JPH054076U - Route predictor - Google Patents

Abstract

(57) [Summary] [Purpose] In a route prediction device that monitors a flight target and predicts its route, when the target that is flying due to the terrain becomes hidden in the shade of the mountain and becomes unobservable. To predict the flight path. [Structure] Observation unit 1 that observes the target and outputs the observed value.
And a terrain information section 2 that accumulates terrain information and outputs it as a pattern signal, and outputs the predicted observation value of the target by inputting the observation value and the terrain information into a neural network, and presenting them in the unit time past. It comprises a prediction unit 3 which inputs an error signal between a predicted observation value at which time is predicted and an observed value at the current time and learns the predicted value. [Effect] It is possible to predict the route that changes depending on the terrain when predicting the route of the target that will fly due to the influence of the terrain.

Description

[Detailed description of the device]

[0001]

[Industrial applications]

 This device is a route predicting device that predicts the route between flight targets even if they are hidden in unobservable places such as San'in, and predicts where they will be next observed.

[0002]

[Prior Art]

 Conventionally, a tracking filter has been used in a route prediction device for the purpose of target tracking when it is always observable.

FIG. A. Singer and K.K. W. behnke; Real-time tracking filter evaluation and selection for technical applications, IEEE TRANS. , VOL. AES7, NO. 1 (1971) is a block diagram of the tracking filter. This R. A. The Singer tracking filter is an application of the Kalman filter and is the basis of the subsequent tracking filtering. In order to use this tracking filter as a route predicting device, the target is observable and is operated as it is in FIG.

First, the configuration of a conventional route prediction device when a target is observable will be described with reference to FIG. In FIG. 4, 1 is an observing unit that observes the target and outputs the observed value, and 3 is a predicting unit that inputs the observed value and outputs the estimated state and predicted state of the target. In the observation unit 1, 4 is a sensor for observing a target, and a radar or the like is used. A data processor 5 processes an output signal of the sensor 4 and outputs an observation value. Further, in the above-mentioned prediction unit 3, 6 is a delay element, which inputs a prediction state, delays it within a unit time, and outputs it. Here, the above-mentioned “observed value” is information only on the target position, whereas “state” is information to which velocity or acceleration is added. Further, in the figure, k + 1 / k in parentheses in the prediction state symbol means that the state is predicted at time k + 1 based on the observed values obtained up to time k. An observation method processor 7 inputs a prediction state and outputs a prediction observation value. Reference numeral 8 is an addition element, which calculates the difference between the predicted observed value and the observed value. This difference signal is called the innovation process in the Kalman filter. A filter gain processing element 51 calculates the sum of the modified signal and the predicted signal. This signal is in the estimated state. Reference numeral 53 is a system dynamics processing element, which inputs the estimated state and outputs a predicted state ahead of a unit time. The predicted state is input to the delay element again and used for processing after a unit time.

The above-mentioned conventional method will be described with reference to the flowchart of FIG. First, the sensor 4 observes the target, and the data processor 5 processes the output data and outputs it as an observation value. Then, the series of operations ST51 for obtaining the target estimated state and the predicted state from the observed values follow the series of operations of the Kalman filter. A series of calculation formulas of the Kalman filter are shown in Formulas 1 to 5.

[0006]

[Equation 1]

[0007]

[Equation 2]

[0008]

[Equation 3]

[0009]

[Equation 4]

[0010]

[Equation 5]

In ST52, the observation equation processing element 7 inputs the prediction state at the current time obtained from the delay element 6, calculates the second term in [] of the equation 1, and outputs it. This process is the process of extracting information about the position from the predicted values of the target position and velocity, and here we will call this value the predicted observation value. Next, in ST53, the addition / subtraction element 8 inputs the predicted value and the predicted value, calculates the value in [] of the equation 1 and outputs it. This value is called the innovation process in the Kalman filter. Further, in ST54, the filter gain processing element 51 obtains the optimum filter gain in accordance with Formulas 2, 3 and 4, and calculates and outputs the second term on the right side of Formula 1. This value has a meaning as a modified signal. Then, in ST55, the addition / subtraction element 52 inputs the correction signal and the prediction state described later, obtains and outputs the value of the whole equation 1. This value has a meaning as an estimated state. Here, in ST56, the estimated state is set to the input of the system dynamics processing element 53. After that, proceeding to ST57, the system dynamics processing element 53 inputs the estimated state, calculates the equation 5, and outputs the predicted state in the unit time future. The estimated state and the predicted state are outputs of the conventional route prediction device. Finally, in ST6, it is determined whether or not to end. After a unit time, it returns to before ST1 and repeats the same processing. In this case, the prediction state output in ST57 is used as a prediction state in which the current time is predicted after a unit time and a unit time before. This time shift is represented by the delay element 6.

[0012]

[Problems to be solved by the device]

 The conventional route prediction device does not predict the route when the target is unobservable, and does not consider the target that flies due to the influence of the terrain, so the unobservable low altitude is affected by the terrain. However, when estimating the route of the flying target, there was a problem such as lack of flexibility in the terrain.

In order to solve the above problems, the present invention uses a neural network having terrain information and a learning function to fly a flight path in an unobservable area of a target that is affected by terrain and flies. The goal is to make flexible forecasts for.

[0014]

[Means for Solving the Problems]

 The route prediction device according to the present invention uses a neural network having a learning function in the prediction unit, learns a target flight route depending on the terrain by observation, and enters the unobservable region from the learning result. It predicts the route of the target.

[0015]

[Action]

 The route prediction device according to the present invention is a device for automatically predicting a target flight route that becomes unobservable after learning by the prediction unit automatically learning the target flight route depending on the terrain from the target observation value and terrain information. Is.

[0016]

【Example】

 An embodiment in which the target of this invention can be observed will be described below with reference to FIG. In FIG. 1, 1 is an observing unit that observes a target and outputs the observed value, 2 is a terrain information unit that accumulates terrain information and outputs the terrain information as a pattern signal, and 3 is the observed value and the terrain information , And outputs the predicted observed value of the target, calculates the error between the predicted observed value that predicted the current time in the past of the unit time and the observed value at the current time, and based on the error signal, the observed value Is a prediction unit that learns the prediction of. In the observing unit 1, 4 is a sensor for observing a target, and a radar or the like is used. A data processor 5 processes an output signal of the sensor 4 and outputs an observation value. In the prediction unit 3, 9 is a neural network, which inputs the observation value and a pattern signal of topographical information described later, and outputs a prediction observation value in the unit time future. Reference numeral 6 is a delay element, which outputs a prediction value obtained by predicting the present time in the unit time past. Numeral 8 is an addition / subtraction element, which calculates the difference between the predicted observed value at which the current time was predicted in the unit time past and the observed value at the current time. This difference signal is called an error signal in the neural network. 9 learns the output of a more accurate predicted observation value from the error signal. In the terrain information section 2, 10 is a memory for accumulating the terrain information and outputs the terrain information. This topographic information is stored in advance. A patterning element 11 inputs the topographical information, converts the topographical information used in the neural network into a pattern signal, and outputs the pattern signal.

Next, an embodiment of the present invention when the target cannot be observed will be described with reference to FIG. In FIG. 2, 2 is a terrain information section, and 3 is a prediction section. Since the predictor 3 cannot obtain an observed value, it consists only of neural network elements. The neural network element inputs a repeated prediction observation value, inputs a pattern signal of topographical information described later, and outputs the prediction observation value. Further, in the terrain information unit 3, 10 is a memory for accumulating the terrain information, and outputs the terrain information. A patterning element 11 inputs the topographical information and outputs the topographical information used in the neural network as a pattern signal.

A method for realizing the route prediction device will be described with reference to the flowchart of FIG. First, in ST1, it is determined whether or not the target is observable. If it is observable, the process proceeds to ST2, the target is observed by the sensor 4, the output data of the data processor 5 is processed, and output as an observed value. Next, in ST3, the predicted observation value at which the current time is predicted in the unit time past and the observed value at the current time are input to the addition / subtraction element, and the error signal is output. Next, at ST4, the terrain information is converted into a pattern signal. Further, in ST5, the neural network executes a learning algorithm to learn the output of an accurate predicted value. Equations 6 to 8 show a series of equations for learning the neural network by back propagation.

[0019]

[Equation 6]

[0020]

[Equation 7]

[0021]

[Equation 8]

Here, Expression 6 defines the sum of inputs and the output to the i-th unit in the k-th layer of the hierarchical network. The weight of the connection between the i-th unit of the (k-1) -th layer and the j-th unit of the k-th layer is corrected by using Equation 7, and learning is performed. The prediction value is output using the weight of the connection corrected in Expression 7 by Expression 8. Finally, in ST6, it is determined whether or not to end. If not completed, the operation returns to ST1 and the same operation is repeated. Delay element 6 of FIG. 1 represents the same time shift as described in the prior art.

On the other hand, if it is determined in ST1 that the observation is impossible, the process proceeds to ST4 to convert the topographical information into a pattern signal. Further, in ST5, the above-mentioned pattern signal is input to the neural network and the predicted value is output. Finally, in ST6, it is determined whether or not to end. If not completed, the process returns to ST1 and the same operation is repeated.

[0024]

[Effect of the device]

 As described above, according to the present invention, since the prediction unit is a neural network, it is possible to learn the flight route of the target that flies depending on the terrain automatically and according to the terrain. , It is possible to predict the flight path for a target constrained by topography while it is not observable.

[Brief description of drawings]

FIG. 1 is a block diagram showing a configuration in an observable case in an embodiment of a route prediction device according to the present invention.

FIG. 2 is a block diagram showing the configuration of an embodiment of the route prediction apparatus according to the present invention in the case where it is unobservable.

FIG. 3 is a flowchart showing a route prediction method according to the present invention.

FIG. 4 is a block diagram showing a configuration of a conventional route prediction device.

FIG. 5 is a flowchart showing a conventional route prediction method.

[Explanation of symbols]

 1 Observation part 2 Topographic information part 3 Prediction part

Claims (1)

[Claims for utility model registration] [Claim 1] An observation unit comprising a sensor for observing a target and a data processor for processing an output signal of the sensor and outputting an observation value, a terrain information memory and terrain for preliminarily storing terrain information A terrain information section consisting of patterning elements for processing information and converting it into a pattern signal, a neural network for outputting the observed value from the observing section and the terrain information from the terrain information section, and a delay within a unit time The prediction function has the function of learning the predicted observation value by the neural network from the delay element that outputs the predicted observation value that predicts the current time in the past by the unit time and the difference between the predicted observation value output from the delay element and the observed value. A route prediction device comprising a unit.