request, and the preset time point is determined by the user according to his own parking time (or the time of arrival at the parking lot) ) Make settings.
  • the berth prediction request of the parking lot at the preset time point the historical data of the parking lot is acquired, and the number of free berths of the parking lot at the last recorded time is obtained from the historical data.
  • the historical data of the parking lot is the berth occupancy information of the parking lot in the past period of time, and there is a long time interval for updating the historical data.
  • the last recorded time point is the time when the parking lot occupant occupancy information is recorded in the historical data
  • the free berth is the parking space not occupied by the vehicle or other items.
  • the historical data of the parking lot may also be the number of vehicles entering and leaving the parking lot in the past period of time and the recording time points of these data.
  • step S102 the pre-trained hybrid prediction model and the number of free berths at the last recorded time point are used to predict the number of free berths of the parking lot at the preset time point, and the hybrid prediction model passes the preset wavelet neural network and the pre-predetermined The non-stationary stochastic process is combined with training.
  • the wavelet neural network is a one-step prediction model, and the prediction accuracy is high in the short-term prediction, and the prediction value jump is not large, but the prediction time is longer in the medium- and long-term prediction. The longer the length, the more likely the deviation of the predicted value will occur.
  • the stationary stochastic process is suitable for medium and long-term prediction. It can suppress the large deviation of the predicted value to a great extent, and the calculation cost is small, but it is prone to a large jump of the predicted value.
  • the hybrid prediction model obtained by the combination of wavelet neural network and stationary stochastic process combines the advantages of wavelet neural network and stationary stochastic process, which makes the wavelet neural network and stationary stochastic process complement each other and can effectively reduce the medium and long-term prediction. Calculate complexity and improve the accuracy of mid- and long-term forecasts.
  • the time interval from the last recording time point to the preset time point may be divided into a plurality of equally spaced time slices, and the number of time slices, the last recorded time point, and the last recorded time point are idle.
  • the number of berths is input into the hybrid prediction model, and the number of free berths for each time slice from the last recorded time point to the preset time point is obtained, thereby obtaining the number of free berths of the parking lot at the predicted time point.
  • step S103 according to the predicted number of free berths of the parking lot at the preset time point, the berth occupancy rate of the parking lot at the preset time point is obtained and output.
  • the berth occupancy rate of the parking lot at the preset time point can be calculated, and the berth occupied by the output parking lot at the preset time point is occupied. Rate to help users quickly park.
  • the training process of the hybrid prediction model can be implemented by the following steps:
  • step S201 a wavelet neural network is constructed, and the wavelet neural network is trained by the historical training data of the collected parking lot to obtain a trained wavelet neural network.
  • the parking lot is collected for historical data of training, and historical data for training is referred to as historical training data for convenience of distinction.
  • historical training data is collected on a weekly basis, thereby effectively improving the accuracy of subsequent berth occupancy prediction.
  • the mathematical model of the wavelet neural network can be expressed as:
  • E j (t) represents the number of free berths of the parking lot j at the preset time point t
  • f(x) is an analytic function of the wavelet neural network.
  • the Morlet mother wavelet can be set as the excitation function
  • a k is the expansion factor
  • b k is the translation factor
  • ⁇ k is the output weight of the wavelet neural network
  • ⁇ ik is the input weight of the wavelet neural network. value.
  • Figure 3 is a diagram showing an example of the structure of a wavelet neural network.
  • W 11 ... W nm is the input weight of the wavelet neural network
  • W 1 ... W m is the output weight of the wavelet neural network.
  • the training wavelet neural network adjusts a k , b k , ⁇ k and ⁇ ik according to the prediction result of the wavelet neural network. Specifically, the number of free berths of the parking lot at the preset time point t is predicted by the wavelet neural network, and the predicted number of free berths is compared with the number of free berths at the preset time point t in the historical training data to obtain the wavelet neural network.
  • Network prediction error E jn (t) is the number of free berths of parking lot j at time t in the historical training data.
  • the a k , b k , ⁇ k and ⁇ ik in the wavelet neural network are adjusted by the preset gradient descent method until the wavelet neural network The prediction error of the network is less than the preset error threshold.
  • step S202 the time series corresponding to the historical training data is divided into a plurality of equidistant time slices, and the distribution type and distribution parameters obeyed by the change in the number of parking vacancies between adjacent time slices are determined.
  • the time series corresponding to the historical training data is divided into a plurality of equally spaced time slices ⁇ t i
  • i 1...n ⁇ , where n is the number of time slices, and each time is obtained from the historical training data.
  • the occupancy of the berth in the parking lot depends on the arrival rate and the departure rate of the vehicle, and the arrival of the vehicle is generally considered to be subject to the Poisson distribution.
  • the distribution of the number of occupied berths in each time slice is The limit of the Poisson distribution, ie the normal distribution, can be expressed as Where ⁇ i .
  • the distribution parameter ⁇ i of each x i can be solved by the maximum likelihood method (MLE),
  • change in the number of idle berth is a non-stationary Gaussian process.
  • cov(x i , x i+1 ) is a covariance operator of x i+1 and x i .
  • step S203 a medium-and-long-term prediction function is constructed according to the distribution type and distribution parameters obeyed by the change in the number of parking lot vacancies between adjacent time slices.
  • the medium and long term prediction function can be expressed as Where E j (t 0 ) represents the number of free berths of parking lot j at time t 0 , and n is the number of time slices between t and t 0 .
  • step S204 the trained wavelet neural network and the medium-long-term prediction function are weighted and combined to obtain a hybrid prediction model, and the weight parameters corresponding to the wavelet neural network and the medium-long-term prediction function in the hybrid prediction model are determined according to the historical training data.
  • the trained wavelet neural network and the medium- and long-term prediction function are weighted and combined to obtain a mixture.
  • the predictive model, the hybrid predictive model can be expressed as:
  • E j (t) a*g(t 0 ,n)+b*f[g(t 0 ,n-1)], where a and b are the weights corresponding to the wavelet neural network and the medium- and long-term prediction function, respectively.
  • the parameters can be determined according to the least squares method and the historical training data.
  • the mixed prediction model is first predicted by the medium and long-term prediction function, and then the direct pre-order result of the prediction result obtained by the medium- and long-term prediction function is input to the wavelet neural network for one-step prediction, and finally The prediction results of the medium- and long-term prediction function and the prediction results of the wavelet neural network are linearly weighted, and the final prediction results have some randomness and no large deviation.
  • the hybrid prediction model constructed by the wavelet neural network and the non-stationary stochastic process is used to predict the number of free berths of the parking lot at a preset time point, and the long-term prediction of the berth occupancy rate is simultaneously solved.
  • the problem that the prediction value of the wavelet neural network deviates greatly in the medium and long-term prediction also solves the problem that the prediction value jumps greatly during the non-stationary stochastic process, and avoids the calculation using the maximum Lyapunov exponent method. Complexity, which effectively increases the mooring The efficiency and accuracy of mid- and long-term forecasts of bit occupancy.
  • Embodiment 2 is a diagrammatic representation of Embodiment 1:
  • FIG. 4 is a flowchart showing an implementation process of a parking lot berth occupancy prediction method according to Embodiment 2 of the present invention. For convenience of description, only parts related to the embodiment of the present invention are shown, which are as follows:
  • step S401 when the berth prediction request of the parking lot at the preset time point is received, the number of free berths of the parking lot at the last recorded time point is obtained from the historical data of the parking lot.
  • the historical data of the parking lot is acquired, and the number of free berths of the parking lot at the last recorded time is obtained from the historical data.
  • the historical data of the parking lot is the berth occupancy information of the parking lot in the past period of time, and there is a long time interval for updating the historical data.
  • the last recorded time point is the time at which the parking berth occupancy information is last recorded in the historical data.
  • step S402 the pre-trained hybrid prediction model and the number of free berths at the last recorded time point are used to predict the number of free berths of the parking lot at the preset time point, and the hybrid prediction model passes the preset wavelet neural network and the pre-predetermined The non-stationary stochastic process is combined with training.
  • the time interval from the last recording time point to the preset time point may be divided into a plurality of equally spaced time slices, and the number of time slices, the last recorded time point, and the last recorded time point are idle.
  • the number of berths is input into the hybrid prediction model, and the number of free berths for each time slice from the last recorded time point to the preset time point is obtained, thereby obtaining the number of free berths of the parking lot at the predicted time point.
  • step S403 it is detected whether the prediction result of the hybrid detection model is chaotic by a preset maximum Lyapunov exponent method.
  • the largest Lyapunov exponent method uses the calculated maximum Lyapunov exponent as one of the main basis for identifying chaotic characteristics.
  • the maximum Lyapunov exponent method is used to detect whether the prediction result of the hybrid detection model is chaotic, the mutual information method and the pseudo-domain method are used to determine the sequence. The time delay and the embedding dimension are used to reconstruct the phase space of the parking space data.
  • the largest Lyapunov exponent is obtained by the small data volume method. Specifically, the formula for calculating the maximum Lyapunov exponent can be expressed as:
  • V(t) Lyapunov[E(t 0 ), E(t 1 ), ....E(t)], Lyapunov is the largest Lyapunov exponent operator, E(t 0 ), E(t 1 ) , . . . E(t) is the number of berths predicted from time t 0 to time t.
  • V(t)>0 it is considered that the prediction result of the hybrid detection model is chaotic, and step S404 is performed.
  • V(t) ⁇ 0 the prediction result of the hybrid detection model is not chaotic, and the process proceeds to step S402, and continues. The prediction is performed by the chaotic prediction model until the prediction result of the chaotic prediction model is chaotic.
  • step S404 according to the predicted number of free berths of the parking lot at the preset time point, the berth occupancy rate of the parking lot at the preset time point is obtained and output.
  • the berth occupancy rate of the parking lot at the preset time point can be calculated, and the berth occupied by the output parking lot at the preset time point is occupied. Rate to help users quickly park.
  • the hybrid prediction model constructed by the wavelet neural network and the non-stationary stochastic process is used to predict the number of free berths of the parking lot at a preset time point, and then the prediction result is performed by the maximum Lyapunov exponential method.
  • Chaotic detection while realizing long-term prediction of berth occupancy rate, solves the problem that the prediction value of wavelet neural network deviates greatly in medium and long-term prediction, and also solves the problem that non-stationary stochastic process has large jump value during prediction.
  • the problem while avoiding the computational complexity of the prediction using the maximum Lyapunov exponent method, effectively improves the efficiency and accuracy of the mid- and long-term prediction of berth occupancy.
  • Embodiment 3 is a diagrammatic representation of Embodiment 3
  • FIG. 5 is a diagram showing the structure of a parking lot berth occupancy prediction apparatus according to Embodiment 3 of the present invention. For convenience of description, only parts related to the embodiment of the present invention are shown, including:
