A result-selective combined model frequency modulation sequence prediction method and related equipment

Abstract

The present invention relates to the technical field of frequency modulation signal prediction, and discloses a result-selective combined model frequency modulation sequence prediction method and related equipment. The method comprises predicting a frequency modulation instruction sequence and correspondingly generating multiple groups of initial prediction results; calculating the deviations between the multiple groups of initial prediction results and an average prediction sequence, and performing a first-level elimination strategy on the multiple groups of initial preset neural network models based on the deviations to obtain a first-level preset neural network model set; calculating the distinctions between each two in the first-level preset neural network model set, and performing a second-level elimination strategy on the first-level preset neural network model set based on the distinctions to obtain a second-level preset neural network model set; calculating the deviation weights of the second-level preset neural network model set, and performing weighted fusion on the second-level preset neural network model set based on the deviation weights to generate a final prediction sequence. The present invention can improve the overall prediction accuracy.

Description

Technical Field

[0001] The present invention relates to the technical field of frequency modulation signal prediction, and in particular to a result-selective combined model frequency modulation sequence prediction method and related equipment. Background Art

[0002] Frequency-modulated signals are widely used in communications, radar, power system frequency control, and other scenarios. Their frequency variation patterns are often highly nonlinear (e.g., exponential modulation and burst modulation) and exhibit long-period time-dependent characteristics (e.g., periodic frequency modulation or gradual frequency offset). Traditional prediction methods (e.g., autoregressive differential moving average models, hidden Markov models, and Kalman filters) build models based on linear or weakly nonlinear assumptions, making it difficult to accurately fit complex frequency modulation patterns, resulting in significant prediction errors. Specifically, existing technologies have the following drawbacks:

[0003] 1. Insufficient ability to capture nonlinear features

[0004] Traditional statistical models (such as autoregressive differential moving average models) and shallow machine learning methods (such as hidden Markov models and support vector machines) typically assume that data follows a linear or stationary distribution. However, the frequency variations of frequency-modulated signals often contain strong nonlinear characteristics such as sudden changes and oscillations. For example, the frequency of an exponential frequency-modulated signal increases exponentially over time. Traditional methods are unable to effectively fit this pattern due to model structure limitations (such as linear difference equations or fixed state transition probabilities), resulting in significant deviations between predicted results and the actual signal.

[0005] 2. Difficulty in modeling long-term temporal dependencies

[0006] Frequency modulation sequences can vary over time periods of tens of seconds or even longer (e.g., periodic pulse frequency modulation). Traditional methods (such as Markov models and recurrent neural networks (RNNs)) are limited in modeling long-range temporal relationships. Markov models, due to their "memorylessness" assumption, cannot capture inter-period dependencies. While RNNs can process sequential data, their vanishing gradient problem causes lags or divergence in predictions of long-period frequency modulation patterns, making them difficult to achieve high-precision predictions.

[0007] To overcome the limitations of a single model, existing technologies have proposed combined prediction models (such as multi-model weighted fusion), which superimpose the prediction results of different models to improve overall performance. However, while such solutions partially alleviate the shortcomings of a single model, they also introduce new technical problems:

[0008] 1. Lack of dynamic weight allocation mechanism

[0009] Combination models typically use fixed weights or simple weighting strategies (such as equal-weighted averaging), which are unable to adapt to the non-stationary and random nature of frequency modulation sequences. For example, in the case of frequency mutations, static weights based on historical performance may lead to delayed model response. Introducing dynamic weight optimization (such as adjusting weights based on real-time error feedback) requires additional computing resources and may not meet real-time prediction requirements due to algorithmic latency.

[0010] 2. Conflict between overfitting and generalization ability

[0011] Combination models enhance expressiveness by integrating multiple sub-models. However, when training data is insufficient, the model may overfit to the noise in the training set (such as occasional fluctuations in specific frequency modulation patterns), resulting in reduced generalization to unseen data. Especially in the case of frequency modulation signals, the complex structure of combination models (such as multi-layer neural network fusion) under small sample sizes can exacerbate the risk of overfitting, limiting their practical application value.

[0012] Traditional methods are unable to adapt to complex FM patterns due to limited model assumptions. While combined models attempt to improve performance through multi-model fusion, they struggle to meet the demands of high-precision, real-time predictions due to limitations in weight optimization and generalization capabilities. Therefore, a prediction method is urgently needed that can dynamically adapt to FM signal characteristics while balancing model complexity and generalization capabilities. Summary of the Invention

[0013] In order to overcome the defects of the above-mentioned prior art, the purpose of the present invention is to provide a result-selective combined model FM sequence prediction method and related equipment to solve the technical problems of nonlinear modeling, long-term dependency capture and dynamic adaptation of the combined model in FM signal prediction.

