Runoff prediction method, device and equipment based on modal decomposition and bidirectional long-short time neural network, and medium

Abstract

The invention discloses a runoff prediction method, device and equipment based on modal decomposition and a bidirectional long-short term neural network and a medium, and relates to the field of runoff prediction, and the method comprises the steps: decomposing an original monthly runoff sequence based on a complete ensemble empirical mode decomposition method of adaptive noise; obtaining a residual sequence and a plurality of different intrinsic mode function component sequences; inputting the intrinsic mode function component sequence into the optimized bidirectional long-short time neural network to obtain a prediction sequence of intrinsic mode function components; decomposing the residual sequence based on a variational mode decomposition method to obtain a plurality of different intrinsic mode function component sequences; inputting the intrinsic mode function component sequence into the optimized bidirectional long-short time neural network to obtain a prediction sequence of intrinsic mode function components; and adding and reconstructing the prediction sequences of the intrinsic mode function component and the intrinsic mode function component to obtain a monthly runoff sequence in a future period of time. The runoff prediction accuracy is improved.

Description

Technical Field

[0001] The present application relates to the field of runoff prediction, and in particular to a runoff prediction method, device, equipment and medium based on modal decomposition and bidirectional long-short-time neural network. Background Art

[0002] Accurate runoff prediction can provide scientific guidance for flood control and drought relief, water resources management and efficient utilization, and ecological and environmental protection. Under the interference of climate change and human activities, runoff exhibits complex characteristics such as non-stationarity and randomness, making it difficult for a single prediction model to obtain satisfactory prediction results. A large number of studies have shown that stabilizing the runoff series is an effective way to improve prediction accuracy. There are many methods for stabilizing runoff series, such as empirical mode decomposition (EMD) and local mean decomposition (LMD). However, traditional empirical mode decomposition and local mean decomposition methods have problems such as mode aliasing and signal mutation. Summary of the invention

[0003] The purpose of this application is to provide a runoff prediction method, device, equipment and medium based on modal decomposition and bidirectional long-short-time neural network, which can reduce modal aliasing and signal mutation and improve the accuracy of runoff prediction.

[0004] To achieve the above objectives, this application provides the following solutions:

[0005] In a first aspect, the present application provides a runoff prediction method based on modal decomposition and bidirectional long-short time neural network, comprising:

[0006] The original monthly runoff sequence is decomposed by a complete ensemble empirical mode decomposition method based on adaptive noise to obtain a residual sequence and several different intrinsic mode function component sequences; the original monthly runoff sequence includes runoff data for a period of history;

[0007] Inputting the plurality of different intrinsic mode function component sequences into the optimized bidirectional long-short time neural network respectively to obtain the prediction sequences of the plurality of different intrinsic mode function components;

[0008] Decomposing the residual sequence based on the variational mode decomposition method to obtain a number of different intrinsic mode function component sequences; inputting the several different intrinsic mode function component sequences into the optimized bidirectional long-short time neural network respectively to obtain a number of different intrinsic mode function component prediction sequences;

[0009] The prediction sequences of the several different intrinsic mode function components and the prediction sequences of the several different intrinsic mode function components are added and reconstructed to obtain a monthly runoff sequence for a period of time in the future.

[0010] In a second aspect, the present application provides a runoff prediction device based on modal decomposition and bidirectional long-short time neural network, comprising:

[0011] The first data decomposition module is used to decompose the original monthly runoff sequence based on the complete set empirical mode decomposition method with adaptive noise to obtain a residual sequence and a plurality of different intrinsic mode function component sequences; the original monthly runoff sequence includes runoff data for a period of history;

[0012] A first model prediction module is used to input the plurality of different intrinsic modal function component sequences into the optimized bidirectional long-short time neural network respectively to obtain prediction sequences of the plurality of different intrinsic modal function components;

[0013] A second data decomposition module is used to decompose the residual sequence based on a variational mode decomposition method to obtain a plurality of different intrinsic mode function component sequences;

[0014] A second model prediction module is used to input the plurality of different intrinsic mode function component sequences into the optimized bidirectional long-short time neural network respectively to obtain prediction sequences of the plurality of different intrinsic mode function components;

[0015] The data reconstruction module is used to add and reconstruct the prediction sequences of the several different intrinsic mode function components and the prediction sequences of the several different intrinsic mode function components to obtain a monthly runoff sequence for a period of time in the future.

[0016] In a third aspect, the present application provides a computer device, comprising: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the above-mentioned runoff prediction method based on modal decomposition and bidirectional long-short-time neural network.

[0017] In a fourth aspect, the present application provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the above-mentioned runoff prediction method based on modal decomposition and bidirectional long-short-time neural network.