  • the request receiving unit 51 is configured to obtain the number of free berths of the parking lot at the last recorded time point from the historical data of the parking lot when receiving the berth prediction request of the parking lot at the preset time point.
  • the berth prediction request of the parking lot at the preset time point may be sent, and the preset time point is determined by the user according to the user. Set the parking time (or the time to arrive at the parking lot).
  • the historical data of the parking lot is acquired, and the number of free berths of the parking lot at the last recorded time is obtained from the historical data.
  • the historical data of the parking lot is the berth occupancy information of the parking lot in the past period of time, and there is a long time interval for updating the historical data.
  • the last recorded time point is the time when the parking lot occupant occupancy information is recorded in the historical data
  • the free berth is the parking space not occupied by the vehicle or other items.
  • the historical data of the parking lot may also be the number of vehicles entering and leaving the parking lot in the past period of time and the recording time points of these data.
  • the berth prediction unit 52 is configured to predict the number of free berths of the parking lot at a preset time point by using the pre-trained hybrid prediction model and the number of idle berths at the last recorded time point, and the hybrid prediction model passes the preset wavelet neural network. It is combined with a preset non-stationary random process.
  • the wavelet neural network is a one-step prediction model, and the prediction accuracy is high in the short-term prediction, and the prediction value jump is not large, but the prediction time is longer in the medium- and long-term prediction. The longer the length, the more likely the deviation of the predicted value will occur.
  • the stationary stochastic process is suitable for medium and long-term prediction. It can suppress the large deviation of the predicted value to a great extent, and the calculation cost is small, but it is prone to a large jump of the predicted value.
  • the hybrid prediction model obtained by the combination of wavelet neural network and stationary stochastic process combines the advantages of wavelet neural network and stationary stochastic process, which makes the wavelet neural network and stationary stochastic process complement each other and can effectively reduce the medium and long-term prediction. Calculate complexity and improve the accuracy of mid- and long-term forecasts.
  • the time interval from the last recording time point to the preset time point may be divided into a plurality of equally spaced time slices, and the number of time slices, the last recorded time point, and the last recorded time point are idle.
  • the number of berths is input into the hybrid prediction model, and the number of free berths for each time slice from the last recorded time point to the preset time point is obtained, thereby obtaining the number of free berths of the parking lot at the predicted time point.
  • the occupancy rate output unit 53 is configured to obtain the berth occupancy rate of the parking lot at the preset time point according to the predicted number of free berths of the parking lot at the preset time point, and output the berth.
  • the berth occupancy rate of the parking lot at the preset time point can be calculated, and the berth occupied by the output parking lot at the preset time point is occupied. Rate to help users quickly park.
  • the parking lot berth occupancy prediction device further includes:
  • the network training unit 61 is configured to construct a wavelet neural network, and train the wavelet neural network through the collected historical training data of the parking lot to obtain a trained wavelet neural network.
  • historical training data is collected.
  • historical training data is collected on a weekly basis, thereby effectively improving the prediction of subsequent berth occupancy. Accuracy.
  • the mathematical model of the wavelet neural network can be expressed as:
  • E j (t) represents the number of free berths of the parking lot j at the preset time point t
  • f(x) is an analytic function of the wavelet neural network.
  • the Morlet mother wavelet can be set as the excitation function
  • a k is the expansion factor
  • b k is the translation factor
  • ⁇ k is the output weight of the wavelet neural network
  • ⁇ ik is the input weight of the wavelet neural network. value.
  • Figure 3 is a diagram showing an example of the structure of a wavelet neural network.
  • W 11 ... W nm is the input weight of the wavelet neural network
  • W 1 ... W m is the output weight of the wavelet neural network.
  • the training wavelet neural network adjusts a k , b k , ⁇ k and ⁇ ik according to the prediction result of the wavelet neural network. Specifically, the number of free berths of the parking lot at the preset time point t is predicted by the wavelet neural network, and the predicted number of free berths is compared with the number of free berths at the preset time point t in the historical training data to obtain the wavelet neural network.
  • Network prediction error E jn (t) is the number of free berths of parking lot j at time t in the historical training data.
  • the a k , b k , ⁇ k and ⁇ ik in the wavelet neural network are adjusted by the preset gradient descent method until the wavelet neural network The prediction error of the network is less than the preset error threshold.
  • the data analyzing unit 62 is configured to divide the time series corresponding to the historical training data into a plurality of equidistant time slices, and determine a distribution type and a distribution parameter obeyed by the change in the number of free parking spaces in the parking lot between adjacent time slices.
  • the time series corresponding to the historical training data is divided into a plurality of equally spaced time slices ⁇ t i
  • i 1...n ⁇ , where n is the number of time slices, and each time is obtained from the historical training data.
  • the occupancy of the berth in the parking lot depends on the arrival rate and the departure rate of the vehicle, and the arrival of the vehicle is generally considered to be subject to the Poisson distribution.
  • the distribution of the number of occupied berths in each time slice is The limit of the Poisson distribution, ie the normal distribution, can be expressed as Where ⁇ i .
  • the distribution parameter ⁇ i of each x i can be solved by the maximum likelihood method (MLE),
  • the change in the number of free berths on the time axis of the historical training data is a non-stationary Gaussian process.
  • Where cov(x i , x i+1 ) is a covariance operator of x i+1 and x i .
  • the medium-and-long-term prediction construction unit 63 is configured to construct a medium-and-long-term prediction function according to the distribution type and distribution parameters obeyed by the change in the number of parking spaces in the parking lot between adjacent time slices.
  • the medium and long term prediction function can be expressed as Where E j (t 0 ) represents the number of free berths of parking lot j at time t 0 , and n is the number of time slices between t and t 0 .
  • the hybrid model generating unit 64 is configured to perform weighted combination of the trained wavelet neural network and the medium-long-term prediction function to obtain a hybrid prediction model, and determine, according to the historical training data, the corresponding functions of the wavelet neural network and the medium- and long-term prediction functions in the hybrid prediction model. Value parameter.
  • the trained wavelet neural network and the medium- and long-term prediction function are weighted and combined to obtain a mixture.
  • the predictive model, the hybrid predictive model can be expressed as:
  • E j (t) a*g(t 0 ,n)+b*f[g(t 0 ,n-1)], where a and b are the weights corresponding to the wavelet neural network and the medium- and long-term prediction function, respectively.
  • the parameters can be determined according to the least squares method and the historical training data.
  • the mixed prediction model is first predicted by the medium and long-term prediction function, and then the direct pre-order result of the prediction result obtained by the medium- and long-term prediction function is input to the wavelet neural network for one-step prediction, and finally The prediction results of the medium- and long-term prediction function and the prediction results of the wavelet neural network are linearly weighted, and the final prediction results have some randomness and no large deviation.
  • the parking lot berth occupancy prediction device further comprises:
  • the chaos detection unit is configured to detect the prediction result of the hybrid detection model by using a preset maximum Lyapunov exponential method to determine whether the prediction result of the hybrid prediction model is chaotic;
  • the prediction jump unit is configured to perform an operation of predicting the number of free berths of the parking lot at the preset time point by the berth prediction unit 52 when the prediction result of the hybrid prediction model is not chaotic.
  • the mutual information method and the pseudo-domain method are first used to determine the time delay and the embedded dimension of the sequence, and phase space reconstruction is performed on the parking space data.
  • the largest Lyapunov exponent is obtained by the small data method. Specifically, the formula for calculating the maximum Lyapunov exponent can be expressed as:
  • V(t) Lyapunov[E(t 0 ), E(t 1 ), ....E(t)], Lyapunov is the largest Lyapunov exponent operator, E(t 0 ), E(t 1 ) , . . . E(t) is the number of berths predicted from time t 0 to time t.
  • the occupancy rate output unit 53 performs an operation of obtaining the berth occupancy rate of the parking lot at the preset time point and outputting, when V(t) ⁇ 0
  • the berth prediction unit 52 performs an operation of predicting the number of free berths of the parking lot at a preset time point.
  • the hybrid prediction model constructed by the wavelet neural network and the non-stationary stochastic process is used to predict the number of free berths of the parking lot at a preset time point, and the long-term prediction of the berth occupancy rate is simultaneously solved.
  • the problem that the prediction value of the wavelet neural network deviates greatly in the medium and long-term prediction also solves the problem that the prediction value jumps greatly during the non-stationary stochastic process, and avoids the calculation using the maximum Lyapunov exponent method. Complexity, which effectively increases the mooring The efficiency and accuracy of mid- and long-term forecasts of bit occupancy.
  • each unit of the parking lot berth occupancy prediction device may be implemented by a corresponding hardware or software unit, and each unit may be an independent software and hardware unit, or may be integrated into a soft and hardware unit. To limit the invention.
  • Embodiment 4 is a diagrammatic representation of Embodiment 4:
  • FIG. 7 shows the structure of a computing device according to Embodiment 4 of the present invention. For the convenience of description, only parts related to the embodiment of the present invention are shown.
  • the computing device 7 of an embodiment of the present invention includes a processor 70, a memory 71, and a computer program 72 stored in the memory 71 and executable on the processor 70.
  • the processor 70 executes the computer program 72 to implement the steps in the various method embodiments described above, such as steps S101 through S103 shown in FIG.
  • processor 70 when executing computer program 72, implements the functions of the various units of the various apparatus embodiments described above, such as the functions of units 51 through 53 of FIG.
  • the hybrid prediction model constructed by the wavelet neural network and the non-stationary stochastic process is used to predict the number of free berths of the parking lot at a preset time point, and the long-term prediction of the berth occupancy rate is simultaneously solved.
  • the problem that the prediction value of the wavelet neural network deviates greatly in the medium and long-term prediction also solves the problem that the prediction value jumps greatly during the non-stationary stochastic process, and avoids the calculation using the maximum Lyapunov exponent method. Complexity, which effectively improves the efficiency and accuracy of mid- and long-term prediction of berth occupancy.
  • Embodiment 5 is a diagrammatic representation of Embodiment 5:
  • a computer readable storage medium storing a computer program, the computer program being executed by a processor to implement the steps in the foregoing method embodiments, for example, FIG. Steps S101 to S103 are shown.
  • the computer program when executed by the processor, implements the functions of the various units in the various apparatus embodiments described above, such as the functions of units 51 through 53 shown in FIG.
  • the hybrid prediction model constructed by the wavelet neural network and the non-stationary stochastic process is used to predict the number of free berths of the parking lot at a preset time point, and the long-term prediction of the berth occupancy rate is simultaneously solved.
  • the problem that the prediction value of the wavelet neural network deviates greatly in the medium and long-term prediction also solves the problem that the prediction value jumps greatly during the non-stationary stochastic process, and avoids the calculation using the maximum Lyapunov exponent method. Complexity, which effectively improves the efficiency and accuracy of mid- and long-term prediction of berth occupancy.
  • the computer readable storage medium of the embodiments of the present invention may include any entity or device capable of carrying computer program code, a recording medium such as a ROM/RAM, a magnetic disk, an optical disk, a flash memory, or the like.

