CN116485003A - Multi-step channel water level prediction method and device based on echo algorithm and storage medium - Google Patents
Multi-step channel water level prediction method and device based on echo algorithm and storage medium Download PDFInfo
- Publication number
- CN116485003A CN116485003A CN202310199177.8A CN202310199177A CN116485003A CN 116485003 A CN116485003 A CN 116485003A CN 202310199177 A CN202310199177 A CN 202310199177A CN 116485003 A CN116485003 A CN 116485003A
- Authority
- CN
- China
- Prior art keywords
- water level
- state
- prediction
- kernel
- echo
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
- XLYOFNOQVPJJNP-UHFFFAOYSA-N water Substances O XLYOFNOQVPJJNP-UHFFFAOYSA-N 0.000 title claims abstract description 147
- 238000004422 calculation algorithm Methods 0.000 title claims abstract description 76
- 238000000034 method Methods 0.000 title claims abstract description 61
- 238000012549 training Methods 0.000 claims abstract description 45
- 125000004122 cyclic group Chemical group 0.000 claims description 33
- 239000011159 matrix material Substances 0.000 claims description 22
- 238000010276 construction Methods 0.000 claims description 5
- 230000003595 spectral effect Effects 0.000 claims description 5
- 230000004913 activation Effects 0.000 claims description 3
- 238000009825 accumulation Methods 0.000 abstract description 10
- 230000007774 longterm Effects 0.000 abstract description 9
- 238000013528 artificial neural network Methods 0.000 abstract description 6
- 230000002265 prevention Effects 0.000 abstract description 5
- 230000005540 biological transmission Effects 0.000 abstract description 2
- 230000008569 process Effects 0.000 description 9
- 230000006870 function Effects 0.000 description 7
- 238000002474 experimental method Methods 0.000 description 6
- 210000002569 neuron Anatomy 0.000 description 5
- 230000008859 change Effects 0.000 description 4
- 230000000694 effects Effects 0.000 description 4
- 238000005516 engineering process Methods 0.000 description 4
- 230000008901 benefit Effects 0.000 description 3
- 238000012545 processing Methods 0.000 description 3
- 230000006403 short-term memory Effects 0.000 description 3
- 230000006399 behavior Effects 0.000 description 2
- 238000004364 calculation method Methods 0.000 description 2
- 230000008878 coupling Effects 0.000 description 2
- 238000010168 coupling process Methods 0.000 description 2
- 238000005859 coupling reaction Methods 0.000 description 2
- 238000011161 development Methods 0.000 description 2
- 230000007787 long-term memory Effects 0.000 description 2
- 238000007726 management method Methods 0.000 description 2
- 230000015654 memory Effects 0.000 description 2
- 230000000306 recurrent effect Effects 0.000 description 2
- 238000012360 testing method Methods 0.000 description 2
- 101100136092 Drosophila melanogaster peng gene Proteins 0.000 description 1
- WQZGKKKJIJFFOK-GASJEMHNSA-N Glucose Natural products OC[C@H]1OC(O)[C@H](O)[C@@H](O)[C@@H]1O WQZGKKKJIJFFOK-GASJEMHNSA-N 0.000 description 1
- 238000013459 approach Methods 0.000 description 1
- 238000009360 aquaculture Methods 0.000 description 1
- 244000144974 aquaculture Species 0.000 description 1
- 238000013473 artificial intelligence Methods 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 238000004891 communication Methods 0.000 description 1
- 238000007796 conventional method Methods 0.000 description 1
- 230000001351 cycling effect Effects 0.000 description 1
- 238000011550 data transformation method Methods 0.000 description 1
- 230000003247 decreasing effect Effects 0.000 description 1
- 238000013461 design Methods 0.000 description 1
- 238000010586 diagram Methods 0.000 description 1
- 239000008103 glucose Substances 0.000 description 1
- 238000012423 maintenance Methods 0.000 description 1
- 238000005259 measurement Methods 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000012544 monitoring process Methods 0.000 description 1
- 230000003287 optical effect Effects 0.000 description 1
- 238000011176 pooling Methods 0.000 description 1
- 230000008092 positive effect Effects 0.000 description 1
- 238000001228 spectrum Methods 0.000 description 1
- 238000006467 substitution reaction Methods 0.000 description 1
- 238000012546 transfer Methods 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q10/00—Administration; Management
- G06Q10/04—Forecasting or optimisation specially adapted for administrative or management purposes, e.g. linear programming or "cutting stock problem"
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F17/00—Digital computing or data processing equipment or methods, specially adapted for specific functions
- G06F17/10—Complex mathematical operations
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q50/00—Information and communication technology [ICT] specially adapted for implementation of business processes of specific business sectors, e.g. utilities or tourism
- G06Q50/40—Business processes related to the transportation industry
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02A—TECHNOLOGIES FOR ADAPTATION TO CLIMATE CHANGE
- Y02A90/00—Technologies having an indirect contribution to adaptation to climate change
- Y02A90/30—Assessment of water resources
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Business, Economics & Management (AREA)
- General Physics & Mathematics (AREA)
- Mathematical Physics (AREA)
- Strategic Management (AREA)
- Human Resources & Organizations (AREA)
- Data Mining & Analysis (AREA)
- Economics (AREA)
- General Engineering & Computer Science (AREA)
- Marketing (AREA)
- General Health & Medical Sciences (AREA)
- Tourism & Hospitality (AREA)
- General Business, Economics & Management (AREA)
- Software Systems (AREA)
- Health & Medical Sciences (AREA)
- Operations Research (AREA)
- Game Theory and Decision Science (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Evolutionary Computation (AREA)
- Quality & Reliability (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- Entrepreneurship & Innovation (AREA)
- Artificial Intelligence (AREA)
- Development Economics (AREA)
- Algebra (AREA)
- Computational Mathematics (AREA)
- Mathematical Analysis (AREA)
- Mathematical Optimization (AREA)
- Pure & Applied Mathematics (AREA)
- Databases & Information Systems (AREA)
- Primary Health Care (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The invention discloses a multi-step channel water level prediction method, a device and a storage medium based on an echo algorithm, wherein a multi-step water level prediction model based on the echo algorithm is constructed by utilizing the characteristics of a water level data time sequence, and the random selected connection weight in the traditional echo algorithm is replaced by a Gaussian kernel method, so that the problem of prediction uncertainty is solved; a brand new network structure is constructed, the problem of low information transmission efficiency of storage states in the neural network is solved, so that the kernel storage states can express richer information characteristics, and information is transmitted between the kernel storage states more effectively; the problem of error accumulation in medium-long-term water level prediction is solved by an updating method of output errors on training states. The prediction model has high training efficiency and small parameter dependence, can predict the long-term water level condition through the historical water level information, can obtain high precision in the long-term water level prediction, can provide data support for channel management and ship running safety, and can also provide early warning information for flood prevention and drought resistance.