[0014] The present invention is achieved through the following technical solutions:

[0015] In a first aspect, the present invention provides a method for predicting a frequency modulation sequence using a combination model with selected results, comprising:

[0016] Utilize multiple sets of initial preset neural network models to predict the frequency modulation instruction sequence and generate multiple sets of initial prediction results accordingly;

[0017] An average prediction sequence is calculated based on multiple sets of initial prediction results;

[0018] Calculating the deviations between the multiple groups of initial prediction results and the average prediction sequence, and performing a first-level elimination strategy on the multiple groups of initial preset neural network models based on the deviations to obtain a first-level preset neural network model set;

[0019] Calculating the discrimination between each pair of the first-level preset neural network model set, and performing a second-level elimination strategy on the first-level preset neural network model set based on the discrimination to obtain a second-level preset neural network model set;

[0020] Calculate the deviation weights of the secondary preset neural network model set, perform weighted fusion on the secondary preset neural network model set based on the deviation weights, and generate the final prediction sequence.

[0021] Preferably, the multiple groups of initial preset neural network models include at least 5 groups of models selected from the group consisting of Informer model, Autoformer model, N-HiTS model, PatchTST model, TimeGrad model, DLinear model, TFT model, Graph WaveNet model, TimesNet model or TiDE model.

[0022] Preferably, the calculation formula of the average prediction sequence includes:

[0023]

[0024] in, M The number of groups of the initial preset neural network model currently participating in the calculation; For the i The initial prediction results of the initial preset neural network model.

[0025] Preferably, the deviations of multiple groups of initial prediction results from the average prediction sequence are calculated, and a first-level elimination strategy is performed on multiple groups of initial preset neural network models based on the deviations to obtain a first-level preset neural network model set. The calculation formula of the deviations includes:

[0026]

[0027] Where L is the total length of the predicted sequence; For the i The initial prediction results of an initial preset neural network model; is the average prediction sequence; h is the first prediction result of the prediction sequence; N is the number of initial prediction results; For the i The deviation of the neural network model; || is the absolute value;

[0028] The first-level elimination strategy includes:

[0029] Comparing the deviations of multiple groups of initial preset neural network models, deleting the neural network model corresponding to the maximum deviation, and obtaining a first-level preset neural network model set;

[0030] When the maximum deviations in multiple groups of initial preset neural network models are the same, any group of neural network models corresponding to the multiple groups with the same maximum deviations is randomly deleted to obtain a first-level preset neural network model set.

[0031] Preferably, the discrimination between each pair of the first-level preset neural network model set is calculated, and a second-level elimination strategy is performed on the first-level preset neural network model set based on the discrimination to obtain a second-level preset neural network model set. The calculation formula of the discrimination includes:

[0032]

[0033] in, For the i The initial prediction results of an initial preset neural network model; For the j The initial prediction result of the initial preset neural network model; h is the first prediction result of the prediction sequence; N is the number of initial prediction results; For the i and j The discrimination of the initial preset neural network model; || is the absolute value;

[0034] The secondary elimination strategy includes:

[0035] Comparing the discrimination between each pair of the first-level preset neural network model set, and determining the two groups of first-level preset neural network models corresponding to the maximum discrimination among all discriminations;

[0036] Based on the two groups of first-level preset neural network models corresponding to the maximum discrimination, the average values ​​of the discriminations associated with the initial prediction results of the two groups of first-level preset neural network models are calculated respectively, and the group of first-level preset neural network models with the larger average value is deleted to obtain a set of second-level preset neural network models.

[0037] Preferably, the calculation formula of the deviation weight includes:

[0038]

[0039] Among them, S is the set of secondary preset neural network models; For the i The deviation of a neural network model; k is any neural network model; is the weight.

[0040] Preferably, the calculation formula of weighted fusion includes:

[0041]

[0042] in, is the prediction result of the remaining neural network; is the weight; is the weighted fusion result; i Represents the first iAn initial preset neural network model.

