A target long track prediction method, device, medium and product

Abstract

The application discloses a target long track prediction method and device, medium and product, relates to the target long track prediction field, and through simultaneously predicting original values, differential (or difference) and integral (or accumulation) three forms of a target sequence, error compensation is carried out on the prediction result of a main model by using inverse operation of the differential and the integral, so that the prediction accuracy is improved. Unlike the existing LSTM model, the triple LSTM model target long track prediction framework proposed by the application carries out error compensation by combining the inverse operation characteristics of the differential and the integral. The prior art does not have a similar prediction method using differential and integral compensation. The method can effectively reduce the prediction error, especially when processing sequence data with complex dynamic characteristics.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of target long track prediction, and in particular to a target long track prediction method, device, medium and product. BACKGROUND

[0002] The statements in this section merely provide background information related to the present disclosure and can not constitute the prior art.

[0003] Time series prediction technology plays a key role in air traffic control, maritime monitoring and other fields, especially in the target long track prediction scenario, which needs to balance short-term dynamic characteristics and long-term trend evolution. Traditional methods are mainly based on statistical principles and physical motion models, among which the most representative are autoregressive integrated moving average model (ARIMA) and Kalman filter algorithm. ARIMA model processes non-stationary sequence through difference operation, but its linear assumption is difficult to adapt to the nonlinear characteristics of target maneuvering behavior; Kalman filter performs well in handling noise, but it relies on accurate system state equation and has limited adaptability in complex environments.

[0004] With the development of deep learning technology, recurrent neural network (RNN) and its variant long short-term memory network (LSTM) have gradually become the mainstream method of time series prediction. LSTM effectively captures long-term dependencies through the gating mechanism and performs better than traditional methods in the track prediction task. However, existing researches mostly focus on the optimization and improvement of single models, and the main technical paths include: 1) increasing network depth to build multi-layer LSTM structure; 2) introducing attention mechanism to strengthen key time step features; 3) combining convolutional neural network to extract spatial features. These improvements have improved the prediction accuracy to some extent, but they are essentially improvements within a single model framework and have two inherent defects: first, a single model is difficult to balance the instantaneous volatility and long-term trend of sequence data, resulting in a trade-off between stability (reflected in trajectory jitter when speed / acceleration suddenly changes) and trend consistency (reflected in cumulative bias over a long time span); second, the fixed weight fusion method does not fully consider the error propagation characteristics in different prediction stages, which easily causes error accumulation and amplification.

[0005] Current multi-model fusion strategy research shows that using parallel prediction architecture has theoretical value to improve robustness, but there are still obvious limitations in practical application: 1) sub-models all take original sequences as training targets, and fail to take advantage of feature complementary effects brought by sequence transformation; 2) weight distribution mostly uses empirical values or simple optimization algorithms, lacking mathematical modeling of error propagation mechanism. Experimental data shows that in scenarios with more than 50 prediction steps, the prediction error of traditional methods will show a nonlinear growth trend, which exposes the lack of current technology in recognizing error dynamics.

[0006] It's worth noting that differential operations (reflecting the local rate of change of a series) and integral operations (reflecting the global cumulative amount of a series) have been shown to enhance features in traditional time series analysis. For example, differential integration methods can significantly improve the stationarity handling capabilities of ARIMA models, while integral transformations offer unique advantages in trend forecasting. However, within the deep learning framework, how to organically combine differential / integral operations with LSTM models to construct a forecasting system with self-error compensation capabilities remains a technical challenge. Summary of the Invention

[0007] The purpose of the present invention is to address the problems existing in the prior art and provide a target long-track prediction method, device, medium, and product, which simultaneously predicts the original value, differential and integral forms of the target sequence, and uses the inverse operations of differential and integral to dynamically compensate for the prediction results of the main model to improve the prediction accuracy; specifically: first, by establishing a triple LSTM model, it is used to train and predict the original output value sequence, the differential sequence of the original output value, and the integral sequence of the original output value of the sliding window generated data set, wherein the differential and integral prediction sequences are inversely transformed and restored to the 0th order sequence, and finally, by dynamically and reasonably adjusting the weights of the three models according to the sequence order, a good prediction effect can be achieved.

[0008] The technical solutions of the present invention are as follows:

[0009] A target long track prediction method, comprising:

[0010] Step S1: Generate a data set containing input data and output data from the original data sequence in a sliding window manner;

[0011] Step S2: Establish three independent LSTM models, which serve as the original value prediction model, the differential sequence prediction model, and the integral sequence prediction model respectively. The inputs of the three LSTM models are all the input data of the dataset, and the outputs correspond to the original output data, the differential sequence of the original output data, and the integral sequence of the original output data respectively.