[0018] According to the specific embodiments provided in this application, this application discloses the following technical effects:

[0019] The present application provides a runoff prediction method, device, equipment, medium and product based on modal decomposition and bidirectional long-short time neural network. The original monthly runoff sequence is decomposed by the complete set empirical mode decomposition method of adaptive noise to obtain a residual sequence and several different intrinsic mode function component sequences; the residual sequence is further decomposed by the variational mode decomposition method to obtain several different intrinsic mode function component sequences. The two different decomposition methods are used to effectively reduce modal aliasing and signal mutation. And the optimized bidirectional long-short time neural network is used to perform multiple predictions on several different intrinsic mode function component sequences and several different intrinsic mode function component sequences, so that the runoff prediction is more accurate. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the drawings required for use in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0021] Figure 1 This is a flow chart of a runoff prediction method based on modal decomposition and bidirectional long-short time neural network in one embodiment of the present application;

[0022] Figure 2 A schematic diagram of a process of optimizing a bidirectional long-short time neural network using a particle swarm algorithm with adaptive adjustment of inertia weights provided in an embodiment of the present application;

[0023] Figure 3 A schematic diagram of the structure of a computer device provided in one embodiment of the present application. DETAILED DESCRIPTION

[0024] The following will be combined with the drawings in the embodiments of the present application to clearly and completely describe the technical solutions in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.

[0025] In order to make the purpose, features and advantages of the present application more obvious and easy to understand, the present application is further described in detail below in conjunction with the accompanying drawings and specific implementation methods.

[0026] In an exemplary embodiment, Figure 1As shown, a runoff prediction method based on modal decomposition and bidirectional long-short-time neural network is provided, which is executed by a computer device, specifically, it can be executed by a computer device such as a terminal or a server alone, or it can be executed by a terminal and a server together. In the embodiment of the present application, the method includes the following steps 101 to 104. Among them:

[0027] Step 101, decomposing the original monthly runoff sequence based on the complete ensemble empirical mode decomposition method with adaptive noise to obtain a residual sequence and a plurality of different intrinsic mode function component sequences; the original monthly runoff sequence includes runoff data for a period of history.

[0028] Step 102, inputting the plurality of different intrinsic modal function component sequences into the optimized bidirectional long-short time neural network respectively to obtain the prediction sequences of the plurality of different intrinsic modal function components.

[0029] Step 103, decomposing the residual sequence based on the variational mode decomposition method to obtain several different intrinsic mode function component sequences; inputting the several different intrinsic mode function component sequences into the optimized bidirectional long-short time neural network respectively to obtain several different prediction sequences of the intrinsic mode function components.

[0030] Step 104 , the prediction sequences of the several different intrinsic mode function components and the prediction sequences of the several different intrinsic mode function components are added and reconstructed to obtain a monthly runoff sequence for a period of time in the future.

[0031] In addition, before the original monthly runoff sequence is decomposed by the complete ensemble empirical mode decomposition method based on adaptive noise to obtain a residual sequence and several different intrinsic mode function component sequences, the runoff prediction method based on mode decomposition and bidirectional long-short-time neural network also includes: normalizing the runoff data of a historical period to obtain processed historical runoff data, and inputting the processed historical runoff data into the bidirectional long-short-time neural network to obtain runoff data for a future period of time.

[0032] The maximum and minimum values ​​in the runoff data for the future period of time are selected and spliced ​​with the processed historical runoff data to obtain the original monthly runoff sequence.

[0033] As an optional implementation, the original monthly runoff series is decomposed based on the ensemble empirical mode decomposition method (EEMD) to obtain a residual sequence and several different intrinsic mode function component sequences, specifically including:

[0034] Step 201, for the first iteration, add the white noise signal v(t) to the original runoff sequence x(t) to obtain the first monthly runoff sequence; the first monthly runoff sequence is x1(t)=x(t)+v1(t); perform empirical mode decomposition (EMD) on the first monthly runoff sequence x1(t) to obtain several different intrinsic mode function component sequences and a residual sequence r1(t); at this time, the first monthly runoff sequence x1(t) is: Among them, h j (t) is the jth intrinsic mode function component sequence.

[0035] Step 202: for the i-th iteration (i=1, 2, ..., k), the signal x to be decomposed is i (t) Add white noise v of different amplitudes i (t), and obtain the second monthly runoff sequence x i (t) = x(t) + v i (t); for the second monthly runoff sequence x i (t) Perform EMD decomposition to obtain several different intrinsic mode function component sequences and a residual sequence r i (t); At this time, the second monthly runoff sequence x i (t) is: Among them, h i,j (t) is the jth intrinsic mode function component sequence of the i-th iteration, r i (t) is the residual sequence of the i-th iteration.