Landscapes

  • Business, Economics & Management (AREA)
  • Engineering & Computer Science (AREA)
  • Human Resources & Organizations (AREA)
  • General Physics & Mathematics (AREA)
  • Physics & Mathematics (AREA)
  • Economics (AREA)
  • Strategic Management (AREA)
  • Development Economics (AREA)
  • Game Theory and Decision Science (AREA)
  • Entrepreneurship & Innovation (AREA)
  • Marketing (AREA)
  • Operations Research (AREA)
  • Quality & Reliability (AREA)
  • Tourism & Hospitality (AREA)
  • General Business, Economics & Management (AREA)
  • Theoretical Computer Science (AREA)
  • Management, Administration, Business Operations System, And Electronic Commerce (AREA)
  • Traffic Control Systems (AREA)

Abstract

本发明适用信息技术领域，提供了一种停车场泊位占用率预测方法、装置、设备及存储介质，该方法包括：当接收到停车场在预设时间点的泊位预测请求时，从停车场的历史数据中获取停车场在最后记录时间点的空闲泊位数量，通过预先训练好的混合预测模型和最后记录时间点的空闲泊位数量，对停车场在预设时间点的空闲泊位数量进行预测，根据预测得到的、停车场在预设时间点的空闲泊位数量，获得停车场在预设时间点的泊位占用率并输出，其中，混合预测模型通过小波神经网络和非平稳随机过程结合训练得到，从而实现泊位占用率的中长期预测，有效地降低了泊位占用率预测的计算复杂度，有效地提高了泊位占用率预测的准确度，进而提高了泊位占用率预测的效率。