[0043] In a second aspect, the present invention further provides a result-selective combined model frequency modulation sequence prediction system, comprising:

[0044] An initial prediction module is used to predict the frequency modulation instruction sequence using multiple sets of initial preset neural network models and generate multiple sets of initial prediction results accordingly;

[0045] An average value calculation processing module is used to calculate an average prediction sequence based on multiple groups of initial prediction results;

[0046] A first-level elimination processing module is used to calculate the deviation between multiple groups of initial prediction results and the average prediction sequence, and to perform a first-level elimination strategy on the multiple groups of initial preset neural network models based on the deviation to obtain a first-level preset neural network model set;

[0047] A secondary elimination processing module is used to calculate the discrimination between each pair of the first-level preset neural network model set, and to perform a secondary elimination strategy on the first-level preset neural network model set based on the discrimination to obtain a second-level preset neural network model set;

[0048] The prediction sequence generation module is used to calculate the deviation weight of the secondary preset neural network model set, perform weighted fusion on the secondary preset neural network model set based on the deviation weight, and generate the final prediction sequence.

[0049] In a third aspect, the present invention also provides a mobile terminal comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the computer program, the result-selective combined model frequency modulation sequence prediction method as described above is implemented.

[0050] In a fourth aspect, the present invention further provides a computer-readable storage medium storing a computer program, which, when executed by a processor, implements the result-selective combined model frequency modulation sequence prediction method as described above.

[0051] Compared with the prior art, the present invention has the following beneficial technical effects:

[0052] The present invention provides a result-selective combined model frequency modulation sequence prediction method. By using multiple sets of preset neural network models to predict the frequency modulation instruction sequence, the method can integrate the nonlinear modeling capabilities of different models; calculating the average prediction sequence of multiple sets of initial prediction results helps to smooth short-term fluctuations and highlight long-term trends. By calculating the deviation between the initial prediction results and the average prediction sequence, and performing a first-level elimination based on the deviation, models that are inaccurate in predicting long-term trends can be eliminated and those models that can better capture long-term dependencies can be retained. On the basis of the first-level elimination, the distinction between the initial prediction results of the remaining models is calculated, and a second-level elimination is performed based on the distinction, which helps to eliminate models whose prediction results are highly similar and lack complementarity, thereby retaining a more diverse and complementary model combination, calculating the deviation weights of the remaining models after the second-level elimination, and performing weighted fusion of the initial prediction results based on these weights, avoiding the limitations of a single model and achieving dynamic adaptation of the combined model.

[0053] Furthermore, by calculating the average prediction sequence from multiple sets of initial prediction results, it is possible to take into account both short-term and long-term prediction needs to a certain extent, thereby improving the overall prediction accuracy.

[0054] Furthermore, the deviation measures the difference between the model's predictions and the average prediction sequence. The greater the deviation, the worse the model's ability to capture the data distribution or trend. By eliminating models with the largest deviations, we can eliminate models that have a low contribution to the overall forecast or even interfere with it, thereby improving the stability of the combined model.

[0055] Furthermore, by calculating the discriminability between the initial predictions of the models, the two-stage elimination strategy can identify models with highly similar predictions. When determining which models to eliminate, the average discriminability values ​​of the two groups of models are compared. The model group with the larger average value indicates that its predictions differ more significantly from the other models. By eliminating redundant models, the two-stage elimination strategy reduces the negative impact of low-quality models on the weighted fusion results, thereby improving the accuracy of the final predictions. BRIEF DESCRIPTION OF THE DRAWINGS

[0056] Figure 1 Flowchart of the combined model frequency modulation sequence prediction method according to an embodiment of the present invention;

[0057] Figure 2 This is a flow chart of the first-level elimination strategy in an embodiment of the present invention;

[0058] Figure 3 Flowchart of the secondary elimination strategy in an embodiment of the present invention;

[0059] Figure 4 Schematic diagram of the principle structure of the combined model frequency modulation sequence prediction system in an embodiment of the present invention;

[0060] In the figure: 1. Initial prediction module; 2. Average value calculation processing module; 3. First-level elimination processing module; 4. Second-level elimination processing module; 5. Prediction sequence generation module. DETAILED DESCRIPTION

[0061] In order to enable those skilled in the art to better understand the solutions of the present invention, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the embodiments described are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts should fall within the scope of protection of the present invention.

[0062] The purpose of the present invention is to provide a result-selective combined model FM sequence prediction method and related equipment to solve the technical problems of nonlinear modeling, long-term dependency capture and dynamic adaptation of the combined model in FM signal prediction.

[0063] The present invention is described in further detail below with reference to the accompanying drawings:

[0064] Example 1

[0065] See also Figure 1 In one embodiment of the present invention, a method for predicting a frequency modulation sequence using a combination model with selected results is provided, comprising:

[0066] Step 1: Use multiple sets of initial preset neural network models to predict the frequency modulation instruction sequence and generate multiple sets of initial prediction results accordingly;

[0067] Specifically, the multiple groups of initial preset neural network models include at least 5 groups of models among the Informer model, Autoformer model, N-HiTS model, PatchTST model, TimeGrad model, DLinear model, TFT model, Graph WaveNet model, TimesNet model or TiDE model.