[0036] Step 203: for the intrinsic mode function component sequence h of the same index in the k iterations, i,j (t) averaging to obtain a final intrinsic modal function component sequence, wherein the final intrinsic modal function component sequence C j (t) is:

[0037] Step 204, using all the final intrinsic mode function component sequences C j (t) and the averaged residual component r(t) to reconstruct the original runoff sequence, the original monthly runoff sequence x(t) is: Among them, the residual component after averaging is

[0038] In another exemplary embodiment of the present application, the original monthly runoff sequence is decomposed based on the complete set empirical mode decomposition method with adaptive noise to obtain a residual sequence and several different intrinsic mode function component sequences, and the above step 101 is replaced by the following steps 301 to 303.

[0039] Step 301, for the i-th iteration, different white noises are added to the original monthly runoff sequence, and the original monthly runoff sequence with different white noises added is decomposed N times using the empirical mode decomposition method to obtain the intrinsic mode function component sequence of the i-th iteration; 1≤i.

[0040] Step 302: obtaining an i-th iteration residual sequence based on the original monthly runoff sequence and the i-th iteration intrinsic mode function component sequence.

[0041] Step 303, determining whether the residual sequence of the i-th iteration can be decomposed, if so, taking the residual sequence of the i-th iteration as the original monthly runoff sequence, and performing the i+1-th iteration.

[0042] If not, the residual sequence of the i-th iteration is taken as the final residual sequence, and the intrinsic mode function component sequence of the 1st iteration to the intrinsic mode function component sequence of the i-th iteration are taken as several different intrinsic mode function component sequences.

[0043] Among them, the intrinsic modal function component sequence includes the intrinsic modal function components obtained by each decomposition. The intrinsic modal function component sequence F1(t) of the first iteration is: Where N is the number of decompositions, x(t) is the original monthly runoff sequence, and v1(t) is the white noise added in the first iteration. is the first intrinsic mode function component obtained by the nth decomposition. The residual sequence r1(t) of the first iteration is: r1(t)=x(t)-F1(t).

[0044] The intrinsic mode function component sequence F2(t) of the second iteration is: Wherein, v2(t) is the white noise added in the second iteration. The residual sequence r2(t) of the second iteration is: r2(t)=r1(t)-F2(t).

[0045] The intrinsic mode function component sequence of the i-th iteration is: Among them, F i (t) is the intrinsic mode function component sequence of the i-th iteration, N is the number of decompositions, is the first intrinsic mode function component obtained by the nth decomposition, r i-1 (t) is the original monthly runoff sequence of the i-1th iteration, v i (t) is the white noise added in the i-th iteration. When i=1, r0(t)=x(t).

[0046] The residual sequence of the i-th iteration is: r i (t) = r i-1 (t)-F i(t); where r i (t) is the residual sequence of the i-th iteration, r i-1 (t) is the residual sequence of the i-1th iteration.

[0047] In another exemplary embodiment of the present application, the several different intrinsic modal function component sequences are respectively input into the optimized bidirectional long-short time neural network, specifically including: truncating the predicted parts of the several different intrinsic modal function component sequences, and inputting the remaining parts of the several different intrinsic modal function component sequences into the optimized bidirectional long-short time neural network; the predicted parts of the several different intrinsic modal function component sequences are the parts after the sudden change of runoff data in the several different intrinsic modal function component sequences.

[0048] In another exemplary embodiment of the present application, the residual sequence is decomposed based on a variational mode decomposition method, specifically comprising: truncating a predicted part of the residual sequence, and decomposing the remaining part of the residual sequence based on a variational mode decomposition method; the predicted part of the residual sequence is the part of the residual sequence after a sudden change in runoff data.

[0049] In another exemplary embodiment of the present application, variational mode decomposition (VMD) is a method for processing non-stationary signal decomposition. The number of modal decompositions of the selected residual sequence is determined according to actual conditions, the residual sequence is split into k intrinsic mode functions with finite bandwidth, and the sum of the bandwidth estimates of all modes is minimized to obtain the corresponding runoff modal component signal and related parameters.

[0050] The residual sequence is decomposed based on the variational modal decomposition method to obtain several different intrinsic mode function component sequences, which specifically includes: constructing a variational modal constraint model and initializing the modal components and center frequencies in the variational modal constraint model. The time domain and frequency domain representations of each modal component are updated through an iterative process, and the center frequency is re-estimated based on the power spectrum of each modal component. A certain iteration accuracy is set, and the iteration is stopped when the change of the modal component is less than a preset threshold. Otherwise, the iteration continues, and each modal component and center frequency are updated, and finally several different intrinsic mode function component sequences are output. The variational modal constraint model is Among them, u k is the decomposed modal component, ω k is the center frequency corresponding to the decomposed modal component; k is the number of modes; f is the residual sequence; δ(t) is the Dirac function; * is the convolution operation; is the gradient operation.

