Reservoir basin daily runoff prediction method, system, equipment and medium

Abstract

The invention provides a reservoir drainage basin daily runoff prediction method, system and device and a medium, and the method comprises the following steps: S1, obtaining daily runoff data in a certain drainage basin, carrying out the abnormal value detection of the daily runoff data through employing an isolated forest, and carrying out the data cleaning; s2, decomposing the cleaned daily runoff data by using data-driven adaptive chirp mode decomposition to obtain a plurality of components; constructing an input matrix for each component, and dividing the input matrix into a training set and a test set; s3, constructing a fusion model of the time convolution network and the non-inverse extreme learning machine, and training the fusion model; s4, introducing a half uniform initialization and variable helix updating strategy into a PID search algorithm to obtain an IPSA algorithm; using an IPSA algorithm to optimize the TCN-IFELM model; and S5, obtaining a prediction result of each component by using the test set, and adding the prediction results to obtain a predicted value of the daily runoff at the future moment. According to the method, the time sequence characteristic and the nonlinear relation of the runoff data can be effectively captured, so that the daily runoff prediction precision is improved.

Description

Technical Field

[0001] The present invention belongs to the technical field of hydrological data prediction, and in particular relates to a method, system, equipment and medium for predicting daily runoff in a reservoir basin. Background Art

[0002] Runoff refers to the flow of water into rivers through the surface and underground after precipitation, excluding evaporation, infiltration and plant interception. Runoff prediction is a key task in hydrology and water resources management. Its purpose is to predict runoff in the future based on previous hydrological and meteorological factors. However, since the runoff process is affected by factors such as human activities and climate change, it is a highly complex nonlinear process. Especially in the context of frequent extreme weather, the runoff sequence shows greater volatility.

[0003] Traditional runoff prediction models include physical cause analysis and mathematical statistics. The physical cause analysis method requires a large amount of measured data and fine parameter adjustment, and is difficult to adapt to complex and changing environmental conditions. Mathematical statistics has limited modeling capabilities for nonlinear and non-stationary time series, and is difficult to capture the dynamic changes of hydrological processes.

[0004] At present, with the advancement of technology, runoff prediction models are increasingly inclined to use big data and machine learning technology to improve the accuracy and efficiency of predictions. For example, the rapid development of deep learning models in time series processing methods has made it possible to accurately predict runoff. At the same time, the combined prediction model based on "decomposition-prediction-reconstruction" shows good practicality and can better handle nonlinear and non-stationary runoff data. By decomposing the original hydrological time series through time-frequency signal decomposition technology, it can effectively extract effective information from the hydrological time series, thereby improving the prediction accuracy of the model. Summary of the invention

[0005] The first object of the present invention is to provide a method for predicting daily runoff in a reservoir basin, in the hope of effectively capturing the nonlinear relationship of runoff data, thereby improving the prediction accuracy of daily runoff.

[0006] To this end, the above-mentioned purpose of the present invention is achieved through the following technical solutions: A reservoir basin daily runoff prediction method comprises the following steps: S1. Obtain daily runoff data from monitoring stations in the reservoir basin in advance, use isolation forest to detect outliers in the daily runoff data, and perform data cleaning; S2, using data-driven adaptive chirp modal decomposition (DD-ACMD) to decompose the daily runoff data processed by step S1 to obtain multiple components; constructing an input matrix for each component and dividing it into a training set and a test set; S3, constructing a TCN-IFELM fusion model that integrates the temporal convolutional network (TCN) and the inverse-free extreme learning machine (IFELM), and inputting the obtained training set into the TCN-IFELM model for training; S4. The half-uniform initialization and variable spiral update strategy are introduced into the PID search algorithm (PSA) to obtain the improved PID search algorithm IPSA; the IPSA algorithm is used to optimize the parameters such as the discard rate, the number of residual blocks and the number of hidden layer neurons of the TCN-IFELM model to improve the prediction performance of the TCN-IFELM model; S5. Input the test set into the optimized fusion model for component prediction, and add the prediction results of each component to obtain the predicted value of daily runoff at a certain moment in the future.

[0007] While adopting the above technical solutions, the present invention may also adopt or combine the following technical solutions: As a preferred technical solution of the present invention: in step S1, the implementation process of using isolation forest to detect outliers in daily runoff data is as follows: S11. Build an Isolation Forest ; ; in, Represents a binary tree; w is the number of binary trees; S12. From daily runoff data V Randomly select Q The data points form a subspace X , randomly selected X A data feature is selected, and a cutting point is randomly selected in the sample interval corresponding to the feature. A hyperplane is constructed from the cutting point to convert X Split into two different branches under the current node; S13, recursively execute step S12 for the child nodes under two different branches until the data cannot be further split; S14. Calculate and identify daily runoff data sets V Medium data point v The degree of abnormality : ; in, For data points v The depth of a node in a binary tree, For data points v exist w The average path length on the tree, Reason Q The average path length of the tree built by points; S15. Cut out outlier and abnormal data points and determine outliers.

[0008] As a preferred technical solution of the present invention: the specific process of step S2 is as follows: S21. The high-frequency mode is enhanced by taking the derivative of the daily runoff time series signal. Its derivative can be expressed as: ; in, represents the original daily runoff time series signal; For signal The derivative of K Represents the total number of modes in the signal; For the i The amplitude of each mode; No. i The phase function of each mode; For the i The instantaneous frequency of each mode; for The derivative of t represents the time variable; S22, estimating the initial instantaneous frequency of the highest frequency mode by normalizing the derivative of the signal, the specific process is as follows: ; in, After normalization, i A frequency modulation signal of a mode; No. i The phase function of each mode; represents the instantaneous frequency of the highest frequency mode; d It means differential, which refers to a very small amount of change; dt Indicates a small interval of time; is a differential operator, which represents the time t The derivative of S23, introduce a time-varying low-pass filter to remove noise above the estimated instantaneous frequency of the current signal, the process is as follows: S231, rewrite the frequency modulation signal as: ; in, Indicates frequency modulated signal; Represents a time interval; express An infinitesimal change in Indicates that the signal is at point The instantaneous frequency of Indicates a time interval On, instantaneous frequency Small changes in represents the initial phase; S232, calculating the analytical signal of the frequency modulation signal: ; in, for The analytical signal of Indicates frequency modulated signal; represents the initial phase; express Hilbert transform; j represents the imaginary part of the signal; S233. Demodulation operator definition and modulation operator : ; ; in, is the demodulation frequency; Indicates the carrier frequency; S234, multiplying the analytical signal by the demodulation operator to obtain a demodulated signal: ; in, represents the demodulated signal; S235, remove the noise higher than the carrier frequency through low-pass filtering, and then multiply it by the modulation operator to obtain a new analytical signal: ; in, Represents the modulated signal; is the demodulated signal after low-pass filtering; S24. After each pattern extraction, a time-varying low-pass filter is used to remove the noise above the pattern to reduce the impact on subsequent pattern extraction. The signal patterns are extracted from high frequency to low frequency through a recursive framework. After each pattern is extracted, the remaining signals are updated and the above process is repeated until the stopping condition is met. DD-ACMD can decompose the daily runoff time series signal into independent components, which is convenient for subsequent analysis and prediction.

[0009] As a preferred technical solution of the present invention: the specific process of step S3 is as follows: S31. TCN is used to capture the dependencies in the daily runoff time series. The specific process of dilated convolution is as follows: ; in, is the input sequence; f is the filter;d is the expansion factor; z is the convolution kernel position; is the convolution kernel size; Indicates the element of the previous layer; S32, using the residual structure to input The output of the convolutional network Perform addition operation, the calculation formula is: ; in, is the activation function; S33. Define the hidden node adding strategy of IFELM: ; in, l is the number of current hidden nodes; and yes l The input weights and biases for hidden nodes, is the newly added input weight, is the newly added bias; T Represents a transpose operation; S34. Update the output weights according to the following rules: ; ; in, The output weight matrix after adding a hidden node; Y Represents the training output matrix; Indicates the current l The pseudo-inverse matrix of hidden nodes; It is the pseudo-inverse matrix after adding a hidden node; and for Two blocks of are used to update the pseudo-inverse matrix; c Represents the vector after adding a hidden node; H represents the hidden layer output matrix; I is the identity matrix; Representation vector c The transpose of S35. Construct a TCN-IFELM fusion model. Use TCN to extract deep features between runoff data, convert the input high-dimensional data into low-dimensional feature representation, and IFELM takes the features extracted by TCN as input and the corresponding runoff as target output in turn. IFELM avoids the matrix inversion operation in the traditional extreme learning machine by gradually adding hidden nodes and updating weights, thereby improving computational efficiency.

[0010] As a preferred technical solution of the present invention: the specific process of step S4 is as follows: S41, setting the population size, maximum number of iterations and interval range of the parameters to be optimized of the IPSA algorithm; S42, using the half-uniform initialization strategy to improve the population initialization part of the original PSA algorithm, the specific process is as follows: ; in, represents the number of iterations, and are the initial values ​​of the individuals, , ; up and low Indicates the upper and lower limits of a variable; n is the population size; is a random number between 0 and 1; the half-uniform initialization strategy not only retains the randomness of the population, but also avoids the concentrated distribution of initialized individuals, which is conducive to improving the diversity of the population; S43, defining the objective function of the algorithm as the root mean square error between the daily runoff prediction value and the actual value, and calculating the fitness value of the population through the objective function; S44. Calculate system deviation: the population in the number of iterations Deviation of time Defined as: ; in, Represents the best individual value in the population; Indicated in Individual values ​​during iteration; S45, PID adjustment process: number of iterations Output value of PID regulation Defined as: ; in, , and is a vector of random numbers between 0 and 1; , and are the adjustment coefficients for proportional, integral and differential respectively; S46, population update: In order to prevent the algorithm from falling into the local optimum, the variable spiral update strategy is used to improve the population update process of the PSA algorithm. The specific process is as follows: ; in, is the value after individual disturbance; Represents the average fitness of the top 10% individuals. is the dynamic weight; b To adjust the parameters of the spiral shape; is a random number uniformly distributed between [-1, 1]; D For individuals and The improved population update formula is: ; in, is a random number between 0 and 1; It means the number of iterations is The output value of PID regulation when ; Indicates the value of the individual at the next iteration; the variable spiral update strategy increases the exploration of local areas while maintaining the global search capability of the algorithm, thereby improving the search accuracy and convergence speed of the algorithm; S47, the IPSA algorithm is used to optimize the TCN-IFELM model's discard rate, the number of residual blocks, the number of hidden layer neurons and other parameters, so that the root mean square error between the daily runoff prediction value and the actual value is continuously reduced; S48. Determine whether the algorithm termination condition is reached through a given maximum number of iterations, and finally output the optimal parameters of the TCN-IFELM model within the maximum number of iterations.