[0068] The Informer model is used as an improved model based on Transformer, and a "ProbSparse" attention mechanism is proposed to significantly reduce computational complexity and is suitable for long sequence prediction.

[0069] The Autoformer model is used to introduce an auto-correlation mechanism to replace traditional attention and improve prediction accuracy by periodic decomposition of sequences.

[0070] The N-HiTS model is used in a multi-layer hierarchical architecture and combined with multi-scale interpolation technology to improve long-term prediction efficiency.

[0071] The PatchTST model is used to segment time series into local segments (Patch), drawing on the ViT idea to enhance the ability to capture local features.

[0072] The TimeGrad model is used to generate probability predictions through step-by-step denoising based on a diffusion model.

[0073] The DLinear model is used for a simple linear layer architecture, challenging complex models through sequence decomposition (trend + residual), emphasizing the "simplicity advantage".

[0074] The TFT model (Temporal Fusion Transformer model) is used to fuse LSTM and Transformer, supporting multimodal modeling of static features, known future inputs, and dynamic features.

[0075] The Graph WaveNet model is used to combine graph neural networks with dilated causal convolution to model spatiotemporal dependencies.

[0076] The TimesNet model is used to convert time series into two-dimensional space and use CNN to capture multi-cycle changes.

[0077] The TiDE model is used in an MLP-based encoder-decoder architecture to achieve high-speed inference through feature densification.

[0078] Step 2: Calculate the average prediction sequence based on multiple sets of initial prediction results;

[0079] Specifically, the calculation formula of the average prediction sequence includes:

[0080]

[0081] in, M The number of groups of the initial preset neural network model currently participating in the calculation; For the i The initial prediction results of the initial preset neural network model.

[0082] Step 3: Calculate the deviation between the multiple groups of initial prediction results and the average prediction sequence, and perform a first-level elimination strategy on the multiple groups of initial preset neural network models based on the deviation to obtain a first-level preset neural network model set;

[0083] Specifically, the calculation formula of the deviation includes:

[0084]

[0085] Where L is the total length of the predicted sequence; For the i The initial prediction results of an initial preset neural network model; is the average prediction sequence; h is the first prediction result of the prediction sequence; N is the number of initial prediction results; For the i The deviation of the neural network model; || is the absolute value;

[0086] according to Figure 2 As shown, the first-level elimination strategy includes:

[0087] Step 31, comparing the deviations of multiple groups of initial preset neural network models, deleting the neural network model corresponding to the maximum deviation, and obtaining a first-level preset neural network model set;

[0088] Step 32: When the maximum deviations in multiple groups of initial preset neural network models are the same, any group of neural network models corresponding to the same maximum deviations is randomly deleted to obtain a first-level preset neural network model set.

[0089] Step 4, calculating the discrimination between each pair of the first-level preset neural network model set, and performing a second-level elimination strategy on the first-level preset neural network model set based on the discrimination to obtain a second-level preset neural network model set;

[0090] Specifically, the calculation formula of the distinctiveness includes:

[0091]

[0092] in, For the i The initial prediction results of an initial preset neural network model; For the j The initial prediction result of the initial preset neural network model; h is the first prediction result of the prediction sequence; N is the number of initial prediction results; For the i and j The discrimination of the initial preset neural network model; || is the absolute value;

[0093] according to Figure 3 As shown, the secondary elimination strategy includes:

[0094] Step 41, comparing the discrimination between each pair of the first-level preset neural network model set, and determining the two sets of first-level preset neural network models corresponding to the maximum discrimination among all the discriminations;

[0095] Step 42, based on the two groups of first-level preset neural network models corresponding to the maximum discrimination, respectively calculate the average values ​​of the discriminations associated with the initial prediction results of the two groups of first-level preset neural network models, and delete the group of first-level preset neural network models with the larger average value to obtain a set of second-level preset neural network models.

[0096] Step 5: Calculate the deviation weight of the secondary preset neural network model set, perform weighted fusion on the secondary preset neural network model set based on the deviation weight, and generate the final prediction sequence.

[0097] Specifically, the calculation formula of the deviation weight includes:

[0098]

[0099] Among them, S is the set of secondary preset neural network models; For the i The deviation of a neural network model; k is any neural network model; is the weight.