[0012] Step S3: Based on the data set, the three LSTM models are trained separately until the preset convergence conditions are met;

[0013] Step S4: The three LSTM models perform actual predictions respectively. The output of the original value prediction model is the 0th order prediction sequence. The output of the differential sequence prediction model and the integral sequence prediction model are inversely transformed to restore to the 0th order prediction sequence.

[0014] Step S5: Dynamically adjust and assign the optimal weights of the three LSTM models according to the order of the predicted sequence;

[0015] Step S6: Based on the weight adjusted in step S5, the three 0-order prediction sequences are weighted and fused to obtain a final prediction result.

[0016] Further, the original value prediction model directly predicts the original value of the target sequence ;

[0017] The differential sequence prediction model predicts the differential sequence of the target sequence ; ;

[0018] The integral sequence prediction model predicts the integral sequence of the target sequence ; ;

[0019] The input of the three LSTM models is the sequence , where is the input sequence length;

[0020] The outputs of the three LSTM models are respectively:

[0021] The original value prediction model: ; where is the output sequence length;

[0022] The differential sequence prediction model: , where ;

[0023] The integral sequence prediction model: , where .

[0024] Further, the output result of the differential sequence prediction model is inversely transformed, including:

[0025] Differential sequence inverse operation: from restore :

[0026]

[0027] Where:

[0028] , the initial condition is .

[0029] Further, the output result of the integral sequence prediction model is inversely transformed, including:

[0030] Integral sequence inverse operation: from restore :

[0031]

[0032] wherein:

[0033] , initial condition is .

[0034] Further, the weight of the differential sequence prediction model decreases with time step, and the weight of the integral sequence prediction model increases with time step.

[0035] Further, the weight distribution method is specifically as follows:

[0036]

[0037]

[0038]

[0039] wherein:

[0040] is the current time;

[0041] is the weight of the original value prediction model;

[0042] is the weight of the differential sequence prediction model

[0043] is the weight of the integral sequence prediction model.

[0044] Further, the weighted fusion includes:

[0045]

[0046] wherein:

[0047] represents the final prediction result.

[0048] The application further provides an electronic device, comprising:

[0049] at least one processor; and a memory connected with the at least one processor in communication;

[0050] wherein the memory stores instructions executable by the at least one processor, and the at least one processor executes the method as described above by executing the instructions stored in the memory.

[0051] The application further provides a computer readable storage medium for storing instructions, when the instructions are executed, the method as described above is realized.

[0052] The application further provides a computer program product, which realizes the method as described above when the computer program is executed by a processor.

[0053] The application has the following advantages compared with the prior art:

[0054] 1. The prior art mainly trains and predicts multiple original output sequences respectively, and then weights the final prediction sequence. It cannot solve and balance the volatility and non-stationarity of the prediction result. The application expands the target long track prediction idea and method by introducing differential sequence prediction compensation, so that the prediction result is more stable.

[0055] 2. The prior art mainly trains and predicts multiple original output sequences respectively, and then weights the final prediction sequence. It cannot solve and balance the deviation of the overall prediction result caused by the poor historical trend accumulation effect of the prediction result. The application expands the target long track prediction idea and method by introducing integral sequence prediction compensation, so that the prediction result has the trend accumulation effect of the historical sequence.

[0056] 3. The prior art mainly trains and predicts multiple original output sequences respectively, and then weights the final prediction sequence. Since the weighting is a fixed coefficient distribution, it cannot solve and balance the error accumulation of the prediction sequence. The application expands the target long track prediction idea and method by dynamically distributing appropriate weights according to the prediction sequence order, so that the prediction result is more accurate.

[0057] 4. The prior art mainly trains and predicts multiple original output sequences respectively, and then weights the final prediction sequence. Since the influence of the model error on the overall error is not considered, it cannot solve and balance the problem of minimizing the final prediction result error. The application dynamically distributes the model weights according to the error accumulation and error offset law of the differential sequence and integral sequence prediction values in the process of restoring to 0 order, so as to minimize the final error. The application expands the target long track prediction idea and method by introducing dynamic weight distribution to reduce the final error, so that the prediction result is more accurate. BRIEF DESCRIPTION OF DRAWINGS

[0058] Figure 1 It is a target long track prediction method flowchart;

[0059] Figure 2 It is a structural schematic diagram of an electronic device provided in an embodiment of the application. DETAILED DESCRIPTION

[0060] It is to be understood that the terms "first", "second", and so on used in the description and claims are used to distinguish one entity or action from another, and do not necessarily require or imply any actual relationship or order between such entities or actions. Also, the terms "comprises", "comprising", or any other variant thereof are intended to cover non-exclusive inclusions, such that a process, method, article, or apparatus that comprises a list of elements does not include only those elements, but can also include other elements not expressly listed or inherent to such process, method, article, or apparatus. Without more limitations, an element defined by the phrase "comprising a" does not exclude the existence of additional identical elements in the process, method, article, or apparatus that includes the element.