[0011] The second object of the present invention is to provide a reservoir basin daily runoff prediction system, comprising the following modules: A daily runoff data acquisition module, wherein the daily runoff data is used to pre-acquire the daily runoff data of the reservoir basin monitoring station, use the isolation forest to perform outlier detection on the daily runoff data, and perform data cleaning; A data processing module, wherein the data processing module is used to decompose the daily runoff data processed by the daily runoff data acquisition module using data-driven adaptive chirp modal decomposition to obtain multiple components; construct an input matrix for each component and divide it into a training set and a test set; A TCN-IFELM fusion model construction and training module, wherein the TCN-IFELM fusion model construction and training module is used to construct a TCN-IFELM fusion model that integrates a temporal convolutional network (TCN) and an inverse extreme learning machine (IFELM), and input the training set obtained by the data processing module into the TCN-IFELM model for training; A TCN-IFELM fusion model optimization module, wherein the TCN-IFELM fusion model optimization module is used to introduce half uniform initialization and variable spiral update strategies into the PID search algorithm to obtain an improved PID search algorithm IPSA; the IPSA algorithm is used to optimize the discard rate, the number of residual blocks, and the number of hidden layer neurons of the TCN-IFELM model constructed by the TCN-IFELM fusion model construction and training module; The prediction module is used to input the test set of each component obtained by the data processing module into the TCN-IFELM fusion model optimized by the TCN-IFELM fusion model optimization module to perform component prediction, add the prediction results of each component, and obtain the predicted value of daily runoff at a certain time in the future.

[0012] The third object of the present invention is to provide an electronic device, the electronic device comprising a processor, a communication interface, a memory and a communication bus, wherein the processor, the communication interface and the memory communicate with each other via the communication bus. A memory, the memory being used to store a computer program; A processor is used to execute the computer program stored in the memory to implement the steps of the reservoir basin daily runoff prediction method as described above.

[0013] Another object of the present invention is to provide a non-transitory readable storage medium, in which a program is stored. When the program is executed by a processor, the steps of the reservoir basin daily runoff prediction method described above are implemented.

[0014] The present invention provides a reservoir basin daily runoff prediction method, system, device and medium, the method comprising the following steps: S1, obtaining daily runoff data in a certain basin, using isolation forest to detect outliers on the daily runoff data, and performing data cleaning; S2, using data-driven adaptive chirp modal decomposition (DD-ACMD) to decompose the cleaned daily runoff data to obtain multiple components; constructing an input matrix for each component and dividing it into a training set and a test set; S3, constructing a temporal convolutional network (TCN) and an inverse-free extreme learning machine (IF S4, the half-uniform initialization and variable spiral update strategy are introduced into the PID search algorithm (PSA) to obtain the improved PID search algorithm (IPSA); the IPSA algorithm is used to optimize the TCN-IFELM model's parameters such as the discard rate, the number of residual blocks and the number of hidden layer neurons; S5, the test set is used to obtain the prediction results of each component and add them up to obtain the predicted value of daily runoff at the future moment.

[0015] Compared with the prior art, the present invention has the following beneficial effects: 1) The present invention proposes to use data-driven adaptive chirp modal decomposition (DD-ACMD) to decompose daily runoff data. Considering that runoff data usually contains signals of multiple time scales and frequencies, DD-ACMD can decompose these complex signals into independent components, which is convenient for subsequent analysis and prediction. It not only has high adaptability and can adapt to different signal characteristics, but also has good noise robustness.

[0016] 2) The present invention constructs a fusion model of TCN-IFELM. TCN can extract the time series characteristics of long-term dependencies in runoff data. IFELM avoids the matrix inversion operation in traditional ELM by gradually adding hidden nodes and updating weights, thereby improving computational efficiency. The TCN-IFELM model combines the advantages of both and improves the generalization ability of the model.

[0017] 3) The present invention proposes an IPSA algorithm that combines multiple improvement strategies. It uses a half-uniform initialization strategy, which not only retains the randomness of the population, but also avoids the concentrated distribution of initialized individuals, which is conducive to improving the diversity of the population. At the same time, the variable spiral update strategy is used to improve the population update process of the PSA algorithm. While maintaining the global search capability of the algorithm, it increases the exploration of local areas, thereby improving the search accuracy and convergence speed of the algorithm. IPSA is used to optimize the TCN-IFELM model's discard rate, the number of residual blocks, and the number of hidden layer neurons, which optimizes the structure of the model and further improves the prediction performance of the model.

[0018] 4) The present invention can effectively capture the time series characteristics and nonlinear relationships of runoff data, thereby improving the prediction accuracy of daily runoff. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] Figure 1 The present invention provides a flow chart of the reservoir basin daily runoff prediction method.

[0020] Figure 2 This is the structural diagram of the TCN model.

[0021] Figure 3 This is the structural diagram of the IFELM model.

[0022] Figure 4 The radar chart is a comparison of the NSE indicators of each model.

[0023] Figure 5 Box plot comparing the prediction errors of each model.

[0024] Figure 6 It is a line comparison chart of the predicted value and the measured value of the TCN-IFELM fusion model provided by the present invention. DETAILED DESCRIPTION

[0025] The present invention is further described in detail below with reference to the accompanying drawings and specific embodiments.

[0026] like Figure 1 As shown, a reservoir basin daily runoff prediction method specifically includes the following steps: (1) Obtain daily runoff data from monitoring stations in the reservoir basin in advance, use the isolation forest method to detect outliers in the daily runoff data, and perform data cleaning; The implementation process of using isolation forest to detect outliers in daily runoff data is as follows: (11) Building an Isolation Forest ; (1) in, Represents a binary tree; w is the number of binary trees; (12) From daily runoff data V Randomly select Q The data points form a subspace X , randomly selected X A data feature is selected, and a cutting point is randomly selected in the sample interval corresponding to the feature. A hyperplane is constructed from the cutting point to convert X Split into two different branches under the current node; (13) Recursively execute step (12) for the child nodes under two different branches until the data cannot be further split; (14) Calculate and identify daily runoff data sets V Medium data point v The degree of abnormality : (2) in, For data points v The depth of a node in a binary tree, For data points v exist w The average path length on the tree, Reason Q The average path length of the tree built by points; (15) Cut out outlier and abnormal data points and identify outliers.

[0027] (2) Decompose the daily runoff data in step (1) using data-driven adaptive chirp modal decomposition (DD-ACMD) to obtain multiple components; construct an input matrix for each component and divide it into a training set and a test set; (21) The high-frequency pattern is enhanced by taking the derivative of the daily runoff time series signal. Its derivative can be expressed as: (3) in, represents the original daily runoff time series signal; For signal The derivative of K Represents the total number of modes in the signal; For the i The amplitude of each mode; No. i The phase function of each mode; For the i The instantaneous frequency of each mode; for The derivative of t represents the time variable; (22) The initial instantaneous frequency of the highest frequency mode is estimated by normalizing the derivative of the signal; the specific process is as follows: (4) in, After normalization, i A frequency modulation signal of a mode; represents the instantaneous frequency of the highest frequency mode; d It means differential, which refers to a very small amount of change; dt Indicates a small interval of time; is a differential operator, which represents the time t The derivative of (23) A time-varying low-pass filter is introduced to remove noise above the estimated instantaneous frequency of the current signal; the process is as follows: (231) Rewrite the FM signal as: (5) in, Indicates frequency modulated signal; Represents a time interval; express An infinitesimal change in Indicates that the signal is at point The instantaneous frequency of Indicates a time interval On, instantaneous frequency Small changes in represents the initial phase; (232) Calculate the analytical signal of the FM signal: (6) in, for The analytical signal of represents the initial phase; express Hilbert transform; j represents the imaginary part of the signal; (233) Define the demodulation operator and modulation operator : (7) (8) in, is the demodulation frequency; Indicates the carrier frequency; (234) Multiplying the analytical signal by the demodulation operator yields the demodulated signal: (9) in, represents the demodulated signal; (235) The noise higher than the carrier frequency is removed by low-pass filtering, and then multiplied by the modulation operator to obtain a new analytical signal: (10) in, Represents the modulated signal; is the demodulated signal after low-pass filtering; (24) After each pattern extraction, a time-varying low-pass filter is used to remove the noise above the pattern to reduce the impact on subsequent pattern extraction. The signal patterns are extracted from high frequency to low frequency through a recursive framework. After each pattern is extracted, the remaining signal is updated and the above process is repeated until the stopping condition is met. DD-ACMD can decompose the daily runoff time series signal into independent components, which is convenient for subsequent analysis and prediction.

[0028] (3) Construct a temporal convolutional network (TCN) and an inverse-free extreme learning machine (IFELM) fusion model (TCN-IFELM), and input the obtained training set into the TCN-IFELM model for training; (31) TCN is used to capture the dependencies in the daily runoff time series. The specific process of dilated convolution is as follows: (11) in, is the input sequence; f is the filter; d is the expansion factor; z is the convolution kernel position; is the convolution kernel size; Indicates the element of the previous layer; (32) Using the residual structure, the input The output of the convolutional network Perform addition operation, the calculation formula is: (12) in, is the activation function; the structure of TCN is as follows Figure 2 shown.

[0029] (33) Define the hidden node addition strategy of IFELM (13) in, l is the number of current hidden nodes; and yes l The input weights and biases for hidden nodes, is the newly added input weight, is the newly added bias; T Represents a transpose operation.

[0030] (34) Update the output weights using the following rules: (14) (15) in, The output weight matrix after adding a hidden node; Y Represents the training output matrix; Indicates the current l The pseudo-inverse matrix of hidden nodes; It is the pseudo-inverse matrix after adding a hidden node; and for Two blocks of are used to update the pseudo-inverse matrix; c Represents the vector after adding a hidden node; H represents the hidden layer output matrix; I is the identity matrix; Representation vector c The transpose of (35) Construct a TCN-IFELM fusion model; use TCN to extract deep features between runoff data, convert the input high-dimensional data into low-dimensional feature representation, and IFELM takes the features extracted by TCN as input and the corresponding runoff as target output; IFELM avoids the matrix inversion operation in the traditional extreme learning machine by gradually adding hidden nodes and updating weights, thereby improving the computational efficiency; the structure of IFELM is as follows Figure 3 shown.

[0031] (4) The half-uniform initialization and variable spiral update strategy are introduced into the PID search algorithm (PSA) to obtain the improved PID search algorithm (IPSA); the IPSA algorithm is used to optimize the parameters of the TCN-IFELM model such as the discard rate, the number of residual blocks, and the number of hidden layer neurons to improve the prediction performance of the TCN-IFELM model; Figure 1 As shown; (41) Set the population size, maximum number of iterations, and range of parameters to be optimized for the IPSA algorithm; (42) The population initialization part of the original PSA algorithm is improved by using the half-uniform initialization strategy; the specific process is as follows: (16) in, represents the number of iterations, and are the initial values ​​of the individuals, , ; up and low Indicates the upper and lower limits of a variable; n is the population size; is a random number between 0 and 1; the half-uniform initialization strategy not only retains the randomness of the population, but also avoids the concentrated distribution of initialized individuals, which is conducive to improving the diversity of the population; (43) The objective function of the algorithm is defined as the root mean square error between the daily runoff prediction value and the actual value, and the fitness value of the population is calculated through the objective function; (44) Calculate the system deviation; the population in the number of iterations Deviation of time Defined as: (17) in, Represents the best individual value in the population; Indicated in Individual values ​​during iteration; (45) PID adjustment process; number of iterations Output value of PID regulation Defined as: (18) in, , and is a vector of random numbers between 0 and 1; , and are the adjustment coefficients for proportional, integral and differential respectively; (46) Population update: In order to prevent the algorithm from falling into the local optimum, the variable spiral update strategy is used to improve the population update process of the PSA algorithm. The specific process is as follows: (19) in, is the value after individual disturbance; Represents the average fitness of the top 10% individuals. is the dynamic weight; b To adjust the parameters of the spiral shape; is a random number uniformly distributed between [-1, 1]; D For individuals and The improved population update formula is: (20) in, is a random number between 0 and 1; It means the number of iterations is The output value of PID regulation when ; Indicates the value of the individual at the next iteration; the variable spiral update strategy increases the exploration of local areas while maintaining the global search capability of the algorithm, thereby improving the search accuracy and convergence speed of the algorithm; (47) The IPSA algorithm was used to optimize the TCN-IFELM model parameters such as the discard rate, the number of residual blocks, and the number of hidden layer neurons, so that the root mean square error between the daily runoff prediction value and the actual value continued to decrease; (48) Whether the algorithm termination condition is reached is determined by the given maximum number of iterations, and finally the optimal parameters of the TCN-IFELM model are output within the maximum number of iterations.

[0032] (5) The test set is input into the optimized fusion model for component prediction, and the prediction results of each component are added together to obtain the predicted value of daily runoff at the future time.

[0033] In order to verify the accuracy of the daily runoff prediction model (DD-ACMD-IPSA-TCN-IFELM) provided by the present invention in predicting the daily runoff time series, and to prove the effectiveness and significance of the model in the field of hydrological prediction, the daily runoff data of the Wujiangdu Hydrological Observation Station on the main stream of the Wujiang River in Zunyi County, Guizhou Province were taken as the research object, and the daily runoff (m 3 / s) data, a total of 7304 data points; the algorithm program was written in MATLAB language to construct the daily runoff prediction model (DD-ACMD-IPSA-TCN-IFELM) provided by the present invention, and compared with six other prediction models, namely, error back propagation algorithm (BP), extreme learning machine (ELM), inverse extreme learning machine (IFELM), time convolution network (TCN), TCN-IFELM and DD-ACMD-TCN-IFELM. Through comparative analysis, it was found that the DD-ACMD-IPSA-TCN-IFELM model provided by the present invention has higher accuracy than the other six models, and has certain advantages in daily runoff prediction. Table 1 shows the evaluation indicators of various different models, including root mean square error (RMSE), mean absolute error (MAE), correlation coefficient (R), symmetric mean absolute percentage error (SMAPE) and Nash efficiency coefficient (NSE).

[0034] Table 1 Performance index statistics between the model provided by the present invention and the control group model Model RMSE MAE R SMAPE NSE Error back propagation algorithm (BP) 191.3771 115.5188 0.9202 30.4079 0.8407 Extreme Learning Machine (ELM) 180.8141 102.6418 0.9264 26.4456 0.8578 Inverse Extreme Learning Machine (IFELM) 177.7680 102.1859 0.9288 26.3197 0.8626 Temporal Convolutional Network (TCN) 171.7469 102.0259 0.9352 26.2213 0.8717 TCN-IFELM 169.0253 99.5835 0.9366 26.0195 0.8757 DD-ACMD-TCN-IFELM 109.8021 72.6003 0.9741 21.3179 0.9476 DD-ACMD-IPSA-TCN-IFELM 98.3114 59.7773 0.9790 20.2754 0.9580 It can be seen from Table 1 that the DD-ACMD-PSA-TCN-IFELM model is superior to other models in the control group in all indicators. Its RMSE, MAE and SMAPE are the smallest, which are 98.3114, 59.7773 and 20.2754 respectively, which are lower than the BP, ELM, IFELM, TCN, TCN-IFELM and DD-ACMD-TCN-IFELM models. This shows that the DD-ACMD-PSA-TCN-IFELM model has significant advantages in prediction accuracy. The constructed TCN-IFELM model combines the advantages of TCN and IFELM, and its prediction effect is better than that of BP, ELM, IFELM and TCN models. In addition, the NSE indicators of each model are as follows Figure 4 As shown in Figure 2, the NSE index of the model proposed in this invention reaches 0.9580, which is better than other comparison models. It is worth noting that after the original data is preprocessed by DD-ACMD decomposition technology, the prediction effect of the DD-ACMD-TCN-IFELM model is greatly improved. Figure 4 It is intuitively reflected that its NSE index reaches 0.9476, which verifies the effectiveness of the DD-ACMD decomposition technology adopted in the present invention.

[0035] from Figure 5It can be seen that the TCN-IFELM fusion model provided by the present invention has high robustness in runoff prediction, with the least outliers. From the comparison between the prediction results and the measured values, the DD-ACMD-IPSA-TCN-IFELM model (that is, the TCN-IFELM fusion model) provided by the present invention shows excellent fitting ability, such as Figure 6 As shown in the figure, the model is highly consistent with the measured values ​​during the period of flat runoff. When the runoff surges, although there is a certain deviation, the overall predicted value is basically consistent with the actual value, which fully proves the effectiveness of the model.

[0036] The present invention also provides a reservoir basin daily runoff prediction system, comprising the following modules: A daily runoff data acquisition module, wherein the daily runoff data is used to pre-acquire the daily runoff data of the reservoir basin monitoring station, use the isolation forest to perform outlier detection on the daily runoff data, and perform data cleaning; A data processing module, wherein the data processing module is used to decompose the daily runoff data processed by the daily runoff data acquisition module using data-driven adaptive chirp modal decomposition to obtain multiple components; construct an input matrix for each component and divide it into a training set and a test set; A TCN-IFELM fusion model construction and training module, wherein the TCN-IFELM fusion model construction and training module is used to construct a temporal convolutional network and an inverse extreme learning machine to obtain a TCN-IFELM fusion model, and input the training set obtained by the data processing module into the TCN-IFELM model for training; A TCN-IFELM fusion model optimization module, wherein the TCN-IFELM fusion model optimization module is used to introduce half uniform initialization and variable spiral update strategies into the PID search algorithm to obtain an improved PID search algorithm IPSA; the IPSA algorithm is used to optimize the discard rate, the number of residual blocks, and the number of hidden layer neurons of the TCN-IFELM model constructed by the TCN-IFELM fusion model construction and training module; The prediction module is used to input the test set of each component obtained by the data processing module into the TCN-IFELM fusion model optimized by the TCN-IFELM fusion model optimization module to perform component prediction, add the prediction results of each component, and obtain the predicted value of daily runoff at a certain time in the future.

[0037] The present invention also provides an electronic device, the electronic device comprises a processor, a communication interface, a memory and a communication bus, wherein the processor, the communication interface and the memory communicate with each other via the communication bus. A memory, the memory being used to store a computer program; A processor is used to execute the computer program stored in the memory to implement the steps of the reservoir basin daily runoff prediction method as described above.

[0038] The present invention also provides a non-transitory readable storage medium, in which a program is stored. When the program is executed by a processor, the steps of the reservoir basin daily runoff prediction method described above are implemented.

[0039] So far, the technical solutions of the present invention have been described in conjunction with the specific experimental processes shown in the accompanying drawings, but the protection scope of the present invention is not limited to these specific implementations. Without departing from the principles of the present invention, those skilled in the art may make equivalent changes or substitutions to the relevant technical features, and the technical solutions after these changes or substitutions will fall within the protection scope of the present invention.

Claims

1. A reservoir basin daily runoff prediction method, characterized in that: The steps include: S1. Obtain daily runoff data from monitoring stations in the reservoir basin in advance, use isolation forest to detect outliers in the daily runoff data, and perform data cleaning; S2, using data-driven adaptive chirp modal decomposition to decompose the daily runoff data processed by step S1 to obtain multiple components; constructing an input matrix for each component and dividing it into a training set and a test set; S3, constructing a TCN-IFELM fusion model that integrates the temporal convolutional network and the non-inverse extreme learning machine, and inputting the obtained training set into the TCN-IFELM model for training; S4. Introduce half-uniform initialization and variable spiral update strategy into the PID search algorithm to obtain the improved PID search algorithm IPSA; use the IPSA algorithm to optimize the discard rate, the number of residual blocks and the number of hidden layer neurons of the TCN-IFELM model to improve the prediction performance of the TCN-IFELM model; S5. Input the test set into the optimized fusion model for component prediction, and add the prediction results of each component to obtain the predicted value of daily runoff at a certain moment in the future.

2. The method according to claim 1, characterized in that In step S1, the implementation process of using isolation forest to detect outliers in daily runoff data is as follows: S11. Build an Isolation Forest ; in, Represents a binary tree; w is the number of binary trees; S12. From daily runoff data V Randomly select Q The data points form a subspace X , randomly selected X A data feature is selected, and a cutting point is randomly selected in the sample interval corresponding to the feature. A hyperplane is constructed from the cutting point to convert X Split into two different branches under the current node; S13, recursively execute step S12 for the child nodes under two different branches until the data cannot be further split; S14. Calculate and identify daily runoff data sets V Medium data point v The degree of abnormality : in, For data points v The depth of a node in a binary tree, For data points v exist w The average path length on the tree, Reason Q The average path length of the tree built by points; S15. Cut out outlier and abnormal data points and determine outliers.

3. The method according to claim 1, characterized in that The specific process of step S2 is as follows: S21. The high-frequency mode is enhanced by taking the derivative of the daily runoff time series signal. Its derivative can be expressed as: in, represents the original daily runoff time series signal; For signal The derivative of K Represents the total number of modes in the signal; For the i The amplitude of each mode; No. i The phase function of each mode; For the i The instantaneous frequency of each mode; for The derivative of t represents the time variable; S22, estimating the initial instantaneous frequency of the highest frequency mode by normalizing the derivative of the signal, the specific process is as follows: in, After normalization, i A frequency modulation signal of a mode; No. i The phase function of each mode; represents the instantaneous frequency of the highest frequency mode; d It means differential, which refers to a very small amount of change; dt Indicates a small interval of time; is a differential operator, which represents the time t The derivative of S23, introduce a time-varying low-pass filter to remove noise above the estimated instantaneous frequency of the current signal, the process is as follows: S231, rewrite the frequency modulation signal as: in, Indicates frequency modulated signal; Represents a time interval; express An infinitesimal change in Indicates that the signal is at point The instantaneous frequency of Indicates a time interval On, instantaneous frequency Small changes in represents the initial phase; S232, calculating the analytical signal of the frequency modulation signal: in, for The analytical signal of Indicates frequency modulated signal; represents the initial phase; express Hilbert transform; j represents the imaginary part of the signal; S233. Demodulation operator definition and modulation operator : in, is the demodulation frequency; Indicates the carrier frequency; S234, multiplying the analytical signal by the demodulation operator to obtain a demodulated signal: in, represents the demodulated signal; S235, remove the noise higher than the carrier frequency through low-pass filtering, and then multiply it by the modulation operator to obtain a new analytical signal: in, Represents the modulated signal; is the demodulated signal after low-pass filtering; S24. After each pattern extraction, a time-varying low-pass filter is used to remove the noise above the pattern to reduce the impact on subsequent pattern extraction. The signal patterns are extracted from high frequency to low frequency through a recursive framework. After each pattern is extracted, the remaining signals are updated and the above process is repeated until the stopping condition is met. DD-ACMD can decompose the daily runoff time series signal into independent components, which is convenient for subsequent analysis and prediction.

4. The method according to claim 1, characterized in that: The specific process of step S3 is as follows: S31. TCN is used to capture the dependencies in the daily runoff time series. The specific process of dilated convolution is as follows: in, is the input sequence; f is the filter; d is the expansion factor; z is the convolution kernel position; is the convolution kernel size; Indicates the element of the previous layer; S32, using the residual structure to input The output of the convolutional network Perform addition operation, the calculation formula is: in, is the activation function; S33. Define the hidden node adding strategy of IFELM: in, l is the number of current hidden nodes; and yes l The input weights and biases for hidden nodes are is the newly added input weight, is the newly added bias; T Represents a transpose operation; S34. Update the output weights according to the following rules: in, The output weight matrix after adding a hidden node; Y Represents the training output matrix; Indicates the current l The pseudo-inverse matrix of hidden nodes; It is the pseudo-inverse matrix after adding a hidden node; and for Two blocks of are used to update the pseudo-inverse matrix; c Represents the vector after adding a hidden node; H represents the hidden layer output matrix; I is the identity matrix; Representation vector c The transpose of S35. Construct a TCN-IFELM fusion model. Use TCN to extract deep features between runoff data, convert the input high-dimensional data into low-dimensional feature representation, and IFELM takes the features extracted by TCN as input and the corresponding runoff as target output in turn. IFELM avoids the matrix inversion operation in the traditional extreme learning machine by gradually adding hidden nodes and updating weights, thereby improving computational efficiency.

5. The method according to claim 1, characterized in that: The specific process of step S4 is as follows: S41, setting the population size, maximum number of iterations and interval range of the parameters to be optimized of the IPSA algorithm; S42, using the half-uniform initialization strategy to improve the population initialization part of the original PSA algorithm, the specific process is as follows: in, represents the number of iterations, and are the initial values ​​of the individuals, , ; up and low Indicates the upper and lower limits of a variable; n is the population size; is a random number between 0 and 1; the half-uniform initialization strategy not only retains the randomness of the population, but also avoids the concentrated distribution of initialized individuals, which is conducive to improving the diversity of the population; S43, defining the objective function of the algorithm as the root mean square error between the daily runoff prediction value and the actual value, and calculating the fitness value of the population through the objective function; S44. Calculate system deviation: the population in the number of iterations Deviation of time Defined as: in, Represents the best individual value in the population; Indicated in Individual values ​​during iteration; S45, PID adjustment process: number of iterations Output value of PID regulation Defined as: in, , and is a vector of random numbers between 0 and 1; , and are the adjustment coefficients for proportional, integral and differential respectively; S46, population update: In order to prevent the algorithm from falling into the local optimum, the variable spiral update strategy is used to improve the population update process of the PSA algorithm. The specific process is as follows: in, is the value after individual disturbance; Represents the average fitness of the top 10% individuals. is the dynamic weight; b To adjust the parameters of the spiral shape; is a random number uniformly distributed between [-1, 1]; D For individuals and The improved population update formula is: in, is a random number between 0 and 1; It means the number of iterations is The output value of PID regulation when ; Indicates the value of the individual at the next iteration; the variable spiral update strategy increases the exploration of local areas while maintaining the global search capability of the algorithm, thereby improving the search accuracy and convergence speed of the algorithm; S47, the IPSA algorithm is used to optimize the discard rate, the number of residual blocks and the number of hidden layer neurons of the TCN-IFELM model, so that the root mean square error between the daily runoff prediction value and the actual value is continuously reduced; S48. Determine whether the algorithm termination condition is reached through a given maximum number of iterations, and finally output the optimal parameters of the TCN-IFELM model within the maximum number of iterations.

6. A reservoir basin daily runoff prediction system, characterized in that: Includes the following modules: A daily runoff data acquisition module, wherein the daily runoff data is used to pre-acquire the daily runoff data of the reservoir basin monitoring station, use the isolation forest to perform outlier detection on the daily runoff data, and perform data cleaning; A data processing module, wherein the data processing module is used to decompose the daily runoff data processed by the daily runoff data acquisition module using data-driven adaptive chirp modal decomposition to obtain multiple components; construct an input matrix for each component and divide it into a training set and a test set; A TCN-IFELM fusion model construction and training module, wherein the TCN-IFELM fusion model construction and training module is used to construct a TCN-IFELM fusion model that integrates a temporal convolutional network and an inverse extreme learning machine, and input the training set obtained by the data processing module into the TCN-IFELM model for training; A TCN-IFELM fusion model optimization module, wherein the TCN-IFELM fusion model optimization module is used to introduce half uniform initialization and variable spiral update strategies into the PID search algorithm to obtain an improved PID search algorithm IPSA; the IPSA algorithm is used to optimize the discard rate, the number of residual blocks, and the number of hidden layer neurons of the TCN-IFELM model constructed by the TCN-IFELM fusion model construction and training module; The prediction module is used to input the test set of each component obtained by the data processing module into the TCN-IFELM fusion model optimized by the TCN-IFELM fusion model optimization module to perform component prediction, add the prediction results of each component, and obtain the predicted value of daily runoff at a certain time in the future.

7. An electronic device, comprising a processor, a communication interface, a memory and a communication bus, wherein the processor, the communication interface and the memory communicate with each other via the communication bus, characterized in that: A memory, the memory being used to store a computer program; A processor, wherein the processor is used to execute a computer program stored in a memory to implement the steps of the reservoir basin daily runoff prediction method as described in any one of claims 1 to 5.

8. A non-transitory readable storage medium, characterized in that: The non-transitory readable storage medium stores a program, and when the program is executed by the processor, the steps of the reservoir basin daily runoff prediction method are implemented as described in any one of claims 1-5.

Images