Description
Technical Field
The invention relates to the technical field of intelligent water level prediction, in particular to a multi-step channel water level prediction method and device based on an echo algorithm and a storage medium.
Background
In the transportation industry, waterway transportation is one of the most important transportation modes in commodity exchange due to the characteristics of large carrying capacity and low carrying cost. The river resources in China are rich, including natural rivers such as Yangtze river, zhujiang river and yellow river, and artificially excavated rivers such as Beijing Hangzhou canal. The river not only provides water for life and agriculture, but also has the functions of waterway transportation, flood control, drought resistance, aquaculture, sightseeing and travelling and the like. The Yangtze river is the first Yangtze river in China, and the main stream and most of tributaries of the Yangtze river are suitable for ship traffic, so that the Yangtze river has the natural advantage of developing waterway transportation. At present, along with the development and application of technologies such as an electronic channel chart, navigation mark remote sensing and control, water level remote sensing and control and the like, the channel state information acquisition capability and navigation information service capability are further improved.
Meanwhile, the water level of the channel is an important factor affecting the safe navigation of the ship, and also affects the navigation capacity of the channel. The change of the water level not only determines the size of the channel dimension to a certain extent, thereby affecting the tonnage of the navigable ship and the load of ship goods, but also reflects the change of the flood season and the dead water season, and has important reference value for flood prevention and drought resistance departments. Therefore, the real-time observation and accurate prediction of the water level state have important reference significance for ship safety sailing in a channel, flood prevention work in flood season, ecological management work and the like.
In addition, channel water level prediction is a challenging task, and has a profound effect on inland waterway traffic. The method plays an important role in cargo allocation, sailing efficiency and sailing safety of inland sailing ships. The inland water level prediction is scientifically and reasonably carried out, the navigation efficiency of inland waterways can be improved, and the navigation safety of ships is ensured. With the development of telemetry and remote control technology, a large amount of water level data can be acquired through water level observation stations arranged along the inland river. Through the accumulation of large amounts of data, today's theories and techniques in the field of artificial intelligence provide a way to learn and build models from large amounts of data. The method provides a new idea for researching the water level prediction technology.
Journal paper: poyang lake level prediction based on long and short term memory neural network (Guo Yan, lai Xijun. Poyang lake level prediction based on long and short term memory neural network [ J ]. Lake science, 2020,32 (3): 12.DOI: CNKI: SUN: FLKX.0.2020-03-025). The paper adopts a Long short term memory neural network (LSTM-term memory network) to construct a water level prediction model of the Poyang lake. The special processing capacity of LSTM on time series data is utilized, and lake entering flow of 'five rivers' nearby the Yanghu and the Yanghu dry flow are used as inputs, and water levels of different water level stations of the Yanghu are used as outputs to train the model. And finally, the construction of the water level prediction model is completed by adjusting model parameters. However, when facing mass data, the computation complexity of LSTM in each layer of network will increase layer by layer, and LSTM iterates the weights of the network in a gradient-decreasing manner back to propagate the loss function, which also results in great influence on the training efficiency of LSTM; the LSTM also needs to find the optimal parameters in the training process, and the number of the parameters is also more, so that the model also has the problem of parameter dependence; in this method, only a single-step prediction is performed. The prediction of the general time sequence is to predict the medium-long term change condition of the observation target, and only the multi-step prediction has practical application significance.
Journal paper: panjia kou reservoir level prediction study based on PCA-ESN model (Sha, peng Hongyu. Panjia kou reservoir level prediction study based on PCA-ESN model [ J ]. Tangshan university journal, 2020,33 (3): 6.DOI: 10.16160/j.cnki.tsxyxb.2020.03.008). The paper proposes an algorithm that preprocesses data by using a PCA algorithm and predicts reservoir water level through an echo network. The algorithm has the prediction capability of small error and high precision, and has stronger effectiveness and feasibility in water level prediction. Although factors that mainly affect the water level are found by processing the data by PCA and efficient water level prediction is performed using the ESN model, it has the following problems in water level prediction: because of the randomly selected feature of the model weights, the model prediction has uncertainty. The model is in multi-step prediction because the prediction range of the training target data is fixed, which results in a limitation of the number of prediction steps in the prediction. Too many parameters need to be defined, resulting in dependency of the model on the parameters.
Recording the water level of the inland waterway is an important task, and plays a vital role in guiding navigation safety and flood prevention. High-precision water level prediction is helpful for improving the navigation safety of the channel and arranging the maintenance channel. However, in the existing medium-long term prediction algorithm, the problems of unstable prediction, low training efficiency, strong parameter dependence, accumulated errors and the like still exist.
Disclosure of Invention
In view of this, the present invention provides a multi-step channel water level prediction method, device and storage medium based on echo algorithm, and provides a novel network model structure called Weighted Error output cyclic echo kernel-state network (WER-EKSN) for medium-long term water level prediction, which utilizes the advantages of pooling calculation (Reservoir Computing, RC), besides improving the accuracy of medium-long term water level prediction, overcomes the problem of unstable prediction, and reduces the problem of Error accumulation and parameter dependence in multi-step prediction, so that the method can be applied in real scenes.