Description

停车场泊位占用率预测方法、装置、设备及存储介质 技术领域
本发明属于信息技术领域，尤其涉及一种停车场泊位占用率预测方法、装置、设备及存储介质。
背景技术
由于车辆数目的激增、以及国内城市早期规划未长远考虑到车辆停放问题，国内大中城市热点区域所提供的停车位远远少于进入的车辆，使得车辆在寻找停车位的过程中，花费大量时间、浪费不必要的能源，甚至引发交通堵塞，在短期内增加这些区域的停车位供应比较困难，所以提高这些区域内的停车位利用率就变得非常重要。
提高停车位利用率需要通过将车位信息实时推送给有需要的车辆来帮助车辆快速找到车位，即研究人员提出的停车诱导系统(PGIS，Parking Guidance and Information)。PGIS在对车辆进行诱导时，车辆距离停车场还有一定距离，PGIS需要估算车辆抵达停车场的时间点、以及这个时间点上停车场内空闲泊位的数量，因此PGIS需要实现未来一段时间内停车场的空闲泊位预测，这本质上是一个基于时间序列的预测问题，在这类预测问题上自回归积分滑动平均模型(ARIMA)、小波神经网络(WNN)或者长短期记忆网络(LSTM)都能实现较为准确的短临预测，然而这些模型需要实时数据的支撑，下一时刻预测的精准度同刚过去的几个连续时间步的数据关联性很高，停车场产权分散，不同停车场间的设备难以互联，且缺乏统一的城市级泊位监测平台，使得大量的停车场实时数据难以获得。
此外，基于停车场的历史数据进行分析和预测的方法需要结合中长期预测技术实现，目前用于中长期预测的方法主要为最大李雅普诺夫指数法，最大李 雅普诺夫指数法的本质是对混沌性的检测，在预测周期持续增长时，会出现很大的偏差，而且该方法每次预测时均需对相空间进行重构，计算复杂度较高。
发明内容
本发明的目的在于提供一种停车场泊位占用率预测方法、装置、设备及存储介质，旨在解决现有技术中停车场空闲泊位中长期预测的计算复杂度较高、容易出现较大偏差，导致停车场空闲泊位中长期预测的效率不高、准确度不高的问题。
一方面，本发明提供了一种停车场泊位占用率预测方法，所述方法包括下述步骤：
当接收到停车场在预设时间点的泊位预测请求时，从所述停车场的历史数据中获取所述停车场在最后记录时间点的空闲泊位数量；
通过预先训练好的混合预测模型和所述最后记录时间点的空闲泊位数量，对所述停车场在所述预设时间点的空闲泊位数量进行预测，所述混合预测模型通过预设的小波神经网络和预设的非平稳随机过程结合训练得到；
根据预测得到的、所述停车场在所述预设时间点的空闲泊位数量，获得所述停车场在所述预设时间点的泊位占用率并输出。
另一方面，本发明提供了一种停车场泊位占用率预测装置，所述装置包括：
泊位数量获取单元，用于当接收到停车场在预设时间点的泊位预测请求时，从所述停车场的历史数据中获取所述停车场在最后记录时间点的空闲泊位数量；
泊位预测单元，用于通过预先训练好的混合预测模型和所述最后记录时间点的空闲泊位数量，对所述停车场在所述预设时间点的空闲泊位数量进行预测，所述混合预测模型通过预设的小波神经网络和预设的非平稳随机过程结合训练得到；以及
占用率输出单元，用于根据预测得到的、所述停车场在所述预设时间点的 空闲泊位数量，获得所述停车场在所述预设时间点的泊位占用率并输出。
另一方面，本发明还提供了一种计算设备，包括存储器、处理器以及存储在所述存储器中并可在所述处理器上运行的计算机程序，所述处理器执行所述计算机程序时实现如上述停车场泊位占用率预测方法所述的步骤。
另一方面，本发明还提供了一种计算机可读存储介质，所述计算机可读存储介质存储有计算机程序，所述计算机程序被处理器执行时实现如上述停车场泊位占用率预测方法所述的步骤。
本发明在接收到停车场在预设时间点的泊位预测请求时，从停车场的历史数据中获取停车场在最后记录时间点的空闲泊位数量，通过训练好的混合预测模型和该空闲泊位数量，对停车场在预设时间点的空闲泊位数量进行预测，根据预测得到的、停车场在预设时间点的空闲泊位数量，获得停车场在预设时间点的泊位占用率并输出，从而通过由小波神经网络和非平稳随机过程结合的混合预测模型，实现了停车场泊位占用率的中长期预测，有效地降低了停车场泊位占用率中长期预测的计算复杂度，有效地提高了停车场泊位占用率中长期预测的准确度，进而提高了泊位占用率预测的效率。
附图说明
图1是本发明实施例一提供的停车场泊位占用率预测方法的实现流程图；
图2是本发明实施例一提供的停车场泊位占用率预测方法中混合预测模型训练过程的实现流程图；
图3是本发明实施例一提供的停车场泊位占用率预测方法中小波神经网络的结构示例图；
图4是本发明实施例二提供的停车场泊位占用率预测方法的实现流程图；
图5是本发明实施例三提供的停车场泊位占用率预测装置的结构示意图；
图6是本发明实施例三提供的停车场泊位占用率预测装置的优选结构示意图；以及
图7是本发明实施例四提供的计算设备的结构示意图。
具体实施方式
为了使本发明的目的、技术方案及优点更加清楚明白，以下结合附图及实施例，对本发明进行进一步详细说明。应当理解，此处所描述的具体实施例仅仅用以解释本发明，并不用于限定本发明。
以下结合具体实施例对本发明的具体实现进行详细描述：
实施例一：
图1示出了本发明实施例一提供的停车场泊位占用率预测方法的实现流程，为了便于说明，仅示出了与本发明实施例相关的部分，详述如下：
在步骤S101中，当接收到停车场在预设时间点的泊位预测请求时，从停车场的历史数据中获取停车场在最后记录时间点的空闲泊位数量。
本发明实施例适用于车载设备或停车管理服务器等计算设备，具体地，适用于这些设备上的停车诱导系统或平台。当用户需要得知停车场在未来某个时间的车位数量时，可发送停车场在预设时间点的泊位预测请求时，预设时间点由用户根据自身的停车时间(或到达停车场的时间)进行设置。在接收到停车场在预设时间点的泊位预测请求时，获取该停车场的历史数据，并从历史数据中获取停车场在最后记录时刻的空闲泊位数量。目前缺乏统一的城市级泊位检测系统、且不同停车场间的设备难以互联，因此停车场的历史数据为停车场过去一段时间内的泊位占用信息，历史数据的更新存在较长的时间间隔，无法实时更新，最后记录时间点为历史数据中最后记录停车场泊位占用信息的时刻，空闲泊位为未被车辆或其它物品占用的停车位。此外，停车场的历史数据还可为过去一段时间内停车场进出车辆的数目和这些数据的记录时间点。
在步骤S102中，通过预先训练好的混合预测模型和最后记录时间点的空闲泊位数量，对停车场在预设时间点的空闲泊位数量进行预测，混合预测模型通过预设的小波神经网络和预设的非平稳随机过程结合训练得到。
在本发明实施例中，小波神经网络为一步预测模型，在短临预测时预测精准度较高，不会出现预测值跳变较大的情况，但用于中长期预测时随着预测时间越长越容易出现预测值的大幅偏离。平稳随机过程适用于中长期预测，在极大程度上能够抑制预测值的大幅偏离，计算开销小，但容易出现预测值跳变较大的情况。因此，由小波神经网络和平稳随机过程结合得到的混合预测模型，结合了小波神经网络和平稳随机过程的优点，使得小波神经网络和平稳随机过程进行优劣互补，能够有效地降低中长期预测的计算复杂度，提高中长期预测的精准程度。
在本发明实施例中，可将最后记录时间点到预设时间点之间的时间间隔划分为多个等距的时间片，将时间片的数量、最后记录时间点、最后记录时间点的空闲泊位数量输入混合预测模型，得到停车场在从最后记录时间点到预设时间点间每个时间片的空闲泊位数量，进而得到停车场在预测时间点的空闲泊位数量。
在步骤S103中，根据预测得到的、停车场在预设时间点的空闲泊位数量，获得停车场在预设时间点的泊位占用率并输出。
在本发明实施例中，根据预测得到的、停车场在预设时间点的空闲泊位数量，可计算得到停车场在预设时间点的泊位占用率，输出停车场在预设时间点的泊位占用率，以帮助用户快速停车。
优选地，如图2所示，混合预测模型的训练过程可通过下述步骤实现：
在步骤S201中，构建小波神经网络，通过采集到的停车场的历史训练数据对小波神经网络进行训练，获得训练好的小波神经网络。
在本发明实施例中，采集停车场用于训练的历史数据，为了便于区分，将用于训练的历史数据称为历史训练数据。优选地，由于人们往往以周为单位进行生活和工作的安排，以周为单位进行历史训练数据的采集，从而有效地提高后续泊位占用率预测的准确度。
在本发明实施例中，小波神经网络的数学模型可表示为：
Figure PCTCN2017117570-appb-000001
其中，Ej(t)表示停车场j在预设时间点t时的空闲泊位数量，f(x)为小波神经网络的解析函数，
Figure PCTCN2017117570-appb-000002
为小波神经网络的激励函数，可将Morlet母小波设置为激励函数，ak为伸缩因子，bk为平移因子，ωk为小波神经网络的输出权值，ωik为小波神经网络的输入权值。图3为小波神经网络的结构示例图，图中W11...Wnm为小波神经网络的输入权值，图中W1...Wm为小波神经网络的输出权值。