[0100] Specifically, the calculation formula of weighted fusion includes:

[0101]

[0102] in, is the prediction result of the remaining neural network; is the weight; is the weighted fusion result; i Represents the first i An initial preset neural network model.

[0103] According to this embodiment, the frequency modulation instruction is set to Pt=[X1, X2, X3, .X i ..,X N ] (N=1000);

[0104] With [X1,X2,X3,.X i ..,X N ] as input, predict the future sequence of length 0.1N, that is, when N=1000, [X1,X2,X3,.X i ..,X 1000 ] into the neural network to predict the future [X 1001 ,X 1002 ,X 1003 ,...,X 1100 ] value.

[0105] The prediction result of the first neural network model Informer is [X 1001（1） ,X 1002 （1） ,X 1003 （1） ,...,X 1100 （1） ];

[0106] The prediction result of the second neural network model Autoformer is [X 1001 （2） ,X 1002 （2） ,X 1003 （2） ,...,X 1100 （2） ];

[0107] The prediction result of the third neural network model N-HiTS is [X 1001 （3） ,X 1002 （3） ,X 1003 （3） ,...,X 1100 （3） ];

[0108] The prediction result of the i-th neural network model is [X 1001 （i） ,X 1002 （i） ,X 1003 （i） ,...,X 1100 （i） ];

[0109] The prediction result of the 10th neural network model TiDE is [X 1001 （10） ,X 1002 （10） ,X 1003 （10） ,...,X 1100 （10） ].

[0110] Assume that the average prediction result of 10 neural network models is: , , ,..., ].

[0111] by For example, the specific calculation method is: = The same applies to other numerical calculations.

[0112] Taking the prediction result of the first neural network as an example, the calculation method of the deviation D1 of the first neural network is:

[0113]

[0114] According to this method, [D1, D2, D3, ..., D 10 ];

[0115] After calculating the deviation, the FM sequence with the largest deviation indicates the worst prediction effect, so it is eliminated and deleted, that is, for example, [D1, D2, D3, ..., D 10 ], the value of D5 is the largest, then the prediction result of the 5th neural network TimeGrad [X 1001 （5） ,X 1002 （5） ,X 1003 （5） ,...,X 1100 （5） ]Abandoned in the follow-up and no longer involved in the follow-up.

[0116] The remaining 9 initial prediction results are used to calculate the discrimination Z between each two. The greater the discrimination, the greater the difference between the two prediction results, and the accuracy of at least one of the prediction results is poor. 12 The calculation formula is as follows:

[0117]

[0118] According to the above formula, the discrimination between the remaining 9 initial prediction results is calculated, which is 36 in total.

[0119] 36 discriminations were calculated, namely [Z 12 ,Z 13 ,Z 14 ,...,Z 89 ]; find [Z 12 ,Z 13 ,Z 14 ,...,Z 89 ], assuming the maximum value is Z 67 , it means that there is a difference in the accuracy of the prediction results between the initial prediction result 6 and the initial prediction result 7, where the discrimination of the initial prediction result 6 is [Z 16 ,Z 26 ,Z 36 ,Z 46 , Z 67， Z 68 ,Z 69], the discrimination of the initial prediction results 7 are [Z 17 ,Z 27 ,Z 37 ,Z 47 , Z 67， Z 78 ,Z 79 ]. Compare the sizes of the two groups of average values. If the average value of the initial prediction result 6 is greater than the average value of the initial prediction result 7, the initial prediction result 6 will be eliminated and deleted.

[0120] According to the above, the initial prediction result of the fifth preset neural network model is deleted in the first-level elimination strategy, and the initial prediction result of the sixth preset neural network model is deleted in the second-level elimination strategy. The final prediction result is [X 1001 ,X 1002 ,X 1003 ,...,X 1100 ]

[0121] X 1001 Explain how the final sequence is generated.

[0122]

[0123] The other results are similar.

[0124] The present invention provides a result-selective combined model frequency modulation sequence prediction method, which predicts the frequency modulation instruction sequence by utilizing multiple sets of preset neural network models. The method can integrate the nonlinear modeling capabilities of different models; the average prediction sequence of multiple sets of initial prediction results is calculated, which helps to smooth short-term fluctuations and highlight long-term trends. By calculating the deviation between the initial prediction results and the average prediction sequence, and performing a first-level elimination based on the deviation, models that are inaccurate in predicting long-term trends can be eliminated and those models that can better capture long-term dependencies can be retained. On the basis of the first-level elimination, the distinction between the initial prediction results of the remaining models is calculated, and a second-level elimination is performed based on the distinction, which helps to eliminate models whose prediction results are highly similar and lack complementarity, thereby retaining a more diverse and complementary model combination. The deviation weights of the remaining models after the second-level elimination are calculated, and the initial prediction results are weighted and fused based on these weights to avoid the limitations of a single model and achieve dynamic adaptation of the combined model.