For this purpose, the invention provides the following technical scheme:
the invention provides a multi-step channel water level prediction method based on an echo algorithm, which comprises the following steps:
acquiring a water level data set of a channel water level station recorded in a preset time period;
building and training a weighted error output cyclic echo kernel state network based on the water level dataset; the weighted error output cyclic echo kernel state network is formed by constructing a kernel storage state by connecting an input layer and a hidden layer by using a Gaussian kernel method on the basis of an echo state network, and single-step prediction of channel water level is performed by using the echo kernel state network to obtain a predicted value of each step; carrying out multi-step prediction on the channel water level by using a weighted error output circulation multi-step algorithm; the weighted error output circulation multi-step algorithm calculates a weight system of the current state by using the prediction error and the weighting coefficient of the last state, and updates the output weight of the current state;
acquiring characteristic data of a channel water level station to be predicted;
and outputting a cyclic echo kernel state network to conduct multi-step channel water level prediction by using a trained weighting error based on the characteristic data of the channel water level station to be predicted, so as to obtain a water level prediction result of the channel water level station.
Further, training the weighted error output cyclic echo kernel state network comprises: an initialization phase and a multi-step prediction loop phase;
in the initialization stage, training the echo kernel state network model through training input features and corresponding target values, and calculating a predicted value, a predicted error and an output weight in the first stage;
and in the multi-step prediction circulation stage, starting from the second prediction state, constructing a new kernel matrix through a kernel storage state algorithm and the weighted error output circulation multi-step algorithm, updating the output weight of each step of state until the circulation is completed with the maximum number of steps of the prediction state, and stopping the training of the ending model.
Further, the kernel stores a state algorithm, including:
s(n+1)＝(1α)s(n)+αF(G(Tr x ，x n+1 ，σ))；
wherein s (n+1) stores a state vector for the n+1th core based on the previous stateInformation and current sample characteristics converted by a kernel method are updated; s (n) represents the n-th core storage state vector; alpha represents a spectral radius parameter; x is x n+1 Represents the n+1st training sample; f (-) represents the activation function tanh of the model; sigma represents a kernel parameter; g (·) represents a Gaussian function; tr x Representing model training features.
Further, constructing a new kernel matrix and updating the output weight of each step of state by a kernel storage state algorithm and the weighted error output cyclic multi-step algorithm, including:
calculating a kernel matrix in the current state by using a kernel storage state algorithm;
a combined storage state generated by combining the input characteristics in the current state with the kernel matrix;
calculating the output weight of the last state;
and updating the output weight of the current state by utilizing the combined storage state, the output weight of the last state and the error.
Further, updating the output weight of the current state by using the combined storage state, the output weight of the previous state and the error comprises:
wherein e p-1 Representing the error in the p-1 state, beta p-1 Is the output weight in the p-1 state, beta p Is the output weight in the p state, S F Representing the combined storage state, y 1 For the target value corresponding to the input characteristic of the initialization stage, p represents the number of steps of the predicted state.
Further, calculating the output weight of the last state includes:
based on the combined stored state matrix, the output weight of the last state is calculated by the following formula:
β＝(S F S F T +I) -1 S F y；
wherein, beta is the output weight of the last state, I is the identity matrix, and y represents the training target value.
Further, the echo kernel state network is utilized to conduct single-step prediction of the channel water level, and a predicted value of each step is obtained, and the method comprises the following steps:
taking the product of the output weight of the current state and the combined storage state matrix as the predicted value of the current step.
The invention also provides a multi-step channel water level prediction device based on the echo algorithm, which comprises:
the data set acquisition unit is used for acquiring a water level data set of the channel water level station recorded in a preset time period;
the model building unit is used for building and training a weighted error output cyclic echo kernel state network based on the water level data set obtained by the data set obtaining unit; the weighted error output cyclic echo kernel state network is formed by constructing a kernel storage state by connecting an input layer and a hidden layer by using a Gaussian kernel method on the basis of an echo state network, and single-step prediction of channel water level is performed by using the echo kernel state network to obtain a predicted value of each step; carrying out multi-step prediction on the channel water level by using a weighted error output circulation multi-step algorithm; the weighted error output circulation multi-step algorithm calculates a weight system of the current state by using the prediction error and the weighting coefficient of the last state, and updates the output weight of the current state;
the characteristic obtaining unit to be tested is used for obtaining characteristic data of the channel water level station to be predicted;
the prediction unit is used for outputting a cyclic echo kernel state network to perform multi-step channel water level prediction by utilizing the weighted error trained by the model construction unit based on the characteristic data of the channel water level station to be predicted acquired by the characteristic acquisition unit to be detected, so as to obtain a water level prediction result of the channel water level station.
The invention also provides a channel water level prediction service interface, which comprises: the input module is used for acquiring characteristic data of the channel water level station;
the prediction module is used for predicting the water level of the channel water level station based on a water level prediction model of the cyclic echo kernel state network output by the weighted error;
and the output module is used for outputting a water level prediction result.
The invention also provides a computer readable storage medium, wherein a computer instruction set is stored in the computer readable storage medium, and when the computer instruction set is executed by a processor, the multi-step channel water level prediction method based on the echo algorithm is realized.
The invention has the advantages and positive effects that:
according to the invention, by utilizing the characteristics of the water level data time sequence, a multi-step water level prediction model based on an echo algorithm is constructed, and the random connection weight selected in the traditional echo algorithm is replaced by a Gaussian kernel method, so that the problem of prediction uncertainty is solved; a brand new network structure is constructed, the problem of low information transmission efficiency of storage states in the neural network is solved, so that the kernel storage states can express richer information characteristics, and information is transmitted between the kernel storage states more effectively; the problem of error accumulation in medium-long-term water level prediction is solved by an updating method of output errors on training states. The prediction model has high training efficiency and small parameter dependence, can predict the long-term water level condition through the historical water level information, can obtain high precision in the long-term water level prediction, can provide data support for channel management and ship running safety, and can also provide early warning information for flood prevention and drought resistance. The water level prediction model can be deployed as a service interface to provide convenient water level data for channel bureaus or crews, and improves the channel information service level and the ship navigation safety.
Drawings
In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings that are required in the embodiments or the description of the prior art will be briefly described, and it is obvious that the drawings in the following description are some embodiments of the present invention, and other drawings may be obtained according to the drawings without inventive effort to a person skilled in the art.