[0061] The features and performances of the present application will be further described in detail below in conjunction with embodiments.

[0062] Embodiment One

[0063] In order to facilitate understanding of the technical solutions of the present application, the LSTM is described as follows.

[0064] The LSTM is a special recurrent neural network that can effectively solve the gradient vanishing or gradient explosion problem of traditional RNN by introducing memory cells and gating mechanisms. Its basic structure includes input gate, forget gate, output gate, etc., and the principle is as follows:

[0065] The input gate controls how new input information is added to the cell state. The calculation formula is:

[0066]

[0067] wherein, is the current input, is the hidden state at the previous time, and are weight matrices, is a bias term, is a sigmoid activation function.

[0068] The forget gate determines how much information in the cell state at the previous time is retained. The calculation formula is:

[0069]

[0070] wherein, is the current input, is the hidden state at the previous time, and are weight matrices, is a bias term, is a sigmoid activation function.

[0071] The output gate controls how information in the cell state is transformed into the hidden state. The calculation formula is:

[0072]

[0073] where, is the current input, is the hidden state at the previous time step, and are weight matrices, is a bias term, is a sigmoid activation function.

[0074] The candidate cell state is the source of new information, generated by the hyperbolic tangent function tanh, with the calculation formula:

[0075]

[0076] where, is the current input, is the hidden state at the previous time step, and are weight matrices, is a bias term.

[0077] Update the cell state, which is updated through the forget gate and the input gate, with the formula as follows:

[0078]

[0079] where, denotes element-wise multiplication.

[0080] Calculate the current hidden state, which is determined by the output gate and the current cell state, with the formula as follows:

[0081]

[0082] Generate the output, which is generated by the hidden state, usually through a fully connected layer:

[0083]

[0084] where, is the output weight matrix, is the output bias term.

[0085] Please refer to Figure 1 , the embodiment proposes a target long track prediction method, which specifically includes the following steps:

[0086] Step S1: generate a data set containing input data and output data from the original data sequence in a sliding window manner;

[0087] Step S2: three independent LSTM models are established as an original value prediction model (0 order), a differential sequence prediction model (1 order) and an integral sequence prediction model (-1 order), respectively, the input of the three LSTM models is the input data of the data set, and the output corresponds to the original output data, the differential sequence of the original output data and the integral sequence (or cumulative sequence) of the original output data, respectively; wherein the original value prediction model is used as the main model;

[0088] Step S3: based on the data set, the three LSTM models are trained respectively until the preset convergence condition is met, such as reaching the number of iterations or reaching the expected accuracy; it should be noted that after the training of the three models, they are used to predict the original value of the future track sequence, the differential value of the future track sequence and the integral value of the future track sequence;

[0089] Step S4: actual prediction is performed by the three LSTM models, the output result of the original value prediction model is the 0 order prediction sequence, and the output results of the differential sequence prediction model and the integral sequence prediction model are inversely transformed to restore the 0 order prediction sequence;

[0090] Step S5: the best weight of the three LSTM models is dynamically adjusted and distributed according to the order of the prediction sequence; that is, the weight parameters of the three LSTM models need to be adjusted according to the characteristics of the data sequence to balance the influence strength of each model on the final prediction result, and the best parameter value suitable for the data sequence can obtain the best prediction effect;

[0091] Step S6: based on the weight adjusted in step S5, the three 0 order prediction sequences are weighted and fused to obtain the final prediction result.

[0092] In this embodiment, it should be noted that the differential term introduced by the above method can maintain the stability and smoothness of the prediction result;

[0093] In this embodiment, it should be noted that the integral term introduced by the above method can capture the long-term trend and cumulative effect of the data.

[0094] Further, the differential term and the integral term compensate for the problems of result fluctuation and long-term cumulative trend deviation of the original value prediction, and by appropriately adjusting the weight coefficients of the three parts, a high-precision prediction result can be obtained.