[0125] Example 2

[0126] according to Figure 4 As shown, the present invention also provides a result-selective combined model frequency modulation sequence prediction system, comprising:

[0127] Initial prediction module 1, used to predict the frequency modulation instruction sequence using multiple sets of initial preset neural network models and generate multiple sets of initial prediction results accordingly;

[0128] An average value calculation processing module 2 is used to calculate an average prediction sequence based on multiple groups of initial prediction results;

[0129] A first-level elimination processing module 3 is used to calculate the deviation between multiple groups of initial prediction results and the average prediction sequence, and to perform a first-level elimination strategy on multiple groups of initial preset neural network models based on the deviation to obtain a first-level preset neural network model set;

[0130] Secondary elimination processing module 4 is used to calculate the discrimination between each pair of the first-level preset neural network model set, and perform a secondary elimination strategy on the first-level preset neural network model set based on the discrimination to obtain a secondary preset neural network model set;

[0131] The prediction sequence generation module 5 is used to calculate the deviation weight of the secondary preset neural network model set, perform weighted fusion on the secondary preset neural network model set based on the deviation weight, and generate a final prediction sequence.

[0132] Example 3

[0133] The present invention also provides a mobile terminal comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, such as a result-selective combined model frequency modulation sequence prediction program.

[0134] When the processor executes the computer program, the above-mentioned result-selective combined model frequency modulation sequence prediction method is implemented, for example:

[0135] Utilize multiple sets of initial preset neural network models to predict the frequency modulation instruction sequence and generate multiple sets of initial prediction results accordingly;

[0136] An average prediction sequence is calculated based on multiple sets of initial prediction results;

[0137] Calculating the deviations between the multiple groups of initial prediction results and the average prediction sequence, and performing a first-level elimination strategy on the multiple groups of initial preset neural network models based on the deviations to obtain a first-level preset neural network model set;

[0138] Calculating the discrimination between each pair of the first-level preset neural network model set, and performing a second-level elimination strategy on the first-level preset neural network model set based on the discrimination to obtain a second-level preset neural network model set;

[0139] Calculate the deviation weights of the secondary preset neural network model set, perform weighted fusion on the secondary preset neural network model set based on the deviation weights, and generate the final prediction sequence.

[0140] Alternatively, when the processor executes the computer program, the functions of each module in the above system are realized, for example:

[0141] Initial prediction module 1, used to predict the frequency modulation instruction sequence using multiple sets of initial preset neural network models and generate multiple sets of initial prediction results accordingly;

[0142] An average value calculation processing module 2 is used to calculate an average prediction sequence based on multiple groups of initial prediction results;

[0143] A first-level elimination processing module 3 is used to calculate the deviation between multiple groups of initial prediction results and the average prediction sequence, and to perform a first-level elimination strategy on multiple groups of initial preset neural network models based on the deviation to obtain a first-level preset neural network model set;

[0144] Secondary elimination processing module 4 is used to calculate the discrimination between each pair of the first-level preset neural network model set, and perform a secondary elimination strategy on the first-level preset neural network model set based on the discrimination to obtain a secondary preset neural network model set;

[0145] The prediction sequence generation module 5 is used to calculate the deviation weight of the secondary preset neural network model set, perform weighted fusion on the secondary preset neural network model set based on the deviation weight, and generate a final prediction sequence.

[0146] Exemplarily, the computer program may be divided into one or more modules / units, which are stored in the memory and executed by the processor to implement the present invention. The one or more modules / units may be a series of computer program instruction segments capable of implementing specific functions, and the instruction segments are used to describe the execution process of the computer program in the mobile terminal.

[0147] For example, the computer program may be divided into an initial prediction module 1, an average value calculation processing module 2, a first-level elimination processing module 3, a second-level elimination processing module 4, and a prediction sequence generation module 5;

[0148] The specific functions of each module are as follows:

[0149] Initial prediction module 1, used to predict the frequency modulation instruction sequence using multiple sets of initial preset neural network models and generate multiple sets of initial prediction results accordingly;

[0150] An average value calculation processing module 2 is used to calculate an average prediction sequence based on multiple groups of initial prediction results;

[0151] A first-level elimination processing module 3 is used to calculate the deviation between multiple groups of initial prediction results and the average prediction sequence, and to perform a first-level elimination strategy on multiple groups of initial preset neural network models based on the deviation to obtain a first-level preset neural network model set;

[0152] Secondary elimination processing module 4 is used to calculate the discrimination between each pair of the first-level preset neural network model set, and perform a secondary elimination strategy on the first-level preset neural network model set based on the discrimination to obtain a secondary preset neural network model set;

[0153] The prediction sequence generation module 5 is used to calculate the deviation weight of the secondary preset neural network model set, perform weighted fusion on the secondary preset neural network model set based on the deviation weight, and generate a final prediction sequence.

[0154] The mobile terminal may be a computing device such as a desktop computer, a notebook, a PDA, a cloud server, etc. The mobile terminal may include, but is not limited to, a processor and a memory.

[0155] The processor may be a central processing unit (CPU), other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor may be a microprocessor or any conventional processor, etc. The processor is the control center of the mobile terminal, connecting various parts of the entire mobile terminal using various interfaces and lines.

[0156] The memory may be used to store the computer programs and / or modules, and the processor implements various functions of the mobile terminal by running or executing the computer programs and / or modules stored in the memory and calling the data stored in the memory.

[0157] The memory may primarily include a program storage area and a data storage area. The program storage area may store an operating system and at least one application required for a function (such as sound playback or image playback); the data storage area may store data generated based on the use of the mobile phone (such as audio data and a phone book). Furthermore, the memory may include high-speed random access memory (RAM) and non-volatile memory, such as a hard disk, internal memory, a plug-in hard disk, a Smart Media Card (SMC), a Secure Digital (SD) card, a Flash Card, at least one disk storage device, a flash memory device, or other volatile solid-state storage device.

[0158] Example 4

[0159] The present invention also provides a computer-readable storage medium storing a computer program, wherein the computer program, when executed by a processor, implements the result-selective combined model frequency modulation sequence prediction method.

[0160] If the module / unit integrated in the mobile terminal is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium.

[0161] Based on this understanding, the present invention can implement all or part of the processes in the above-mentioned method by means of a computer program instructing related hardware. The computer program can be stored in a computer-readable storage medium. When executed by a processor, the computer program can implement the above-mentioned aggregated reinforcement learning resource scheduling method. The computer program includes computer program code, which can be in source code form, object code form, executable file, or some intermediate form.

[0162] The computer-readable medium may include: any entity or device that can carry the computer program code, recording medium, USB flash drive, mobile hard disk, magnetic disk, optical disk, computer memory, read-only memory (ROM), random access memory (RAM), electrical carrier signal, telecommunication signal and software distribution medium, etc.

[0163] It should be noted that the content contained in the computer-readable medium can be appropriately increased or decreased according to the requirements of legislation and patent practices in the jurisdiction. For example, in some jurisdictions, according to legislation and patent practices, computer-readable media do not include electrical carrier signals and telecommunication signals.

[0164] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit it. Although the present invention has been described in detail with reference to the above embodiments, ordinary technicians in the field should understand that the specific implementation methods of the present invention can still be modified or replaced by equivalents. Any modification or equivalent replacement that does not depart from the spirit and scope of the present invention should be covered by the scope of protection of the present invention.

Claims

1. A method for predicting frequency modulation sequences using a combination model with selective results, characterized in that: include: Utilize multiple sets of initial preset neural network models to predict the frequency modulation instruction sequence and generate multiple sets of initial prediction results accordingly; An average prediction sequence is calculated based on multiple sets of initial prediction results; Calculating the deviations between the multiple groups of initial prediction results and the average prediction sequence, and performing a first-level elimination strategy on the multiple groups of initial preset neural network models based on the deviations to obtain a first-level preset neural network model set; Calculating the discrimination between each pair of the first-level preset neural network model set, and performing a second-level elimination strategy on the first-level preset neural network model set based on the discrimination to obtain a second-level preset neural network model set; The calculation formula of the discrimination degree is include: in, For the i The initial prediction results of an initial preset neural network model; For the j The initial prediction result of the initial preset neural network model; h is the first prediction result of the prediction sequence; N is the number of initial prediction results; For the i and j The discrimination of the initial preset neural network model; || is the absolute value; The secondary elimination strategy includes: Comparing the discrimination between each pair of the first-level preset neural network model set, and determining the two groups of first-level preset neural network models corresponding to the maximum discrimination among all discriminations; Based on the two groups of first-level preset neural network models corresponding to the maximum discrimination, the average values ​​of the discriminations associated with the initial prediction results of the two groups of first-level preset neural network models are calculated respectively, and the group of first-level preset neural network models with the larger average value is deleted to obtain a set of second-level preset neural network models; Calculate the deviation weights of the secondary preset neural network model set, perform weighted fusion on the secondary preset neural network model set based on the deviation weights, and generate the final prediction sequence.

2. The result-selective combined model FM sequence prediction method according to claim 1, characterized in that: The multiple groups of initial preset neural network models include at least five groups of models selected from the group consisting of an Informer model, an Autoformer model, an N-HiTS model, a PatchTST model, a TimeGrad model, a DLinear model, a TFT model, a Graph WaveNet model, a TimesNet model, and a TiDE model.

3. The method for predicting FM sequences with a combination model and selective results according to claim 1, wherein: The calculation formula of the average prediction sequence includes: in, M The number of groups of the initial preset neural network model currently participating in the calculation; For the i The initial prediction results of the initial preset neural network model.

4. The method for predicting a frequency modulation sequence with a combination model and selective results according to claim 1, wherein: The deviation between the multiple groups of initial prediction results and the average prediction sequence is calculated, and the multiple groups of initial preset neural network models are subjected to a first-level elimination strategy based on the deviation to obtain a first-level preset neural network model set. The calculation formula of the deviation is include: Where L is the total length of the predicted sequence; For the i The initial prediction results of an initial preset neural network model; is the average prediction sequence; h is the first prediction result of the prediction sequence; N is the number of initial prediction results; For the i The deviation of the neural network model; || is the absolute value; The first-level elimination strategy includes: Comparing the deviations of multiple groups of initial preset neural network models, deleting the neural network model corresponding to the maximum deviation, and obtaining a first-level preset neural network model set; When the maximum deviations in multiple groups of initial preset neural network models are the same, any group of neural network models corresponding to the multiple groups with the same maximum deviations is randomly deleted to obtain a first-level preset neural network model set.

5. The method for predicting frequency modulation sequences using a combination model with selective results according to claim 1, wherein: The calculation formula of the deviation weight includes: Among them, S is the set of secondary preset neural network models; For the i The deviation of a neural network model; k is any neural network model; is the weight.

6. The method for predicting frequency modulation sequences with a combination model and selective results according to claim 1, wherein: The calculation formula of the weighted fusion includes: in, is the prediction result of the remaining neural network; is the weight; is the weighted fusion result; i Represents the first i An initial preset neural network model.

7. A result-selective combined model frequency modulation sequence prediction system, characterized in that: include: An initial prediction module is used to predict the frequency modulation instruction sequence using multiple sets of initial preset neural network models and generate multiple sets of initial prediction results accordingly; An average value calculation processing module is used to calculate an average prediction sequence based on multiple groups of initial prediction results; A first-level elimination processing module is used to calculate the deviation between multiple groups of initial prediction results and the average prediction sequence, and to perform a first-level elimination strategy on the multiple groups of initial preset neural network models based on the deviation to obtain a first-level preset neural network model set; A secondary elimination processing module is used to calculate the discrimination between each pair of the first-level preset neural network model set, and to perform a secondary elimination strategy on the first-level preset neural network model set based on the discrimination to obtain a second-level preset neural network model set; The calculation formula of the discrimination degree is include: in, For the i The initial prediction results of an initial preset neural network model; For the j The initial prediction result of the initial preset neural network model; h is the first prediction result of the prediction sequence; N is the number of initial prediction results; For the i and j The discrimination of the initial preset neural network model; || is the absolute value; The secondary elimination strategy includes: Comparing the discrimination between each pair of the first-level preset neural network model set, and determining the two groups of first-level preset neural network models corresponding to the maximum discrimination among all discriminations; Based on the two groups of first-level preset neural network models corresponding to the maximum discrimination, the average values ​​of the discriminations associated with the initial prediction results of the two groups of first-level preset neural network models are calculated respectively, and the group of first-level preset neural network models with the larger average value is deleted to obtain a set of second-level preset neural network models; The prediction sequence generation module is used to calculate the deviation weight of the secondary preset neural network model set, perform weighted fusion on the secondary preset neural network model set based on the deviation weight, and generate the final prediction sequence.

8. A mobile terminal comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein: When the processor executes the computer program, the result-selective combined model frequency modulation sequence prediction method according to any one of claims 1 to 6 is implemented.

9. A computer-readable storage medium storing a computer program, characterized in that: When the computer program is executed by a processor, the result-selective combined model frequency modulation sequence prediction method according to any one of claims 1 to 6 is implemented.

Images