FIG. 1 is a flow chart of a weighted error output recursive multi-step algorithm in an embodiment of the present invention;
FIG. 2 is a flow chart of a predictive model WER-EKSN in accordance with an embodiment of the invention;
FIG. 3 is a diagram showing a comparison of predicted and actual values of a model WER-EKSN in a fifth step according to an embodiment of the present invention.
Detailed Description
In order that those skilled in the art will better understand the present invention, a technical solution in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings in which it is apparent that the described embodiments are only some embodiments of the present invention, not all embodiments. All other embodiments, which can be made by those skilled in the art based on the embodiments of the present invention without making any inventive effort, shall fall within the scope of the present invention.
It should be noted that the terms "first," "second," and the like in the description and the claims of the present invention and the above figures are used for distinguishing between similar objects and not necessarily for describing a particular sequential or chronological order. It is to be understood that the data so used may be interchanged where appropriate such that the embodiments of the invention described herein may be implemented in sequences other than those illustrated or otherwise described herein. Furthermore, the terms "comprises," "comprising," and "having," and any variations thereof, are intended to cover a non-exclusive inclusion, such that a process, method, system, article, or apparatus that comprises a list of steps or elements is not necessarily limited to those steps or elements expressly listed but may include other steps or elements not expressly listed or inherent to such process, method, article, or apparatus.
The invention provides a novel kernel state network structure for replacing the state of a reserve pool in an echo network, and solves the problem of prediction instability. The invention reduces the error accumulation problem in multi-state prediction by using the proposed weighted error output cyclic multi-step algorithm, and improves the long-term prediction accuracy to a certain extent.
The invention is characterized in that: first, an echo kernel state network (Echo Kernel State Network, EKSN) is formed by connecting the input layer and the hidden layer using a gaussian kernel method based on the existing echo state network (Echo State Network, ESN). The method successfully solves the problem of unstable prediction caused by random weight, and expands the dimension of the input features. And secondly, the kernel storage state is constructed by using a kernel method, so that hidden neurons are connected with each other, information among the neurons is transmitted, and history information carried by the previous neurons can be inherited. On the other hand, the construction of the kernel storage state reduces the number of parameters required to be set, so that the parameter dependence problem of the model is reduced. Finally, the invention also provides a new weighted error output circulation multi-step algorithm which is used for multi-step prediction of water level, and the core idea is to calculate the weight coefficient of the current state by utilizing the previous prediction error and the weight coefficient, so as to update the output weight of the state prediction model, and the problem of error accumulation in multi-step prediction can be reduced. An accurate prediction result can be obtained in the middle-long term prediction.
For ease of understanding, the EKSN model in the embodiments of the present invention is described in detail below.
In order to solve the problem of unstable ESN performance caused by random weight selection in an ESN model, the invention designs a novel calculation method of a storage state. This approach not only eliminates the prediction instability of the ESN, but also reduces the dependency of the parameters. In addition, the kernel method can also convert input features from low dimension to high dimension features, and the prediction capability of the model is improved in turn.
The invention adopts the Gaussian kernel method to connect the input layer and the hidden layer, and successfully replaces the influence of random weight on the model prediction result in the traditional ESN algorithm. The kernel characteristic K can be calculated by using a formula (1):
K＝G(Tr x ，X i ，σ)(1)
wherein G (·) represents a Gaussian function, tr x Representing model training features, X i Representing the ith trainingFeature data, σ, represents kernel parameters.
In another aspect, the present invention provides a novel method of calculating a storage state that can obtain higher dimensional features and less parameter dependence than conventional methods. The novel storage state is called as a kernel storage state (Kernel Reservoir State), and the mathematical expression equation is shown in a formula (2):
s(n+1) ＝ (1-α)s(n)+αF(G(Tr x ，x n+1 ， σ)) (2)
wherein s (n) represents the n-th kernel storage state vector, alpha represents the spectral radius parameter, x n+1 Representing the n+1th training sample, F (-) represents the activation function tanh of the model. This formula updates the kernel storage state based on the previous state information and the current sample characteristics converted by the kernel method. The radius of the spectrum as a scaling parameter determines the percentage of information transferred to the current state. In equation (2), the former half represents information of (1α) percent transferred from the former state to the current state, and the latter represents information of (0) percent extracted from the initial kernel<α<1). The initial kernel storage state may be calculated using equation (3).
s(1)＝F(G(Tr x ，x 1 ，σ))(3)
In addition, in order to enrich the hidden layer characteristic data, the invention uses the original characteristic Tr x Generating a combined storage state S by combining with the kernel storage feature matrix S F . The expression formula is as follows:
S F ＝[Tr x ，S(4)
based on the combined storage state matrix, the output weight β of the EKSN model can be calculated by equation (5):
β＝(S F S F T +I) -1 S F y(5)
where I is the identity matrix and y represents the training target value.
According to the above steps, the present invention constructs a new model structure, called EKSN, and is used for single-step prediction. Predictive value of training partCan be expressed by the formula (6):
for ease of understanding, a novel weighted error output cycling multi-step algorithm in accordance with embodiments of the present invention is described in detail below.
In the conventional multi-step prediction algorithm, more predicted values are used as input features as the number of prediction steps increases as training data, which also causes more error features to be learned by the model, resulting in an error accumulation problem. This is also why the prediction accuracy is lower and lower as the number of prediction steps increases.
The present invention provides a novel recursive Algorithm, called Weighted Error output recursive multi-step Algorithm (WER), which applies the output weight beta of the last state p-1 p-1 And error e p-1 Output weight beta for current state p And updating. The flow chart of this algorithm is shown in figure 1 below.
In the WER algorithm, the loop of the first step is based only on the initial input feature X tr，1 And corresponding predicted valuesTo make the next state input feature X tr，2 Is calculated by the computer. It simply combines the input features with the predicted valuesWhen p is>1, the update of the output weight comprises two parts. The first part applies the sum of the prediction error of the previous state and the target value of the current state to calculate the output weight of the current state; the second part uses the radius of the glucose as the weight, setting the ratio of the current and previous state output weights, respectively. The computational mathematical expression is represented by formula (7):
wherein e p-1 Representing the error in the p-1 state.
Based on the EKSN model and the novel weighted output error recursion multi-step algorithm, the invention constructs a multi-step time sequence prediction model, which is called a weighted error output cyclic echo kernel state network (WER-EKSN). As shown in fig. 2, WER-EKSN mainly includes two phases: initialization and multi-step prediction phases.