[0051] In another exemplary embodiment of the present application, the Long Short-Term Memory (LSTM) network is a special recurrent neural network that adds a "processor" to the algorithm to determine whether the input information is useful. The processor has three gates, namely, the input gate, the forget gate, and the output gate. The calculation formula of the forget gate is: t =σ(W f h t-1 +U f x t +b f ), where W f and U f is the weight matrix, b f is the bias term, and σ is the sigmoid activation function. The calculation formula of the input gate is: t =σ(W i h t-1 +U i x t +b i );c t =tanh(W c h t-1 +U c x t +b c ), where W i , U i , W c and U c is the weight matrix, b i and b c is the bias term. The calculation formula for the new cell unit is: t =f t c t-1 +i t c t ; The calculation formula for the output gate is: t =σ(W o h t-1 +U o x t +b o );h t =o t tanh(c t ), where x t is the input at time t, h t is the output at time t, i t is the output of the input gate, f t is the output of the forget gate, c t is the cell state at time t, o t is the output of the output gate, and b is the parameter matrix.

[0052] The Bidirectional Long Short-Term Memory (BiLSTM) network is composed of two ordinary LSTMs. The unidirectional LSTM can only mine the feature information of time series data from the past to the future, but cannot mine the time series data from the future to the past, which may lead to low learning efficiency of the Bidirectional Long Short-Term Memory network and affect the prediction accuracy of the Bidirectional Long Short-Term Memory network. BiLSTM mines time series data information in two directions, which enables BiLSTM to not only use the features of time series at past moments, but also learn the features of time series at future moments, continuously improving the learning ability of the Bidirectional Long Short-Term Memory network and the utilization rate of time series data, thereby improving the prediction accuracy of the Bidirectional Long Short-Term Memory network.

[0053] Adaptive Particle Swarm Optimization (APSO) is selected to optimize the parameters of the bidirectional long short-term memory network.

[0054] The bidirectional long short-term memory network has strong self-learning and memory functions, which improves the fitting effect and generalization ability of the model. However, in the process of model training, the four parameters of the model: learning rate, number of nodes in the first hidden layer, number of nodes in the second hidden layer and number of iterations have uncertainty, diversity and irrelevance, which makes it difficult to determine the parameters and affects the accuracy and training time of the model. Therefore, it is necessary to select a suitable algorithm to quickly determine the optimal parameters of the model. Selecting the optimal parameters is actually an optimization problem. Establish an optimal objective function, set constraints, and establish multi-objective functions for multiple objectives to solve them. Under the condition that the set constraints remain unchanged, a set of solutions is obtained by solving them. This solution is the optimal parameter. In general, the optimal solution to the problem is described as follows: Among them, F(x) is the objective function, g i (x) is the constraint function, and x is the decision variable. When there are many objectives in the optimization problem and there are some contradictions between the objectives, in order to achieve the optimal solution for multiple objectives at the same time, it is necessary to balance the relationship between the various objective functions.

[0055] Therefore, the particle swarm algorithm with adaptive adjustment of inertia weight is selected to find the optimal solution for the parameters in the BiLSTM model, so that it can achieve the optimal purpose and realize fast and accurate search for the optimal parameters; in the optimization process, the four parameters of the long short-term memory network are no longer assigned based on experience, but the particle swarm algorithm is used to optimize the parameters. This improves the accuracy of prediction, reduces the loss value, speeds up the training time, and reduces the number of model iterations.

[0056] like Figure 2As shown, the optimization process of the bidirectional long and short time neural network specifically includes: optimizing the parameters of the bidirectional long and short time neural network by using a particle swarm algorithm that adaptively adjusts the inertia weight to obtain an optimized bidirectional long and short time neural network; the parameters of the bidirectional long and short time neural network specifically include: learning rate, the number of nodes in the first hidden layer, the number of nodes in the second hidden layer, and the number of iterations.

[0057] Among them, the particle swarm algorithm with adaptive adjustment of inertia weight is used to optimize the parameters of the bidirectional long and short time neural network, and the optimized bidirectional long and short time neural network is obtained, which specifically includes:

[0058] Step 401, setting particle parameters, specifically includes: setting a parameter for a particle, in this application, the number of particles is four, setting the maximum number of iterations T, the maximum value of the inertia weight w max , the minimum value of inertia weight w min , cognitive weight c1, social weight c2 and fitness error threshold. Randomly initialize the particle position x(i) and velocity v i (t), initialize the individual extreme value pbest and the global extreme value gbest, where e is the index of the particle.