在本发明实施例中，训练小波神经网络即根据小波神经网络的预测结果对ak、bk、ωk和ωik进行调整。具体地，通过小波神经网络预测停车场在预设时间点t时的空闲泊位数量，将预测得到的空闲泊位数量与历史训练数据中预设时间点t时的空闲泊位数量比较，可得到小波神经网络的预测误差
Figure PCTCN2017117570-appb-000003
Ejn(t)为历史训练数据中停车场j在t时刻的空闲泊位数量，通过预设的梯度下降法对小波神经网络中ak、bk、ωk和ωik进行调整，直至小波神经网络的预测误差小于预设误差阈值。
在步骤S202中，将历史训练数据对应的时间序列划分多个等距的时间片，确定相邻时间片之间停车场空闲泊位数量变化所服从的分布类型、分布参数。
在本发明实施例中，将从历史训练数据对应的时间序列划分为多个等距的时间片{Δti|i＝1…n}，n为时间片的数量，从历史训练数据中获取每个时间片上的已占用泊位数量xi。停车场中泊位的占用与否取决于车辆的到达率和驶离率，且车辆的到达一般被认为是服从泊松分布的，所以从长期来看每个时间片内已占用泊位数量的分布为泊松分布的极限，即正态分布，可表示为
Figure PCTCN2017117570-appb-000004
其中，μi、
Figure PCTCN2017117570-appb-000005
为xi正态分布的分布参数，具体地，可通过最大似然法(MLE)求解每个xi的分布参数μi、
Figure PCTCN2017117570-appb-000006
在本发明实施例中，随着Δti不同xi的分布参数也不同，因此在历史训练数 据的时间轴上，空闲泊位数量变化是一个非平稳的高斯过程。在两个相邻时间片ti、ti+1之间空闲泊位数量的变化为Δxi＝xi+1-xi，xi+1和xi被视为两个独立同正态分布的变量，所以Δxi也服从正态分布，且
Figure PCTCN2017117570-appb-000007
其中，cov(xi,xi+1)为xi+1和xi的协方差算子。
在步骤S203中，根据相邻时间片之间停车场空闲泊位数量变化所服从的分布类型、分布参数，构建中长期预测函数。
在本发明实施例中，中长期预测函数可表示为
Figure PCTCN2017117570-appb-000008
其中，Ej(t0)表示停车场j在t0时刻的空闲泊位数量，n为t和t0之间时间片的数量。
在步骤S204中，将训练好的小波神经网络和中长期预测函数进行加权组合，获得混合预测模型，根据历史训练数据确定混合预测模型中小波神经网络、中长期预测函数分别对应的权值参数。
在本发明实施例中，为了将小波神经网络和基于非平稳随机过程构建的中长期预测函数进行优势结合和优劣互补，将训练好的小波神经网络和中长期预测函数进行加权组合，得到混合预测模型，混合预测模型可表示为：
Ej(t)＝a*g(t0,n)+b*f[g(t0,n-1)]，其中，a、b分别为小波神经网络、中长期预测函数对应的权值参数，可根据最小二乘法和历史训练数据进行确定，混合预测模型先由中长期预测函数进行预测，再将中长期预测函数所得预测结果的直接前序结果输入到小波神经网络进行一步预测，最后中长期预测函数的预测结果、小波神经网络的预测结果进行线性加权，最终得到的预测结果既有一定的随机性，又不至于出现较大偏离。
在本发明实施例中，通过小波神经网络、非平稳随机过程结合构建的混合预测模型，对停车场在预设时间点的空闲泊位数量进行预测，在实现泊位占用率中长期预测的同时，解决了小波神经网络在中长期预测时预测值大幅偏离的问题，也解决了非平稳随机过程在预测过程中预测值跳变较大的问题，同时避免了采用最大李雅普诺夫指数法进行预测的计算复杂度，从而有效地提高了泊 位占用率中长期预测的效率和准确率。
实施例二：
图4示出了本发明实施例二提供的停车场泊位占用率预测方法的实现流程，为了便于说明，仅示出了与本发明实施例相关的部分，详述如下：
在步骤S401中，当接收到停车场在预设时间点的泊位预测请求时，从停车场的历史数据中获取停车场在最后记录时间点的空闲泊位数量。
在本发明实施例中，在接收到停车场在预设时间点的泊位预测请求时，获取该停车场的历史数据，并从历史数据中获取停车场在最后记录时刻的空闲泊位数量。目前缺乏统一的城市级泊位检测系统、且不同停车场间的设备难以互联，因此停车场的历史数据为停车场过去一段时间内的泊位占用信息，历史数据的更新存在较长的时间间隔，无法实时更新，最后记录时间点为历史数据中最后记录停车场泊位占用信息的时刻。
在步骤S402中，通过预先训练好的混合预测模型和最后记录时间点的空闲泊位数量，对停车场在预设时间点的空闲泊位数量进行预测，混合预测模型通过预设的小波神经网络和预设的非平稳随机过程结合训练得到。
在本发明实施例中，可将最后记录时间点到预设时间点之间的时间间隔划分为多个等距的时间片，将时间片的数量、最后记录时间点、最后记录时间点的空闲泊位数量输入混合预测模型，得到停车场在从最后记录时间点到预设时间点间每个时间片的空闲泊位数量，进而得到停车场在预测时间点的空闲泊位数量。具体地，混合预测模型的训练过程可参照实施例一相应步骤的详细描述，在此不再赘述。
在步骤S403中，通过预设的最大李雅普诺夫指数法检测混合检测模型的预测结果是否具有混沌性。
在本发明实施例中，最大李雅普诺夫指数法将计算得到的最大李雅普诺夫指数作为识别混沌特性的主要依据之一。最大李雅普诺夫指数法在检测混合检测模型的预测结果是否具有混沌性时，先利用互信息法和伪领域法确定序列的 时间延迟和嵌入维数，在对停车位数据进行相空间重构，最后通过小数据量法求得最大李雅普诺夫指数。具体地，最大李雅普诺夫指数的计算公式可表示为：
V(t)＝Lyapunov[E(t0),E(t1),....E(t)]，Lyapunov为最大李雅普诺夫指数算子，E(t0),E(t1),....E(t)为t0到t时刻所预测的泊位数量。当V(t)>0时认为混合检测模型的预测结果具有混沌性，执行步骤S404，当V(t)≤0时为混合检测模型的预测结果不具有混沌性，跳转至步骤S402，继续通过混沌预测模型进行预测，直至混沌预测模型的预测结果具有混沌性。
在步骤S404中，根据预测得到的、停车场在预设时间点的空闲泊位数量，获得停车场在预设时间点的泊位占用率并输出。
在本发明实施例中，根据预测得到的、停车场在预设时间点的空闲泊位数量，可计算得到停车场在预设时间点的泊位占用率，输出停车场在预设时间点的泊位占用率，以帮助用户快速停车。
在本发明实施例中，通过小波神经网络、非平稳随机过程结合构建的混合预测模型，对停车场在预设时间点的空闲泊位数量进行预测，再通过最大李雅普诺夫指数法对预测结果进行混沌性检测，在实现泊位占用率中长期预测的同时，解决了小波神经网络在中长期预测时预测值大幅偏离的问题，也解决了非平稳随机过程在预测过程中预测值跳变较大的问题，同时避免了采用最大李雅普诺夫指数法进行预测的计算复杂度，从而有效地提高了泊位占用率中长期预测的效率和准确率。
实施例三：
图5示出了本发明实施例三提供的停车场泊位占用率预测装置的结构，为了便于说明，仅示出了与本发明实施例相关的部分，其中包括：
请求接收单元51，用于当接收到停车场在预设时间点的泊位预测请求时，从停车场的历史数据中获取停车场在最后记录时间点的空闲泊位数量。
在本发明实施例中，当用户需要得知停车场在未来某个时间的车位数量时，可发送停车场在预设时间点的泊位预测请求时，预设时间点由用户根据自身的 停车时间(或到达停车场的时间)进行设置。在接收到停车场在预设时间点的泊位预测请求时，获取该停车场的历史数据，并从历史数据中获取停车场在最后记录时刻的空闲泊位数量。目前缺乏统一的城市级泊位检测系统、且不同停车场间的设备难以互联，因此停车场的历史数据为停车场过去一段时间内的泊位占用信息，历史数据的更新存在较长的时间间隔，无法实时更新，最后记录时间点为历史数据中最后记录停车场泊位占用信息的时刻，空闲泊位为未被车辆或其它物品占用的停车位。此外，停车场的历史数据还可为过去一段时间内停车场进出车辆的数目和这些数据的记录时间点。
泊位预测单元52，用于通过预先训练好的混合预测模型和最后记录时间点的空闲泊位数量，对停车场在预设时间点的空闲泊位数量进行预测，混合预测模型通过预设的小波神经网络和预设的非平稳随机过程结合训练得到。
在本发明实施例中，小波神经网络为一步预测模型，在短临预测时预测精准度较高，不会出现预测值跳变较大的情况，但用于中长期预测时随着预测时间越长越容易出现预测值的大幅偏离。平稳随机过程适用于中长期预测，在极大程度上能够抑制预测值的大幅偏离，计算开销小，但容易出现预测值跳变较大的情况。因此，由小波神经网络和平稳随机过程结合得到的混合预测模型，结合了小波神经网络和平稳随机过程的优点，使得小波神经网络和平稳随机过程进行优劣互补，能够有效地降低中长期预测的计算复杂度，提高中长期预测的精准程度。
在本发明实施例中，可将最后记录时间点到预设时间点之间的时间间隔划分为多个等距的时间片，将时间片的数量、最后记录时间点、最后记录时间点的空闲泊位数量输入混合预测模型，得到停车场在从最后记录时间点到预设时间点间每个时间片的空闲泊位数量，进而得到停车场在预测时间点的空闲泊位数量。
占用率输出单元53，用于根据预测得到的、停车场在预设时间点的空闲泊位数量，获得停车场在预设时间点的泊位占用率并输出。
在本发明实施例中，根据预测得到的、停车场在预设时间点的空闲泊位数量，可计算得到停车场在预设时间点的泊位占用率，输出停车场在预设时间点的泊位占用率，以帮助用户快速停车。
优选地，如图6所示，停车场泊位占用率预测装置还包括：
网络训练单元61，用于构建小波神经网络，通过采集到的停车场的历史训练数据对小波神经网络进行训练，获得训练好的小波神经网络。
在本发明实施例中，采集历史训练数据，优选地，由于人们往往以周为单位进行生活和工作的安排，以周为单位进行历史训练数据的采集，从而有效地提高后续泊位占用率预测的准确度。
在本发明实施例中，小波神经网络的数学模型可表示为：
Figure PCTCN2017117570-appb-000009
其中，Ej(t)表示停车场j在预设时间点t时的空闲泊位数量，f(x)为小波神经网络的解析函数，
Figure PCTCN2017117570-appb-000010
为小波神经网络的激励函数，可将Morlet母小波设置为激励函数，ak为伸缩因子，bk为平移因子，ωk为小波神经网络的输出权值，ωik为小波神经网络的输入权值。图3为小波神经网络的结构示例图，图中W11...Wnm为小波神经网络的输入权值，图中W1...Wm为小波神经网络的输出权值。
在本发明实施例中，训练小波神经网络即根据小波神经网络的预测结果对ak、bk、ωk和ωik进行调整。具体地，通过小波神经网络预测停车场在预设时间点t时的空闲泊位数量，将预测得到的空闲泊位数量与历史训练数据中预设时间点t时的空闲泊位数量比较，可得到小波神经网络的预测误差
Figure PCTCN2017117570-appb-000011
Ejn(t)为历史训练数据中停车场j在t时刻的空闲泊位数量，通过预设的梯度下降法对小波神经网络中ak、bk、ωk和ωik进行调整，直至小波神经网络的预测误差小于预设误差阈值。
数据分析单元62，用于将历史训练数据对应的时间序列划分多个等距的时间片，确定相邻时间片之间停车场空闲泊位数量变化所服从的分布类型、分布参数。
在本发明实施例中，将从历史训练数据对应的时间序列划分为多个等距的时间片{Δti|i＝1…n}，n为时间片的数量，从历史训练数据中获取每个时间片上的已占用泊位数量xi。停车场中泊位的占用与否取决于车辆的到达率和驶离率，且车辆的到达一般被认为是服从泊松分布的，所以从长期来看每个时间片内已占用泊位数量的分布为泊松分布的极限，即正态分布，可表示为
Figure PCTCN2017117570-appb-000012
其中，μi、
Figure PCTCN2017117570-appb-000013
为xi正态分布的分布参数，具体地，可通过最大似然法(MLE)求解每个xi的分布参数μi、
Figure PCTCN2017117570-appb-000014
在本发明实施例中，随着Δti不同xi的分布参数也不同，因此在历史训练数据的时间轴上，空闲泊位数量变化是一个非平稳的高斯过程。在两个相邻时间片ti、ti+1之间空闲泊位数量的变化为Δxi＝xi+1-xi，xi+1和xi被视为两个独立同正态分布的变量，所以Δxi也服从正态分布，且
Figure PCTCN2017117570-appb-000015
其中，cov(xi,xi+1)为xi+1和xi的协方差算子。
中长期预测构建单元63，用于根据相邻时间片之间停车场空闲泊位数量变化所服从的分布类型、分布参数，构建中长期预测函数。
在本发明实施例中，中长期预测函数可表示为
Figure PCTCN2017117570-appb-000016
其中，Ej(t0)表示停车场j在t0时刻的空闲泊位数量，n为t和t0之间时间片的数量。
混合模型生成单元64，用于将训练好的小波神经网络和中长期预测函数进行加权组合，获得混合预测模型，根据历史训练数据确定混合预测模型中小波神经网络、中长期预测函数分别对应的权值参数。
在本发明实施例中，为了将小波神经网络和基于非平稳随机过程构建的中长期预测函数进行优势结合和优劣互补，将训练好的小波神经网络和中长期预测函数进行加权组合，得到混合预测模型，混合预测模型可表示为：
Ej(t)＝a*g(t0,n)+b*f[g(t0,n-1)]，其中，a、b分别为小波神经网络、中长期预测函数对应的权值参数，可根据最小二乘法和历史训练数据进行确定，混合预测模型先由中长期预测函数进行预测，再将中长期预测函数所得预测结果的直接前序结果输入到小波神经网络进行一步预测，最后中长期预测函数的预测结果、小波神经网络的预测结果进行线性加权，最终得到的预测结果既有一定的随机性，又不至于出现较大偏离。
优选地，停车场泊位占用率预测装置还包括：
混沌性检测单元，用于通过预设的最大李雅普诺夫指数法对混合检测模型的预测结果进行检测，以确定混合预测模型的预测结果是否具有混沌性；以及
预测跳转单元，用于当混合预测模型的预测结果不具有混沌性时，由泊位预测单元52执行对停车场在预设时间点的空闲泊位数量进行预测的操作。
在本发明实施例中，在计算混合检测模型的预测结果是否具有混沌性时，先利用互信息法和伪领域法确定序列的时间延迟和嵌入维数，在对停车位数据进行相空间重构，最后通过小数据量法求得最大李雅普诺夫指数。具体地，最大李雅普诺夫指数的计算公式可表示为：
V(t)＝Lyapunov[E(t0),E(t1),....E(t)]，Lyapunov为最大李雅普诺夫指数算子，E(t0),E(t1),....E(t)为t0到t时刻所预测的泊位数量。当V(t)>0时认为混合检测模型的预测结果具有混沌性，由占用率输出单元53执行获得停车场在预设时间点的泊位占用率并输出的操作，当V(t)≤0时为混合检测模型的预测结果不具有混沌性，由泊位预测单元52执行对停车场在预设时间点的空闲泊位数量进行预测的操作。
在本发明实施例中，通过小波神经网络、非平稳随机过程结合构建的混合预测模型，对停车场在预设时间点的空闲泊位数量进行预测，在实现泊位占用率中长期预测的同时，解决了小波神经网络在中长期预测时预测值大幅偏离的问题，也解决了非平稳随机过程在预测过程中预测值跳变较大的问题，同时避免了采用最大李雅普诺夫指数法进行预测的计算复杂度，从而有效地提高了泊 位占用率中长期预测的效率和准确率。
在本发明实施例中，停车场泊位占用率预测装置的各单元可由相应的硬件或软件单元实现，各单元可以为独立的软、硬件单元，也可以集成为一个软、硬件单元，在此不用以限制本发明。
实施例四：
图7示出了本发明实施例四提供的计算设备的结构，为了便于说明，仅示出了与本发明实施例相关的部分。
本发明实施例的计算设备7包括处理器70、存储器71以及存储在存储器71中并可在处理器70上运行的计算机程序72。该处理器70执行计算机程序72时实现上述各个方法实施例中的步骤，例如图1所示的步骤S101至S103。或者，处理器70执行计算机程序72时实现上述各装置实施例中各单元的功能，例如图5所示单元51至53的功能。
在本发明实施例中，通过小波神经网络、非平稳随机过程结合构建的混合预测模型，对停车场在预设时间点的空闲泊位数量进行预测，在实现泊位占用率中长期预测的同时，解决了小波神经网络在中长期预测时预测值大幅偏离的问题，也解决了非平稳随机过程在预测过程中预测值跳变较大的问题，同时避免了采用最大李雅普诺夫指数法进行预测的计算复杂度，从而有效地提高了泊位占用率中长期预测的效率和准确率。
该处理器70执行计算机程序72时实现上述各个方法实施例中的步骤具体可参考前述方法实施例的描述，在此不再赘述。
实施例五：
在本发明实施例中，提供了一种计算机可读存储介质，该计算机可读存储介质存储有计算机程序，该计算机程序被处理器执行时实现上述各个方法实施例中的步骤，例如，图1所示的步骤S101至S103。或者，该计算机程序被处理器执行时实现上述各装置实施例中各单元的功能，例如图5所示单元51至53的功能。
在本发明实施例中，通过小波神经网络、非平稳随机过程结合构建的混合预测模型，对停车场在预设时间点的空闲泊位数量进行预测，在实现泊位占用率中长期预测的同时，解决了小波神经网络在中长期预测时预测值大幅偏离的问题，也解决了非平稳随机过程在预测过程中预测值跳变较大的问题，同时避免了采用最大李雅普诺夫指数法进行预测的计算复杂度，从而有效地提高了泊位占用率中长期预测的效率和准确率。该计算机程序被处理器执行时实现上述各个方法实施例中的步骤具体可参考前述方法实施例的描述，在此不再赘述。
本发明实施例的计算机可读存储介质可以包括能够携带计算机程序代码的任何实体或装置、记录介质，例如，ROM/RAM、磁盘、光盘、闪存等存储器。
以上所述仅为本发明的较佳实施例而已，并不用以限制本发明，凡在本发明的精神和原则之内所作的任何修改、等同替换和改进等，均应包含在本发明的保护范围之内。