In the initialisation phase, the feature X is input, mainly by training tr，1 And the corresponding target value y 1 Training the EKSN model to calculate the predicted value of the first stagePrediction error e 1 And output weight beta 1 . To enable prediction of subsequent states, a weighted error output cyclic multi-step algorithm is applied in the subsequent training process, which can be used to generate new input feature values. Equation (8) represents a mathematical expression that generates new input features.
Wherein L represents the size of the training data, P represents the maximum predicted state number of steps, and P represents the predicted state number of steps.
The second stage is a multi-step prediction loop process. Starting from the second predicted state until the loop has completed the maximum number of predicted state steps P. The process mainly comprises the steps of constructing a new kernel matrix and updating the output weight of each step of state by a method for calculating the kernel storage state and a new circulation algorithm. Firstly, calculating a kernel matrix in the state by using a formula (2), and then merging the input characteristics in the state with the kernel matrix to generate a new kernel matrix; secondly, carrying out output weight in the state according to a formula (7); finally, for continued training of the next state, input features for the next state are generated using equation (8). And (5) until the maximum prediction state step number is reached, the training of the ending model is stopped in a circulating way.
The training predictive process of WER-EKSN may be simply presented in algorithm 1.
In the above embodiment, the multi-state cyclic network model for water level station water level data prediction constructs a kernel storage state, and compared with the storage state in an echo algorithm, the multi-state cyclic network model not only replaces randomly selected connection weights by a gaussian kernel method, improves the feature dimension, but also connects neurons in a neural network with each other and transmits information. This is more advantageous for the transfer of features between time series data, increasing the effectiveness of long-term prediction. Meanwhile, a circulation algorithm is output aiming at the weighted errors of multi-step prediction, and the method for updating the model prediction state in the current step is utilized by utilizing the errors in the previous step prediction process. The problem of error accumulation is reduced in multi-step prediction by the model, and the accuracy of medium-long term prediction is improved.
In order to verify the predictive capability of the predictive algorithm provided by the invention on water level, the water level data of the Yangtze river water level stations at six different geographic positions are used in the embodiment of the invention, including Anqing, baizhu, jiujiang, nanjing, tuhu and Zhenjiang. These water level data are the water level conditions at the water level station at the time recorded at eight points a day. In the training process, the size of the prediction range P is set to five, and the size of the time window D is set to twenty. Based on the above settings, and through a data transformation method, a data matrix for model training can be generated. The detailed data information is shown in table 1.
TABLE 1
In order to embody the superiority of the prediction performance of the invention and solve the limitation of the original prediction model, in the embodiment of the invention, WER-EKSN is compared with a cyclic echo network (Recurrent Echo State Network, R-ESN) and a cyclic echo kernel state network model (Recurrent Echo Kernel State Network, R-EKSN). In addition, all data were divided in the different models in proportions of 70% and 30% to generate training data sets and test data sets.
Before model training, in order to enable all the comparison models to exert the optimal prediction effect, the parameter selection of each model is also an important consideration factor for model training. In the training of the R-ESN model, the optimal hidden neuron number is searched out by using a grid searching method with [10, 1000] as a range and 10 as intervals. In R-ESN and R-EKSN, there are also values of the spectral radii to be determined. With [0.05,0.95] as a range, searches were performed at 0.05 intervals to determine the value of the optimal spectral radius. In addition, the parameter set of R-EKSN is the same as that of R-EKSN in order to fairly compare the predicted performance of R-EKSN and WER-EKSN.
Finally, in order to evaluate whether the proposed method and prediction model can successfully solve the problem of prediction instability and reduce error accumulation in mid-long multi-step predictions, two experiments are designed in the embodiment of the present invention. Firstly, the EKSN model is proved to have excellent performance in water level prediction, the problem of prediction uncertainty can be solved, and meanwhile, the dependence on model parameters is reduced. It compares the predicted performance of R-EKSN with R-ESN. Second, the difference in water level prediction exhibited by the conventional multi-step cyclic algorithm and the proposed weighted error output cyclic multi-step algorithm is found by comparing the models R-EKSN and WER-EKSN.
The first experiment demonstrates the performance comparison between a conventional echo algorithm and R-EKSN. The prediction level of the mean variance (Mean Square Error, MSE) and the mean absolute percentage of symmetry (Symmetric Mean Absolute Percentage Error, SMAPE) are measured by their measurement criteria. The formula is expressed as follows:
the performance of these two predictive models in multi-step predictions is shown in table 2 and their performance of average prediction periods (1-5 step states) is calculated. In addition, standard deviation values of the predicted values in the 1-5 step states are given in the last column of the table, which represent the predicted performance differences in the 1-5 prediction periods.
Based on the water level prediction results of the six water level stations, the average prediction value of the model R-EKSN in the 1-5 prediction period is as follows: the average value of MSE is 5.12E-02 and the average value of SMAPE is 22.75%. Among these six data, R-EKSN achieved better prediction in 1-5 cycles of water level prediction relative to the performance of model R-ESN. With the best performing appearing in the data of ninety river. In addition, the performance of R-EKSN in the prediction performance of each step is superior to that of R-ESN. Furthermore, based on the predicted behavior of MSE in five predicted states, the SD value of R-EKSN in all water level data is less than R-ESN, while based on the SD value in SMAPE, R-EKSN also has a smaller value in most data. This indicates that the prediction stability of R-EKSN is greater than that of R-EKSN.