[0059] Step 402, calculate the adaptive weight w. The adaptive weight w is Among them, i is the current iteration number.

[0060] Step 403, calculate and compare fitness. Specifically, it includes: calculating the current position x of each particle e The fitness f(x) of (t) e (t)). If the fitness of the current position is less than the best fitness value f(pbest) in the history of the particle e ), then update f(pbest e ) is x e (t). And from the individual extreme values ​​pbest of all particles, find the position with the minimum fitness as the global extreme value gbest.

[0061] Step 404, update the speed and position of the particle. Specifically, for each particle, update its speed v e (t+1) and position x e (t+1). The updated speed is:

[0062] v e (t+1)=w(t)·v e (t)+c1·r1·(pbest e -x e (t))+c2·r2·(gbest e -x e(t)); where r1 and r2 are two numbers randomly generated in the interval [0,1].

[0063] The updated position is: x e (t+1)=x e (t)+v e (t+1).

[0064] Step 405, check whether the fitness error is less than the fitness error threshold. Specifically, it includes: calculating the fitness f(gbest) of the current global extreme value gbest e ), if the fitness of the current global extremum gbest is less than the adaptation error threshold, the algorithm is terminated and the particle parameters at this time are output; otherwise, it is iterated to the maximum number of times T. The output particle parameters are input into the bidirectional long-short time neural network to obtain the optimized bidirectional long-short time neural network.

[0065] The training process of the optimized bidirectional long and short time neural network specifically includes: obtaining historical runoff data, preprocessing the historical runoff data, and obtaining the processed historical runoff data; initializing the bidirectional long and short time neural network, and optimizing the parameters of the bidirectional long and short time neural network by using the particle swarm algorithm with adaptive adjustment of the inertia weight to obtain the optimized bidirectional long and short time neural network; inputting the processed historical runoff data into the optimized bidirectional long and short time neural network for training until the optimized bidirectional long and short time neural network converges. And calculate the loss value after the long and short time memory network training, if the loss value is less than 0.2 at this time, stop training, otherwise re-optimize the parameters and continue training until the loss value is less than 0.2 and stop training. The output bidirectional long and short time neural network at this time is the final optimized bidirectional long and short time neural network. Among them, the threshold of the loss value is set to 0.2 because the loss value obtained without the particle swarm algorithm optimization is 0.201887.

[0066] This application uses the complete set empirical mode decomposition method of adaptive noise to decompose the original monthly runoff sequence to obtain a residual sequence and several different intrinsic mode function component sequences; the optimized bidirectional long-short time neural network is used to make multiple predictions for several different intrinsic mode function component sequences and several different intrinsic mode function component sequences, which improves the prediction accuracy compared with the existing artificial intelligence methods, reduces the network training time, and improves the prediction efficiency. The particle swarm algorithm with adaptive adjustment of inertia weight is used to optimize the parameters of the bidirectional long-short time neural network to obtain the optimal parameters that change over time, enhance the robustness and adaptability of the model, and thus improve the prediction accuracy.

[0067] In another exemplary embodiment of the present application, a particle swarm optimization algorithm (PSO) and a bidirectional long short-term memory network (BiLSTM) are combined into a PSO-BiLSTM model; a variational mode decomposition (VMD), a particle swarm optimization algorithm (PSO) and a long short-term memory network (LSTM) are combined into a VMD-PSO-LSTM model; a local characteristic-scale decomposition (LCD), a sparrow search algorithm (SSA) and a bidirectional long short-term memory network (BiLSTM) are combined into an LCD-SSA-BiLSTM model; an empirical mode decomposition (EMD), an attention mechanism (Attention Mechanism) and a bidirectional long short-term memory network (BiLSTM) are combined into an EMD-ATT-BiLSTM model;Ensemble Empirical Mode Decomposition (EEMD), Variational Mode Decomposition (VMD), Convolutional Neural Network (CNN) and Long Short-Term Memory (LSTM) are combined into the EEMD-VMD-CNN-LSTM model, and Complementary Ensemble Empirical Mode Decomposition with Adaptive Noise (CEEMDAN), Variational Mode Decomposition (VMD), Adaptive Particle Swarm Optimization (APSO) and Bidirectional Long Short-Term Memory (BiLSTM) are combined into the CEEMDAN-VMD-APSO-BiLSTM model. ;

[0068] The example analysis data uses the monthly runoff data of a hydropower station in southwest China as the research data. The first 80% of the 660 monthly runoff data of a hydropower station is used as training samples, and the last 10% is used for testing. Under the same experimental process, it is compared with CEEMDAN-VMD-APSO-BiLSTM, EEMD-VMD-CNN-LSTM, EMD-ATT-BiLSTM, PSO-BiLSTM, VMD-PSO-LSTM, LCD-SSA-BiLSTM, and LSTM.

[0069] In terms of the rating criteria for prediction errors, three rating criteria, namely Nash efficiency coefficient (NSE), root mean square error (RMSE) and mean absolute percentage error (MAPE), were used for evaluation. The comparison between the runoff prediction values ​​and actual values ​​of each model is shown in Table 1; the comparison between the evaluation index results of each model is shown in Table 2.

[0070] Table 1 Comparison between the predicted and actual runoff values ​​of each model

[0071]

[0072] Table 2 Comparison of evaluation index results of each model

[0073]

[0074] (1) Compared with the LSTM model without sequence decomposition, the Nash coefficient NSE of the CEEMDAN-VMD-APSO-BiLSTM hybrid model is improved by 167.56%, the root mean square error RMSE is reduced by 86%, and the mean absolute percentage error MAPE is reduced by 68.6%.

[0075] (2) Compared with the PSO-BiLSTM model without sequence decomposition, the Nash coefficient NSE of the CEEMDAN-VMD-APSO-BiLSTM hybrid model is improved by 90.38%, the root mean square error RMSE is reduced by 83%, and the mean absolute percentage error MAPE is reduced by 60%.

[0076] (3) Compared with the VMD-PSO-LSTM model, the Nash coefficient NSE of the CEEMDAN-VMD-APSO-BiLSTM hybrid model is improved by 3.1%, the root mean square error RMSE is reduced by 41.7%, and the mean absolute percentage error MAPE is reduced by 27.6%.

[0077] (4) Compared with the LCD-SSA-BiLSTM model, the Nash coefficient NSE of the CEEMDAN-VMD-APSO-BiLSTM hybrid model is improved by 3.1%, the root mean square error RMSE is reduced by 40.6%, and the mean absolute percentage error MAPE is reduced by 27%.

[0078] (5) Compared with the EMD-ATT-BiLSTM model, the root mean square error (RMSE) of the CEEMDAN-VMD-APSO-BiLSTM hybrid model is reduced by 38.7% and the mean absolute percentage error (MAPE) is reduced by 26.3%.

[0079] (6) Compared with the EEMD-VMD-CNN-LSTM model, the root mean square error (RMSE) of the CEEMDAN-VMD-APSO-BiLSTM model is reduced by 20% and the mean absolute percentage error (MAPE) is reduced by 13.1%.

[0080] From the results in Table 2, it can be seen that the prediction absolute error of the decomposed model at the peak point of the monthly runoff series is smaller than that of other models, indicating that the hybrid runoff prediction model combined with the decomposition method can improve the prediction accuracy of the peak point of the monthly runoff series.

[0081] This method is based on a hybrid prediction model that combines variational mode decomposition (VMD), complete ensemble empirical mode decomposition with adaptive noise (CEEMDAN) and bidirectional long short-term memory neural network (BiLSTM), and the reliability of the model is verified using data.

[0082] Based on the idea of ​​"decomposition--prediction--reconstruction", CEEMDAN is first used to process the original monthly runoff sequence to obtain a series of modal components and a residual sequence with high complexity. The residual sequence obtained after VMD decomposition is processed by CEEMDAN to reduce its complexity, extract hidden information, and improve the prediction accuracy of the subsequent prediction model. Combined with BiLSTM for prediction, it can not only better extract the latent features of the runoff sequence, but also learn the long-term dependence characteristics of the data and improve the prediction ability of the model. Through example analysis, it is verified that the model of this method can effectively improve the prediction accuracy of the peak point of the runoff sequence and the overall prediction accuracy, which reflects the effectiveness of this method.

[0083] Based on the same inventive concept, an embodiment of the present application also provides a runoff prediction device based on modal decomposition and bidirectional long and short time neural network for implementing the above-mentioned runoff prediction method based on modal decomposition and bidirectional long and short time neural network.

[0084] In an exemplary embodiment, a runoff prediction device based on modal decomposition and bidirectional long-short-time neural network is provided, comprising:

[0085] The first data decomposition module is used to decompose the original monthly runoff sequence based on the complete set empirical mode decomposition method with adaptive noise to obtain a residual sequence and a plurality of different intrinsic mode function component sequences; the original monthly runoff sequence includes runoff data for a period of history.

[0086] The first model prediction module is used to input the several different intrinsic modal function component sequences into the optimized bidirectional long-short time neural network respectively to obtain the prediction sequences of several different intrinsic modal function components.

[0087] The second data decomposition module is used to decompose the residual sequence based on the variational mode decomposition method to obtain a plurality of different intrinsic mode function component sequences.

[0088] The second model prediction module is used to input the several different intrinsic mode function component sequences into the optimized bidirectional long-short time neural network respectively to obtain the prediction sequences of several different intrinsic mode function components.

[0089] The data reconstruction module is used to add and reconstruct the prediction sequences of the several different intrinsic mode function components and the prediction sequences of the several different intrinsic mode function components to obtain a monthly runoff sequence for a period of time in the future.

[0090] In an exemplary embodiment, a computer device is provided. The computer device may be a server or a terminal. The internal structure diagram thereof may be as follows: Figure 3As shown. The computer device includes a processor, a memory, an input / output interface (Input / Output, referred to as I / O) and a communication interface. The processor, the memory and the input / output interface are connected through a system bus, and the communication interface is connected to the system bus through the input / output interface. The processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program and a database. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The database of the computer device is used to store video tag processing data. The input / output interface of the computer device is used to exchange information between the processor and an external device. The communication interface of the computer device is used to communicate with an external terminal through a network connection. When the computer program is executed by the processor, a method for realizing the above-mentioned runoff prediction method based on modal decomposition and bidirectional long-short-time neural network is implemented.

[0091] Those skilled in the art will understand that Figure 3 The structure shown in the figure is only a block diagram of a part of the structure related to the solution of the present application, and does not constitute a limitation on the computer device to which the solution of the present application is applied. The specific computer device may include more or fewer components than those shown in the figure, or combine certain components, or have a different arrangement of components.

[0092] In an exemplary embodiment, a computer device is also provided, including a memory and a processor, wherein a computer program is stored in the memory, and when the processor executes the computer program, the steps in the above-mentioned runoff prediction method embodiment based on modal decomposition and bidirectional long-short-time neural network are implemented.

[0093] In an exemplary embodiment, a computer-readable storage medium is provided, storing a computer program, which, when executed by a processor, implements the steps in the above-mentioned runoff prediction method embodiment based on modal decomposition and bidirectional long-short-time neural network.

[0094] Those of ordinary skill in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing the relevant hardware through a computer program, and the computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to the memory, database or other medium used in the embodiments provided in the present application can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM may be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM).

[0095] The database involved in each embodiment provided in this application may include at least one of a relational database and a non-relational database. The non-relational database may include a distributed database based on blockchain, etc., but is not limited thereto. The processor involved in each embodiment provided in this application may be a general-purpose processor, a central processing unit, a graphics processor, a digital signal processor, a programmable logic device, a data processing logic device based on quantum computing, etc., but is not limited thereto.

[0096] The technical features of the above embodiments may be combined arbitrarily. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0097] This article uses specific examples to illustrate the principles and implementation methods of this application. The description of the above embodiments is only used to help understand the method and core ideas of this application. At the same time, for those skilled in the art, according to the ideas of this application, there will be changes in the specific implementation methods and application scope. In summary, the content of this specification should not be understood as limiting this application.

Claims

1. A runoff prediction method based on modal decomposition and bidirectional long-short-time neural network, characterized in that: The runoff prediction method based on modal decomposition and bidirectional long-short time neural network includes: The original monthly runoff sequence is decomposed by a complete ensemble empirical mode decomposition method based on adaptive noise to obtain a residual sequence and several different intrinsic mode function component sequences; the original monthly runoff sequence includes runoff data for a period of history; Inputting the plurality of different intrinsic mode function component sequences into the optimized bidirectional long-short time neural network respectively to obtain the prediction sequences of the plurality of different intrinsic mode function components; Decomposing the residual sequence based on the variational mode decomposition method to obtain a number of different intrinsic mode function component sequences; inputting the several different intrinsic mode function component sequences into the optimized bidirectional long-short time neural network respectively to obtain a number of different intrinsic mode function component prediction sequences; The prediction sequences of the several different intrinsic mode function components and the prediction sequences of the several different intrinsic mode function components are added and reconstructed to obtain a monthly runoff sequence for a period of time in the future.

2. The runoff prediction method based on modal decomposition and bidirectional long-short time neural network according to claim 1 is characterized in that: Before decomposing the original monthly runoff sequence by the complete ensemble empirical mode decomposition method based on adaptive noise to obtain a residual sequence and a plurality of different intrinsic mode function component sequences, the runoff prediction method based on mode decomposition and bidirectional long-short time neural network also includes: Normalize the runoff data of a period of history to obtain the processed historical runoff data, input the processed historical runoff data into the bidirectional long-short time neural network to obtain the runoff data of a period of time in the future; The maximum and minimum values ​​in the runoff data for the future period of time are selected and spliced ​​with the processed historical runoff data to obtain the original monthly runoff sequence.

3. The runoff prediction method based on modal decomposition and bidirectional long-short time neural network according to claim 1 is characterized in that: The original monthly runoff series is decomposed by the complete ensemble empirical mode decomposition method based on adaptive noise to obtain a residual series and several different intrinsic mode function component series, including: For the i-th iteration, different white noises are added to the original monthly runoff sequence, and the original monthly runoff sequence with different white noises is decomposed N times using the empirical mode decomposition method to obtain the intrinsic mode function component sequence of the i-th iteration; 1≤i; Based on the original monthly runoff sequence and the intrinsic mode function component sequence of the i-th iteration, obtaining the residual sequence of the i-th iteration; Determine whether the residual sequence of the i-th iteration can be decomposed. If so, take the residual sequence of the i-th iteration as the original monthly runoff sequence and perform the i+1-th iteration; If not, the residual sequence of the i-th iteration is taken as the final residual sequence, and the intrinsic mode function component sequence of the 1st iteration to the intrinsic mode function component sequence of the i-th iteration are taken as several different intrinsic mode function component sequences.

4. The runoff prediction method based on modal decomposition and bidirectional long-short time neural network according to claim 3 is characterized in that: The intrinsic mode function component sequence includes the intrinsic mode function components obtained by each decomposition; the intrinsic mode function component sequence of the i-th iteration is: Among them, F i (t) is the intrinsic mode function component sequence of the i-th iteration, N is the number of decompositions, F1 n (t) is the first intrinsic mode function component obtained by the nth decomposition, r i-1 (t) is the original monthly runoff sequence of the i-1th iteration, v i (t) is the white noise added in the i-th iteration; The residual sequence of the i-th iteration is: r i (t)=r i-1 (t)-F i (t); Among them, r i (t) is the residual sequence of the i-th iteration, r i-1 (t) is the residual sequence of the i-1th iteration.

5. The runoff prediction method based on modal decomposition and bidirectional long-short time neural network according to claim 1 is characterized in that: Inputting the plurality of different intrinsic mode function component sequences into the optimized bidirectional long-short time neural network respectively, specifically comprising: The predicted parts of the several different intrinsic modal function component sequences are truncated, and the remaining parts of the several different intrinsic modal function component sequences are respectively input into the optimized bidirectional long-short time neural network; the predicted parts of the several different intrinsic modal function component sequences are the parts after the sudden change of runoff data in the several different intrinsic modal function component sequences.

6. The runoff prediction method based on modal decomposition and bidirectional long-short time neural network according to claim 1 is characterized in that: Decomposing the residual sequence based on the variational mode decomposition method specifically includes: The predicted part of the residual sequence is truncated, and the remaining part of the residual sequence is decomposed based on the variational mode decomposition method; the predicted part of the residual sequence is the part after the runoff data in the residual sequence changes suddenly.

7. The runoff prediction method based on modal decomposition and bidirectional long-short time neural network according to claim 1 is characterized in that: The optimization process of the bidirectional long-short time neural network specifically includes: The parameters of the bidirectional long short time neural network are optimized by using a particle swarm algorithm with adaptive adjustment of inertia weights to obtain an optimized bidirectional long short time neural network; the parameters of the bidirectional long short time neural network specifically include: learning rate, the number of nodes in the first hidden layer, the number of nodes in the second hidden layer, and the number of iterations.

8. A runoff prediction device based on modal decomposition and bidirectional long-short time neural network, applied to the runoff prediction method based on modal decomposition and bidirectional long-short time neural network described in any one of claims 1 to 7, characterized in that: The runoff prediction device based on modal decomposition and bidirectional long-short time neural network includes: The first data decomposition module is used to decompose the original monthly runoff sequence based on the complete set empirical mode decomposition method with adaptive noise to obtain a residual sequence and a plurality of different intrinsic mode function component sequences; the original monthly runoff sequence includes runoff data for a period of history; A first model prediction module is used to input the plurality of different intrinsic modal function component sequences into the optimized bidirectional long-short time neural network respectively to obtain prediction sequences of the plurality of different intrinsic modal function components; A second data decomposition module is used to decompose the residual sequence based on a variational mode decomposition method to obtain a plurality of different intrinsic mode function component sequences; A second model prediction module is used to input the plurality of different intrinsic mode function component sequences into the optimized bidirectional long-short time neural network respectively to obtain prediction sequences of the plurality of different intrinsic mode function components; The data reconstruction module is used to add and reconstruct the prediction sequences of the several different intrinsic mode function components and the prediction sequences of the several different intrinsic mode function components to obtain a monthly runoff sequence for a period of time in the future.

9. A computer device comprising: A memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the runoff prediction method based on modal decomposition and bidirectional long-short-time neural network as described in any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the runoff prediction method based on modal decomposition and bidirectional long-short-time neural network described in any one of claims 1 to 7 is implemented.

Images