Claims (10)

  1. 一种停车场泊位占用率预测方法，其特征在于，所述方法包括下述步骤：
    当接收到停车场在预设时间点的泊位预测请求时，从所述停车场的历史数据中获取所述停车场在最后记录时间点的空闲泊位数量；
    通过预先训练好的混合预测模型和所述最后记录时间点的空闲泊位数量，对所述停车场在所述预设时间点的空闲泊位数量进行预测，所述混合预测模型通过预设的小波神经网络和预设的非平稳随机过程结合训练得到；
    根据预测得到的、所述停车场在所述预设时间点的空闲泊位数量，获得所述停车场在所述预设时间点的泊位占用率并输出。
  2. 如权利要求1所述的方法，其特征在于，对所述停车场在所述预设时间点的空闲泊位数量进行预测的步骤之后，获得所述停车场在所述预设时间点的泊位占用率并输出的步骤之前，所述方法还包括：
    通过预设的最大李雅普诺夫指数法对所述混合检测模型的预测结果进行检测，以确定所述混合预测模型的预测结果是否具有混沌性；
    当所述混合预测模型的预测结果不具有混沌性时，跳转至对所述停车场在所述预设时间点的空闲泊位数量进行预测的步骤。
  3. 如权利要求1所述的方法，其特征在于，当接收到停车场在预设时间点的泊位预测请求时，从所述停车场的历史数据中获取所述停车场在最后记录时间点的空闲泊位数量的步骤之前，所述方法还包括：
    构建小波神经网络，通过采集到的所述停车场的历史训练数据对所述小波神经网络进行训练，获得训练好的所述小波神经网络；
    将所述历史训练数据对应的时间序列划分多个等距的时间片，确定相邻所述时间片之间所述停车场空闲泊位数量变化所服从的分布类型、分布参数；
    根据相邻所述时间片之间所述停车场空闲泊位数量变化所服从的分布类型、分布参数，构建中长期预测函数；
    将所述训练好的小波神经网络和所述中长期预测函数进行加权组合，获得 所述混合预测模型，根据所述历史训练数据确定所述混合预测模型中所述小波神经网络、所述中长期预测函数分别对应的权值参数。
  4. 如权利要求3所述的方法，其特征在于，构建小波神经网络，通过采集到的所述停车场的历史训练数据对所述小波神经网络进行训练，获得训练好的所述小波神经网络的步骤，包括：
    构建所述小波神经网络，所述小波神经网络为：
    Figure PCTCN2017117570-appb-100001
    其中，Ej(t)表示所述停车场j在t时刻的空闲泊位数量，f(x)为所述小波神经网络的解析函数，
    Figure PCTCN2017117570-appb-100002
    为所述小波神经网络的激励函数，ak为伸缩因子，bk为平移因子，ωk为所述小波神经网络的输出权值，ωik为所述小波神经网络的输入权值；
    通过所述历史训练数据和预设的梯度下降法，对所述小波神经网络进行训练，直至所述小波神经网络的训练误差小于预设误差阈值，所述小波神经网络的训练误差通过公式
    Figure PCTCN2017117570-appb-100003
    计算得到，其中，e为所述训练误差，Ejn(t)为所述历史训练数据中所述停车场j在t时刻的空闲泊位数量。
  5. 如权利要求3所述的方法，其特征在于，所述混合预测模型为：
    Ej(t)＝a*g(t0,n)+b*f[g(t0,n-1)]，其中，a、b为所述混合预测模型中所述小波神经网络、所述中长期预测函数分别对应的所述权值参数，g(t0,n)为所述中长期预测函数，
    Figure PCTCN2017117570-appb-100004
    Ej(t0)表示所述停车场j在t0时刻的空闲泊位数量，n为t和t0之间所述时间片的数量，Δxi为所述相邻时间片ti、ti+1之间变化的空闲泊位数量。
  6. 一种停车场泊位占用率预测装置，其特征在于，所述装置包括：
    请求接收单元，用于当接收到停车场在预设时间点的泊位预测请求时，从所述停车场的历史数据中获取所述停车场在最后记录时间点的空闲泊位数量；
    泊位预测单元，用于通过预先训练好的混合预测模型和所述最后记录时间点的空闲泊位数量，对所述停车场在所述预设时间点的空闲泊位数量进行预测，所述混合预测模型通过预设的小波神经网络和预设的非平稳随机过程结合训练得到；以及
    占用率输出单元，用于根据预测得到的、所述停车场在所述预设时间点的空闲泊位数量，获得所述停车场在所述预设时间点的泊位占用率并输出。
  7. 如权利要求6所述的装置，其特征在于，所述装置还包括：
    混沌性检测单元，用于通过预设的最大李雅普诺夫指数法对所述混合检测模型的预测结果进行检测，以确定所述混合预测模型的预测结果是否具有混沌性；以及
    预测跳转单元，用于当所述混合预测模型的预测结果不具有混沌性时，由泊位预测单元执行对所述停车场在所述预设时间点的空闲泊位数量进行预测的操作。
  8. 如权利要求6所述的装置，其特征在于，所述装置还包括：
    网络训练单元，用于构建小波神经网络，通过采集到的所述停车场的历史训练数据对所述小波神经网络进行训练，获得训练好的所述小波神经网络；
    数据分析单元，用于将所述历史训练数据对应的时间序列划分多个等距的时间片，确定相邻所述时间片之间所述停车场空闲泊位数量变化所服从的分布类型、分布参数；
    中长期预测构建单元，用于根据相邻所述时间片之间所述停车场空闲泊位数量变化所服从的分布类型、分布参数，构建中长期预测函数；以及
    混合模型生成单元，用于将所述训练好的小波神经网络和所述中长期预测函数进行加权组合，获得所述混合预测模型，根据所述历史训练数据确定所述混合预测模型中所述小波神经网络、所述中长期预测函数分别对应的权值参数。
  9. 一种计算设备，包括存储器、处理器以及存储在所述存储器中并可在所述处理器上运行的计算机程序，其特征在于，所述处理器执行所述计算机程序 时实现如权利要求1至5任一项所述方法的步骤。
  10. 一种计算机可读存储介质，所述计算机可读存储介质存储有计算机程序，其特征在于，所述计算机程序被处理器执行时实现如权利要求1至5任一项所述方法的步骤。
PCT/CN2017/117570 2017-12-20 2017-12-20 停车场泊位占用率预测方法、装置、设备及存储介质 WO2019119319A1 (zh)

Priority Applications (1)

Application Number Priority Date Filing Date Title
PCT/CN2017/117570 WO2019119319A1 (zh) 2017-12-20 2017-12-20 停车场泊位占用率预测方法、装置、设备及存储介质

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
PCT/CN2017/117570 WO2019119319A1 (zh) 2017-12-20 2017-12-20 停车场泊位占用率预测方法、装置、设备及存储介质

Publications (1)

Publication Number Publication Date
WO2019119319A1 true WO2019119319A1 (zh) 2019-06-27

Family

ID=66992929

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2017/117570 WO2019119319A1 (zh) 2017-12-20 2017-12-20 停车场泊位占用率预测方法、装置、设备及存储介质

Country Status (1)

Country Link
WO (1) WO2019119319A1 (zh)

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN112863235A (zh) * 2021-01-05 2021-05-28 北京智联云海科技有限公司 一种预测车位空闲率的方法和装置
TWI750673B (zh) * 2020-05-26 2021-12-21 恆景科技股份有限公司 環境光估算系統

Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20030107498A1 (en) * 2001-12-07 2003-06-12 Drennan George A. Automated space availability coordinator
CN102024343A (zh) * 2010-12-20 2011-04-20 东南大学 停车场有效泊位占有率短时预测方法
CN102855773A (zh) * 2012-09-13 2013-01-02 东南大学 一种停车场有效泊位占有率短时预测方法
CN102867407A (zh) * 2012-09-13 2013-01-09 东南大学 一种停车场有效泊位占有率多步预测方法
CN106548254A (zh) * 2016-11-16 2017-03-29 上海理工大学 一种有效停车泊位的预测方法

Patent Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20030107498A1 (en) * 2001-12-07 2003-06-12 Drennan George A. Automated space availability coordinator
CN102024343A (zh) * 2010-12-20 2011-04-20 东南大学 停车场有效泊位占有率短时预测方法
CN102855773A (zh) * 2012-09-13 2013-01-02 东南大学 一种停车场有效泊位占有率短时预测方法
CN102867407A (zh) * 2012-09-13 2013-01-09 东南大学 一种停车场有效泊位占有率多步预测方法
CN106548254A (zh) * 2016-11-16 2017-03-29 上海理工大学 一种有效停车泊位的预测方法

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
TWI750673B (zh) * 2020-05-26 2021-12-21 恆景科技股份有限公司 環境光估算系統
CN112863235A (zh) * 2021-01-05 2021-05-28 北京智联云海科技有限公司 一种预测车位空闲率的方法和装置

Similar Documents

Publication Publication Date Title
CN108010378B (zh) 停车场泊位占用率预测方法、装置、设备及存储介质
CN106503840B (zh) 停车场可用车位预测方法及系统
CN105023433B (zh) 一种高速公路交通异常事件影响范围预估方法
Xiao et al. How likely am I to find parking?–A practical model-based framework for predicting parking availability
Zhang et al. An improved k-nearest neighbor model for short-term traffic flow prediction
Christofa et al. Person-based traffic responsive signal control optimization
JP6064437B2 (ja) カーシェアリングシステムの運用管理システムおよび方法
CN105513414A (zh) 基于实时交通查询和云模型的停车场车位预测和推荐方法
CN107610464A (zh) 一种基于高斯混合时间序列模型的轨迹预测方法
CN106933649B (zh) 基于移动平均和神经网络的虚拟机负载预测方法及系统
CN103368787A (zh) 信息处理装置、信息处理方法和程序
CN106887141B (zh) 基于排队论的连续交通节点拥塞程度预测模型、系统及方法
CN110858973A (zh) 小区网络流量预测方法及装置
CN112907970B (zh) 一种基于车辆排队长度变化率的可变车道转向控制方法
CN104217258A (zh) 一种电力负荷条件密度预测方法
WO2019119319A1 (zh) 停车场泊位占用率预测方法、装置、设备及存储介质
CN104851280A (zh) 一种车辆行驶控制方法、装置、系统及相关设备
CN109005223A (zh) 物联网资源调度方法及系统、计算机可读存储介质和终端
CN112561285A (zh) 网点的推荐方法、装置、电子设备及计算机可读存储介质
CN108564810B (zh) 一种车位共享系统及其方法
CN109615885B (zh) 一种智能交通信号控制方法、装置及系统
Haliem et al. AdaPool: An adaptive model-free ride-sharing approach for dispatching using deep reinforcement learning
CN113140108B (zh) 一种网联智能交通系统中的云端交通态势预测方法
CN112287503B (zh) 用于交通需求预测的动态空间网络构建方法
CN105491079A (zh) 云计算环境中调整应用所需资源的方法及装置

Legal Events

Date Code Title Description
121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 17935679

Country of ref document: EP

Kind code of ref document: A1

NENP Non-entry into the national phase

Ref country code: DE

122 Ep: pct application non-entry in european phase

Ref document number: 17935679

Country of ref document: EP

Kind code of ref document: A1