In the second experiment, the performance comparison of R-EKSN and WER-EKSN in five predicted states and average predicted periods thereof was mainly focused on. Table 2 also shows the predicted behavior of the model EKSN based on two different recursive multi-step algorithm types. As in experiment one, the predictive performance of the model was also measured using SMAPE and MSE. Due to the nature of these two types of recursive algorithms, the effect of WER-EKSN and R-EKSN on the first-step prediction state is the same. In the second-step prediction state, the prediction ability of the model WER-EKSN in six data is obviously better than that of the model R-EKSN. Its predicted values at MSE and SMAPE decreased by 1.4E-03 and 1.73%, respectively. In a relatively long-term prediction state (steps 4-5), the predictive power of WER-EKSN is lower than R-EKSN in SMAPE in this prediction state, except for white, but the gap between them is also small: including 0.04% of the fourth step predicted state and 1.73% of the fifth step predicted state. Furthermore, the predictive performance of the model WER-EKSN showed the best performance among the six data sets in the average predictive state. Among these, the largest predicted performance increase occurs in the nine rivers. Compared with the result of R-EKSN, the value of SMAPE in the model provided by the invention is reduced by 0.34%, and the value of MSE is reduced by 0.1E-03. To show the ability of long term water level prediction, the present invention graphically displays the predicted and actual values of the WER-EKSN in the fifth step of the predicted state in FIG. 3. For Anqing and Qing, in the 2500 th sample, the model provided by the invention can accurately predict the place very similar to the actual value, and especially the sample is not in the regular trend of time sequence data. And a linear trend that the predicted value completely coincides with the actual value is shown in a comparison graph of the predicted value and the actual value of the nine river. In the Zhenjiang data table, the prediction value of the model proposed by the present invention does not reach an accurate prediction value, especially the top and bottom regions. The line graph of the remaining data shows almost the same change in test data between actual and predicted values. These demonstrate that the proposed model WER-EKSN has superior capabilities in water level prediction. In addition, the Standard Deviation (SD) values between R-EKSN and WER-EKSN were compared, as shown in the last column of Table 2, in WER-EKSN, only the SD value in the white-ended SMAPE was greater than that of R-EKSN. In the remaining dataset, the SD value of the model WER-EKSN, whether MSE or SMAPE, is less than R-EKSN. Therefore, this also shows that the model predictive stability of WER-EKSN is also superior to R-EKSN. As can be seen from experiments II, the weighted error output cyclic multi-step algorithm not only can improve the water level prediction capability, but also has better prediction stability than the traditional multi-step prediction algorithm.
TABLE 2
/>
/>
/>
The comparison experiment can be used for obtaining: the predictive performance of WER-EKSN is superior to that of R-EKSN and R-ESN, which shows that the EKSN model structure provided by the invention not only improves the predictive accuracy, but also increases the predictive stability. In addition, the weighted error output cyclic multi-step algorithm plays a critical role in multi-step water level prediction. The method reduces the error accumulation of prediction and improves the prediction precision in long-term prediction. Based on the performance of the real water level data in different types, the WER-EKSN shows excellent prediction capability and prediction stability. The method not only obtains accurate prediction effect in different prediction states in all water level data, but also has better stability compared with R-ESN. Therefore, the prediction model provided by the invention is beneficial to monitoring and predicting the water station water level, and can provide data support for keeping sailing safe through the prediction data.
The invention also provides a multi-step channel water level prediction device based on the echo algorithm, which comprises:
the data set acquisition unit is used for acquiring a water level data set of the channel water level station recorded in a preset time period;
the model building unit is used for building and training a weighted error output cyclic echo kernel state network based on the water level data set obtained by the data set obtaining unit; the weighted error output cyclic echo kernel state network is formed by constructing a kernel storage state by connecting an input layer and a hidden layer by using a Gaussian kernel method on the basis of an echo state network, and single-step prediction of channel water level is performed by using the echo kernel state network to obtain a predicted value of each step; carrying out multi-step prediction on the channel water level by using a weighted error output circulation multi-step algorithm; the weighted error output circulation multi-step algorithm calculates a weight system of the current state by using the prediction error and the weighting coefficient of the last state, and updates the output weight of the current state;
the characteristic obtaining unit to be tested is used for obtaining characteristic data of the channel water level station to be predicted;
the prediction unit is used for outputting a cyclic echo kernel state network to perform multi-step channel water level prediction by utilizing the weighted error trained by the model construction unit based on the characteristic data of the channel water level station to be predicted acquired by the characteristic acquisition unit to be detected, so as to obtain a water level prediction result of the channel water level station.
Since the echo algorithm-based multi-step channel water level prediction device according to the embodiment of the present invention corresponds to the echo algorithm-based multi-step channel water level prediction method in the above embodiment, the description is relatively simple, and regarding the similarity, please refer to the description of the echo algorithm-based multi-step channel water level prediction device in the above embodiment, which will not be described in detail herein.
The embodiment of the invention also discloses a computer readable storage medium, wherein a computer instruction set is stored in the computer readable storage medium, and when the computer instruction set is executed by a processor, the multi-step channel water level prediction method based on the echo algorithm provided in any embodiment is realized.
In the several embodiments provided in the present invention, it should be understood that the disclosed technology may be implemented in other manners. The above-described embodiments of the apparatus are merely exemplary, and the division of the units, for example, may be a logic function division, and may be implemented in another manner, for example, a plurality of units or components may be combined or may be integrated into another system, or some features may be omitted, or not performed. Alternatively, the coupling or direct coupling or communication connection shown or discussed with each other may be through some interfaces, units or modules, or may be in electrical or other forms.
The units described as separate parts may or may not be physically separate, and parts displayed as units may or may not be physical units, may be located in one place, or may be distributed on a plurality of units. Some or all of the units may be selected according to actual needs to achieve the purpose of the solution of this embodiment.
In addition, each functional unit in the embodiments of the present invention may be integrated in one processing unit, or each unit may exist alone physically, or two or more units may be integrated in one unit. The integrated units may be implemented in hardware or in software functional units.
The integrated units, if implemented in the form of software functional units and sold or used as stand-alone products, may be stored in a computer readable storage medium. Based on such understanding, the technical solution of the present invention may be embodied essentially or in part or all of the technical solution or in part in the form of a software product stored in a storage medium, including instructions for causing a computer device (which may be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the method according to the embodiments of the present invention. And the aforementioned storage medium includes: a U-disk, a Read-Only Memory (ROM), a random access Memory (RAM, random Access Memory), a removable hard disk, a magnetic disk, or an optical disk, or other various media capable of storing program codes.
Finally, it should be noted that: the above embodiments are only for illustrating the technical solution of the present invention, and not for limiting the same; although the invention has been described in detail with reference to the foregoing embodiments, it will be understood by those of ordinary skill in the art that: the technical scheme described in the foregoing embodiments can be modified or some or all of the technical features thereof can be replaced by equivalents; such modifications and substitutions do not depart from the spirit of the invention.