[0095] In this embodiment, it should be noted that in the long-track target prediction scenario, a differential and integral sequence prediction model is introduced to compensate for the original sequence prediction of the original value prediction model, thereby optimizing the prediction effect. Due to the introduction of differentials and integrals in the prediction, this method is only applicable to multi-step (multi-point) predictions and cannot be used for single-step (single-point) predictions.

[0096] In this embodiment, specifically, the original value prediction model directly predicts the target sequence The original value of

[0097] Differential sequence prediction model predicts target sequence The differential (difference) sequence of ;

[0098] Integral sequence prediction model predicts target sequence The integral (accumulation) sequence of ;

[0099] The input of the three LSTM models is a sequence ,in is the length of the input sequence;

[0100] The outputs of the three LSTM models are:

[0101] Original value prediction model: ;in is the output sequence length;

[0102] Differential sequence prediction model: ,in ;

[0103] Integral sequence prediction model: ,in .

[0104] In this embodiment, specifically, the output result of the differential sequence prediction model is inversely transformed, including:

[0105] Inverse operation of differential sequence: from recover :

[0106]

[0107] in:

[0108] , the initial conditions are .

[0109] In this embodiment, specifically, the output result of the integral sequence prediction model is inversely transformed, including:

[0110] Integral sequence inverse operation: from Recovery :

[0111]

[0112] Wherein:

[0113] The initial condition is .

[0114] In this embodiment, it should be noted that the prediction error of the original value prediction model, the differential sequence prediction model, and the integral sequence prediction model is , , and , respectively:

[0115] Error propagation of differential inverse operation:

[0116]

[0117] Error propagation of integral inverse operation:

[0118]

[0119] Differential sequence inverse operation: the error will accumulate, and the prediction error will be larger at the later stage, so the weight should be gradually reduced.

[0120] Integral sequence inverse operation: the error can be partially offset, and the weight can be appropriately gradually increased.

[0121] In this embodiment, specifically, the weight of the differential sequence prediction model decreases with the time step, and the weight of the integral sequence prediction model increases with the time step.

[0122] That is, the weight distribution of the differential sequence prediction model should be that the weight becomes smaller and smaller from front to back in sequence order to reduce the error accumulation caused by sequence recovery;

[0123] The weight distribution of the integral sequence prediction model should be that the weight becomes larger and larger from front to back in sequence order to enhance the error offset when the adjacent two items are subtracted in the sequence recovery process.

[0124] In this embodiment, it should be noted that the differential sequence inverse operation will cause the accumulation of errors, and the integral sequence inverse operation can partially offset the errors through the error difference of adjacent two items. Therefore, due to the different error characteristics of the differential sequence prediction model and the integral sequence prediction model, the weight is dynamically allocated.

[0125] In this embodiment, specifically, the weight allocation method is as follows:

[0126]

[0127]

[0128]

[0129] wherein:

[0130] is the current time;

[0131] is the weight of the original value prediction model;

[0132] is the weight of the differential sequence prediction model

[0133] is the weight of the integral sequence prediction model.

[0134] , and ;

[0135] which can dynamically adjust the weight according to the error characteristics. By reasonably selecting the fusion weight , the prediction error can be effectively reduced.

[0136] In the embodiment, specifically, the weighted fusion includes:

[0137]

[0138] wherein:

[0139] represents the final prediction result.

[0140] In the embodiment, it should be noted that the target long track prediction method proposed above combines the original value, the differential sequence and the integral sequence of the target sequence, and uses the inverse operation of the differential and integral to compensate the prediction result of the main model to improve the prediction accuracy.

[0141] Based on the same technical concept, the embodiment of the present application also provides an electronic device which can realize the target long track prediction method process provided by the above-mentioned embodiment of the present application. In an embodiment, the electronic device can be a server, or a terminal device or other electronic device. As shown in the figure, Figure 2 the electronic device can include:

[0142] at least one processor, and a memory connected with the at least one processor, and the specific connection medium between the processor and the memory in the embodiment of the present application is not limited, Figure 2 for example, the connection between the processor and the memory through the bus is taken as an example. The bus is in Figure 2The connection between the other components is indicated by a thick line, which is only illustrative and not limited. The bus can be divided into an address bus, a data bus, a control bus, etc. For convenience, Figure 2 The bus is indicated by a thick line, but it does not mean that there is only one bus or only one type of bus. Alternatively, the processor can also be referred to as a controller, and the name is not limited.

[0143] In the embodiment of the application, the memory stores instructions executable by the at least one processor, and the at least one processor can execute the target long track prediction method discussed above by executing the instructions stored in the memory. The processor can realize Figure 2 The functions of each module in the device shown.

[0144] The processor is the control center of the device, and can connect each part of the entire control device through various interfaces and lines. By running or executing the instructions stored in the memory and calling the data stored in the memory, the device processes data and performs various functions, thereby monitoring the entire device.

[0145] In an alternative design, the processor can include one or more processing units, and the processor can integrate an application processor and a modem processor. The application processor mainly processes the operating system, user interface, application program, etc., and the modem processor mainly processes wireless communication. It can be understood that the above-mentioned modem processor can also not be integrated into the processor. In some embodiments, the processor and the memory can be implemented on the same chip, and in some embodiments, they can also be implemented on separate chips respectively.

[0146] The processor can be a general-purpose processor, such as a CPU, a digital signal processor, an application-specific integrated circuit, a field programmable gate array, or other programmable logic device, a discrete gate or transistor logic device, a discrete hardware component, which can realize or execute the methods, steps and logic block diagrams disclosed in the embodiments of the application. The general-purpose processor can be a microprocessor or any conventional processor. The steps of the target long track prediction method disclosed in the embodiments of the application can be directly embodied by the hardware processor for execution, or executed by the combination of hardware and software modules in the processor.

[0147] The memory, as a non-volatile computer readable storage medium, can be used to store non-volatile software programs, non-volatile computer executable programs and modules. The memory can include at least one type of storage medium, for example, can include flash memory, hard disk, multimedia card, card type memory, random access memory (RAM), static random access memory (SRAM), programmable read-only memory (PROM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), magnetic memory, magnetic disk, optical disk, etc. The memory is any other medium capable of carrying or storing desired program code in the form of instructions or data structures and capable of being accessed by a computer, but is not limited thereto. The memory in the embodiments of the present application can also be a circuit or any other device capable of realizing a storage function, used for storing program instructions and / or data.

[0148] By designing and programming the processor, the code corresponding to the target long track prediction method introduced in the foregoing embodiments can be fixed in the chip, so that the chip can execute the steps of the method of the foregoing embodiments when running. How to design and program the processor is a technology known to those skilled in the art, which will not be described here.

[0149] Based on the same inventive concept, the embodiments of the present application also provide a storage medium storing computer instructions, when the computer instructions run on a computer, the computer instructions make the computer execute the target long track prediction method discussed above.

[0150] In some optional embodiments, the present application also provides that various aspects of the target long track prediction method can also be implemented in the form of a program product, which includes program codes, when the program product runs on the device, the program codes are used to make the control device execute the steps of the target long track prediction method according to various exemplary embodiments of the present application described above in the specification.

[0151] It should be noted that although several units or subunits of the device are mentioned in the detailed description above, this division is merely exemplary and not mandatory. In fact, according to an embodiment of the present invention, the features and functions of two or more units described above can be embodied in one unit. Conversely, the features and functions of a unit described above can be further divided into multiple units to be embodied. In addition, although the operations of the method of the present invention are described in a specific order in the accompanying drawings, this does not require or imply that these operations must be performed in this specific order, or that all the operations shown must be performed to achieve the desired results. Additionally or alternatively, certain steps can be omitted, multiple steps can be combined into one step, and / or one step can be decomposed into multiple steps.

[0152] It will be understood by those skilled in the art that embodiments of the present invention may be provided as methods, systems, or computer program products. Thus, the present invention may take the form of an entirely hardware embodiment, an entirely software embodiment, or an embodiment combining software and hardware. Furthermore, the present invention may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to magnetic disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0153] The present invention is described with reference to flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to embodiments of the present invention. It should be understood that each process and / or block in the flowchart and / or block diagram, as well as a combination of processes and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to generate a server, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the process in the flowchart and / or block diagram. Figure 1 a process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0154] Program code for performing the operations of the present invention may be written using any combination of one or more programming languages, including object-oriented programming languages ​​such as Java, C++, and conventional procedural programming languages ​​such as "C" or similar programming languages. The program code may be executed entirely on the user's computing device, partially on the user's device, as a stand-alone software package, partially on the user's computing device and partially on a remote computing device, or entirely on a remote computing device or server.

[0155] In situations where the remote computing device is involved, the remote computing device can be connected to the user computing device through any kind of network, including a local area network (LAN) or a wide area network (WAN), or can be connected to an external computing device (e.g., using an Internet service provider to connect through the Internet).

[0156] These computer program instructions can also be stored in a computer-readable memory that can direct a computer or other programmable data processing apparatus to function in a particular manner, such that the instructions stored in the computer-readable memory produce an article of manufacture including instructions which implement the Figure 1 function specified in the flow or flows and / or blocks Figure 1 of the multiple flows or blocks.

[0157] These computer program instructions can also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer implemented process such that the instructions which execute on the computer or other programmable apparatus provide steps for implementing the Figure 1 function specified in the flow or flows and / or blocks Figure 1 of the multiple flows or blocks.

[0158] Furthermore, in some embodiments, a computer program product is also proposed, which, when executed by a processor, implements the above-mentioned target long track prediction method.

[0159] The above-described embodiments are merely intended to express the specific embodiments of the present application, and the description is relatively specific and detailed, but it should not be understood as a limitation on the protection scope of the present application. It should be noted that, for those skilled in the art, without departing from the technical concept of the present application, a number of modifications and improvements can be made, which all belong to the protection scope of the present application.

[0160] The above-described embodiments are merely intended to express the specific embodiments of the present application, and the description is relatively specific and detailed, but it should not be understood as a limitation on the protection scope of the present application. It should be noted that, for those skilled in the art, without departing from the technical concept of the present application, a number of modifications and improvements can be made, which all belong to the protection scope of the present application.

[0161] This Background section is intended to provide a general context for the present application, and as such the subject matter of the work of the present inventors that was made in the course of the inventors' own research and development, as described in this Background section, as well as the subject matter claimed in this section, is neither expressly nor impliedly admitted as prior art against the present application.

Claims

1. A target long track prediction method, characterized in that: include: Step S1: Generate a data set containing input data and output data from the original data sequence in a sliding window manner; Step S2: Establish three independent LSTM models, which serve as the original value prediction model, the differential sequence prediction model, and the integral sequence prediction model respectively. The inputs of the three LSTM models are all the input data of the dataset, and the outputs correspond to the original output data, the differential sequence of the original output data, and the integral sequence of the original output data respectively. Step S3: Based on the data set, the three LSTM models are trained separately until the preset convergence conditions are met; Step S4: The three LSTM models perform actual predictions respectively. The output of the original value prediction model is the 0th order prediction sequence. The output of the differential sequence prediction model and the integral sequence prediction model are inversely transformed to restore to the 0th order prediction sequence. Step S5: Dynamically adjust and assign the optimal weights of the three LSTM models according to the order of the predicted sequence; Step S6: Based on the weights adjusted in step S5, perform weighted fusion on the three 0th-order prediction sequences to obtain the final prediction result; The weight of the differential sequence prediction model decreases with time steps, and the weight of the integral sequence prediction model increases with time steps; The weight distribution method is as follows: in: For the current moment; The weights of the prediction model are the original values; is the weight of the differential sequence prediction model is the weight of the integrated sequence prediction model.

2. A target long track prediction method according to claim 1, characterized in that: The original value prediction model directly predicts the target sequence The original value of Differential sequence prediction model predicts target sequence Differential sequence of ; Integral sequence prediction model predicts target sequence The integral sequence of ; The input of the three LSTM models is a sequence ,in is the length of the input sequence; The outputs of the three LSTM models are: Original value prediction model: ;in is the output sequence length; Differential sequence prediction model: ,in ; Integral sequence prediction model: ,in .

3. A target long track prediction method according to claim 2, characterized in that: Perform inverse transformation on the output of the differential sequence prediction model, including: Inverse operation of differential sequence: from recover : in: , the initial conditions are .

4. A target long track prediction method according to claim 3, characterized in that: Perform an inverse transformation on the output of the integral sequence prediction model, including: Inverse operation of integral sequence: from recover : in: , the initial conditions are .

5. A target long track prediction method according to claim 1, characterized in that: Weighted fusion, including: in: Indicates the final prediction result.

6. An electronic device, characterized in that: include: at least one processor; and a memory communicatively coupled to the at least one processor; The memory stores instructions that can be executed by the at least one processor, and the at least one processor executes the method according to any one of claims 1 to 5 by executing the instructions stored in the memory.

7. A computer-readable storage medium, characterized in that The computer-readable storage medium is used to store instructions, and when the instructions are executed, the method according to any one of claims 1 to 5 is implemented.

8. A computer program product, characterized in that When the computer program is executed by a processor, the method according to any one of claims 1 to 5 is implemented.

Images