Claims (10)
1. A multi-step channel water level prediction method based on an echo algorithm, the method comprising:
acquiring a water level data set of a channel water level station recorded in a preset time period;
building and training a weighted error output cyclic echo kernel state network based on the water level dataset; the weighted error output cyclic echo kernel state network is formed by constructing a kernel storage state by connecting an input layer and a hidden layer by using a Gaussian kernel method on the basis of an echo state network, and single-step prediction of channel water level is performed by using the echo kernel state network to obtain a predicted value of each step; carrying out multi-step prediction on the channel water level by using a weighted error output circulation multi-step algorithm; the weighted error output circulation multi-step algorithm calculates a weight system of the current state by using the prediction error and the weighting coefficient of the last state, and updates the output weight of the current state;
acquiring characteristic data of a channel water level station to be predicted;
and outputting a cyclic echo kernel state network to conduct multi-step channel water level prediction by using a trained weighting error based on the characteristic data of the channel water level station to be predicted, so as to obtain a water level prediction result of the channel water level station.
2. The method of claim 1, wherein training the weighted error output cyclic echo kernel-state network comprises: an initialization phase and a multi-step prediction loop phase;
in the initialization stage, training the echo kernel state network model through training input features and corresponding target values, and calculating a predicted value, a predicted error and an output weight in the first stage;
and in the multi-step prediction circulation stage, starting from the second prediction state, constructing a new kernel matrix through a kernel storage state algorithm and the weighted error output circulation multi-step algorithm, updating the output weight of each step of state until the circulation is completed with the maximum number of steps of the prediction state, and stopping the training of the ending model.
3. The method for predicting the water level of a multi-step channel based on an echo algorithm of claim 2, wherein the kernel stores a state algorithm comprising:
s(n+1)＝(1α)s(n)+αF(G(Tr x ，x n+1 ，σ))；
wherein s (n+1) stores a state vector for the n+1th core, based on previous state information and a current sample feature update converted by a core method; s (n) represents the n-th core storage state vector; alpha represents a spectral radius parameter; x is x n+1 Represents the n+1st training sample; f (-) represents the activation function tanh of the model; sigma represents a kernel parameter; g (·) represents a Gaussian function; tr x Representing model training features.
4. A multi-step channel water level prediction method based on an echo algorithm according to claim 3, wherein constructing a new kernel matrix and updating the output weight of each step of state by a kernel storage state algorithm and the weighted error output cyclic multi-step algorithm comprises:
calculating a kernel matrix in the current state by using a kernel storage state algorithm;
a combined storage state generated by combining the input characteristics in the current state with the kernel matrix;
calculating the output weight of the last state;
and updating the output weight of the current state by utilizing the combined storage state, the output weight of the last state and the error.
5. The method for predicting the water level of a multi-step channel based on an echo algorithm of claim 4, wherein updating the output weight of the current state by using the combined stored state, the output weight of the previous state and the error comprises:
wherein e p-1 Representing the error in the p-1 state, beta p-1 Is the output weight in the p-1 state, beta p Is the output weight in the p state, S F Representing the combined storage state, y 1 For the target value corresponding to the input characteristic of the initialization stage, p represents the number of steps of the predicted state.
6. The method for predicting the water level of a multi-step channel based on an echo algorithm of claim 5, wherein calculating the output weight of the previous state comprises:
based on the combined stored state matrix, the output weight of the last state is calculated by the following formula:
β＝(S F S F T +I) -1 S F y；
wherein, beta is the output weight of the last state, I is the identity matrix, and y represents the training target value.
7. The method for predicting the water level of a multi-step channel based on an echo algorithm according to claim 6, wherein the step of predicting the water level of the channel in a single step by using the echo kernel state network to obtain a predicted value of each step comprises the steps of:
taking the product of the output weight of the current state and the combined storage state matrix as the predicted value of the current step.
8. A multi-step channel water level prediction apparatus based on an echo algorithm, the apparatus comprising:
the data set acquisition unit is used for acquiring a water level data set of the channel water level station recorded in a preset time period;
the model building unit is used for building and training a weighted error output cyclic echo kernel state network based on the water level data set obtained by the data set obtaining unit; the weighted error output cyclic echo kernel state network is formed by constructing a kernel storage state by connecting an input layer and a hidden layer by using a Gaussian kernel method on the basis of an echo state network, and single-step prediction of channel water level is performed by using the echo kernel state network to obtain a predicted value of each step; carrying out multi-step prediction on the channel water level by using a weighted error output circulation multi-step algorithm; the weighted error output circulation multi-step algorithm calculates a weight system of the current state by using the prediction error and the weighting coefficient of the last state, and updates the output weight of the current state;
the characteristic obtaining unit to be tested is used for obtaining characteristic data of the channel water level station to be predicted;
the prediction unit is used for outputting a cyclic echo kernel state network to perform multi-step channel water level prediction by utilizing the weighted error trained by the model construction unit based on the characteristic data of the channel water level station to be predicted acquired by the characteristic acquisition unit to be detected, so as to obtain a water level prediction result of the channel water level station.
9. A channel water level prediction service interface, the service interface comprising:
the input module is used for acquiring characteristic data of the channel water level station;
a prediction module for predicting the water level of the channel water level station based on the water level prediction model of the weighted error output cyclic echo kernel state network according to claims 1-7;
and the output module is used for outputting a water level prediction result.
10. A computer readable storage medium, wherein a set of computer instructions is stored in the computer readable storage medium, which when executed by a processor implements the echo algorithm based multi-step channel water level prediction method according to any one of claims 1 to 7.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310199177.8A CN116485003A (en) | 2023-03-03 | 2023-03-03 | Multi-step channel water level prediction method and device based on echo algorithm and storage medium |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310199177.8A CN116485003A (en) | 2023-03-03 | 2023-03-03 | Multi-step channel water level prediction method and device based on echo algorithm and storage medium |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN116485003A true CN116485003A (en) | 2023-07-25 |
Family
ID=87212718
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202310199177.8A Pending CN116485003A (en) | 2023-03-03 | 2023-03-03 | Multi-step channel water level prediction method and device based on echo algorithm and storage medium |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN116485003A (en) |
Citations (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2019094729A1 (en) * | 2017-11-09 | 2019-05-16 | Strong Force Iot Portfolio 2016, Llc | Methods and systems for the industrial internet of things |
| CN209085657U (en) * | 2017-08-02 | 2019-07-09 | 强力物联网投资组合2016有限公司 | For data gathering system related or industrial environment with chemical production technology |
| CN110110243A (en) * | 2019-04-27 | 2019-08-09 | 华南理工大学 | A kind of historical track destination prediction technique based on echo state network |
| CN110414742A (en) * | 2019-08-02 | 2019-11-05 | 大连海事大学 | A kind of freshwater line intelligent Forecasting of more gaging station linkages |
| US10514694B1 (en) * | 2015-07-21 | 2019-12-24 | Hrl Laboratories, Llc | System and method for classifying agents based on agent movement patterns |
| CN111737640A (en) * | 2020-08-17 | 2020-10-02 | 深圳江行联加智能科技有限公司 | Water level prediction method, device and computer readable storage medium |
| CN113128781A (en) * | 2021-04-30 | 2021-07-16 | 大连理工大学 | Distributed industrial energy operation optimization platform for automatically constructing intelligent model and algorithm |
| CN115169740A (en) * | 2022-08-08 | 2022-10-11 | 济南大学 | Sequence prediction method and system of pooled echo state network based on compressed sensing |
| WO2022221712A1 (en) * | 2021-04-15 | 2022-10-20 | Curemetrix, Inc. | Detecting, scoring and predicting disease risk using multiple medical-imaging modalities |
-
2023
- 2023-03-03 CN CN202310199177.8A patent/CN116485003A/en active Pending
Patent Citations (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US10514694B1 (en) * | 2015-07-21 | 2019-12-24 | Hrl Laboratories, Llc | System and method for classifying agents based on agent movement patterns |
| CN209085657U (en) * | 2017-08-02 | 2019-07-09 | 强力物联网投资组合2016有限公司 | For data gathering system related or industrial environment with chemical production technology |
| WO2019094729A1 (en) * | 2017-11-09 | 2019-05-16 | Strong Force Iot Portfolio 2016, Llc | Methods and systems for the industrial internet of things |
| CN110110243A (en) * | 2019-04-27 | 2019-08-09 | 华南理工大学 | A kind of historical track destination prediction technique based on echo state network |
| CN110414742A (en) * | 2019-08-02 | 2019-11-05 | 大连海事大学 | A kind of freshwater line intelligent Forecasting of more gaging station linkages |
| CN111737640A (en) * | 2020-08-17 | 2020-10-02 | 深圳江行联加智能科技有限公司 | Water level prediction method, device and computer readable storage medium |
| WO2022221712A1 (en) * | 2021-04-15 | 2022-10-20 | Curemetrix, Inc. | Detecting, scoring and predicting disease risk using multiple medical-imaging modalities |
| CN113128781A (en) * | 2021-04-30 | 2021-07-16 | 大连理工大学 | Distributed industrial energy operation optimization platform for automatically constructing intelligent model and algorithm |
| CN115169740A (en) * | 2022-08-08 | 2022-10-11 | 济南大学 | Sequence prediction method and system of pooled echo state network based on compressed sensing |
Non-Patent Citations (1)
| Title |
|---|
| ZONGYING LIU 等: "Weighted error-output recurrent echo kernel state network for multi-step water level prediction", 《APPLIED SOFT COMPUTING》, vol. 137, 26 February 2023 (2023-02-26), pages 1 - 14 * |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Wang et al. | Deterministic and probabilistic wind power forecasting using a variational Bayesian-based adaptive robust multi-kernel regression model | |
| CN112541839B (en) | Reservoir storage flow prediction method based on neural differential equation | |
| Saeed et al. | Hybrid bidirectional LSTM model for short-term wind speed interval prediction | |
| Cheng et al. | Augmented convolutional network for wind power prediction: A new recurrent architecture design with spatial-temporal image inputs | |
| CN111353652A (en) | Wind power output short-term interval prediction method | |
| CN105760952A (en) | Load prediction method based on Kalman filtering and self-adaptive fuzzy neural network | |
| CN110837915B (en) | Low-voltage load point prediction and probability prediction method for power system based on hybrid integrated deep learning | |
| CN115688579A (en) | Basin multi-point water level prediction early warning method based on generation of countermeasure network | |
| CN109460874B (en) | Sense wave height prediction method based on deep learning | |
| Lopes et al. | Artificial neural networks approaches for predicting the potential for hydropower generation: a case study for Amazon region | |
| CN112381334B (en) | High-steep slope deformation trend prediction method based on multi-factor fuzzy time sequence | |
| Yao et al. | Wave height forecast method with multi-step training set extension LSTM neural network | |
| CN114021836A (en) | Multivariable reservoir water inflow amount prediction system based on different-angle fusion, training method and application | |
| Kosana et al. | Hybrid wind speed prediction framework using data pre-processing strategy based autoencoder network | |
| CN107330538B (en) | Method for compiling reservoir adaptive scheduling rules under climate change condition | |
| CN112241802A (en) | Interval prediction method for wind power | |
| CN114997636A (en) | Reservoir group flood control scheduling risk analysis method considering model structure and hydrologic prediction double uncertainty | |
| Feng et al. | Stream-flow forecasting based on dynamic spatio-temporal attention | |
| Handoyo et al. | The fuzzy inference system with rule bases generated by using the fuzzy C-means to predict regional minimum wage in Indonesia | |
| Mubarak et al. | Short-term electricity price forecasting using interpretable hybrid machine learning models | |
| CN116485003A (en) | Multi-step channel water level prediction method and device based on echo algorithm and storage medium | |
| CN116865232A (en) | Wind speed error correction-based medium-and-long-term wind power prediction method and system | |
| Shterev et al. | Time series prediction with neural networks: a review | |
| Yu et al. | Analysis and Application of the Spatio-Temporal Feature in Wind Power Prediction. | |
| Nazemi et al. | Fuzzy-stochastic linear programming in water resources engineering |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination |