A method for filling tide data of ocean stations based on IALSTM network

Abstract

The present invention discloses a method for filling tide data of ocean stations based on the IALSTM network, comprising: obtaining tide data of a target ocean station and associated ocean stations; using part of the tide data of the associated ocean stations as training data to train an IALSTM model; and completing the filling of missing tide data of the target ocean station based on the IALSTM model. The method is based on the tide data of the target ocean station and the tide data of related stations, and establishes an association relationship through an IALSTM deep learning neural network model, thereby filling the missing tide data of the ocean station with high precision based on neighboring stations, thereby improving the integrity and continuity of the data. The method is significantly superior to traditional interpolation methods, and can effectively fill in missing values ​​or unobserved values ​​of tide time series, greatly improving the quality of tide data and greatly improving data utilization efficiency.

Description

Technical Field

[0001] The invention belongs to the field of marine science and technology, and in particular relates to a method for filling tide data of an ocean station based on an IALSTM network. Background Art

[0002] Tidal data is a time series data observed by ocean stations. It reflects the changes in sea level rise and fall and has very important applications in disaster prevention and mitigation, urban planning, transportation, coastal tourism, etc. However, due to various reasons such as human operation errors, adverse environmental influences, or instrument failures, the collected tidal data will inevitably have data missing problems, thus affecting the quality of the observed data. Therefore, effectively filling in the missing values ​​or unobserved values ​​of the tidal time series can improve the quality of tidal data and enhance the efficiency of data use.

[0003] The lack of tidal data brings inconvenience to the use, and the traditional filling methods include mathematical methods such as linear interpolation, nearest neighbor interpolation, polynomial interpolation and spline interpolation. However, these traditional filling methods are highly dependent, low in accuracy, and greatly affected by the lack of data.

[0004] Therefore, how to solve the missing tidal data and fill it with more accurate data has become an urgent problem to be solved by practitioners in the same field. Summary of the invention

[0005] The object of the present invention is to provide a method for filling tide data of ocean stations based on the IALSTM network, which can solve the problem of missing tide data and can better fill data with higher accuracy.

[0006] To achieve the above object, the technical solution adopted by the present invention is:

[0007] The present invention provides a method for filling tide data of ocean stations based on IALSTM network, comprising:

[0008] Obtain tide data of target ocean stations and associated ocean stations;

[0009] Using part of the tide level data of the associated ocean station as training data to train the IALSTM model;

[0010] The missing tidal data of the target ocean station are filled based on the IALSTM model.

[0011] Furthermore, the tidal data of the target ocean station and the associated ocean stations are obtained, including:

[0012] The tidal level data of the target ocean station and the associated ocean station are obtained and normalized to ensure that the tidal level data of the target ocean station and the tidal level data of the associated ocean station are data at the same time.

[0013] Furthermore, obtaining the tidal data of the target ocean station and the associated ocean stations also includes:

[0014] Through the correlation test method, the correlation coefficient of the tidal data of the target ocean station and the associated ocean station is obtained;

[0015] When the correlation coefficient meets the threshold value, it is confirmed that the tide level data acquired by the target ocean station and the associated ocean station are correlated.

[0016] Furthermore, the IALSTM model includes a multi-dimensional attention module for optimizing the weight vector in the traditional attention mechanism from a scalar to a vector.

[0017] Furthermore, the IALSTM model adopts a single hidden layer neural network with 512 neurons and a learning rate of 0.0001.

[0018] Further, filling in the missing tidal data of the target ocean station is completed based on the IALSTM model; including:

[0019] Inputting the tide level data of the associated ocean stations at the same time into the IALSTM model to obtain output results;

[0020] The output result is used as the filling result of the missing tide level data of the target ocean station at the same time.

[0021] Compared with the prior art, the present invention has the following beneficial effects:

[0022] A method for filling tide data of ocean stations based on the IALSTM network includes: obtaining tide data of a target ocean station and associated ocean stations; using part of the tide data of the associated ocean stations as training data to train an IALSTM model; and completing the filling of missing tide data of the target ocean station based on the IALSTM model. This method is significantly superior to traditional interpolation methods, and can effectively fill in missing values ​​or unobserved values ​​of tide time series, greatly improving the quality of tide data and greatly improving data utilization efficiency. BRIEF DESCRIPTION OF THE DRAWINGS

[0023] Figure 1 A flow chart of the steps of a method for filling tide data of ocean stations based on the IALSTM network provided in an embodiment of the present invention;

[0024] Figure 2 A schematic diagram of an implementation method of a method for filling tide data of an ocean station based on an IALSTM network provided in an embodiment of the present invention;

[0025] Figure 3 A structural diagram of the IALSTM model provided by an embodiment of the present invention;

[0026] Figure 4a A comparison chart of filling effects of a one-hour missing measurement provided by an embodiment of the present invention;

[0027] Figure 4b A comparison chart of filling effects of a 6-hour missing measurement provided by an embodiment of the present invention;

[0028] Figure 4c A comparison chart of filling effects of a 24-hour missing measurement provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0029] In order to make the technical means, creative features, objectives and effects achieved by the present invention easy to understand, the present invention is further explained below in conjunction with specific implementation methods.

[0030] In the description of the present invention, it should be noted that the terms "upper", "lower", "inner", "outer", "front end", "rear end", "two ends", "one end", "the other end" and the like indicate positions or positional relationships based on the positions or positional relationships shown in the drawings, and are only for the convenience of describing the present invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore cannot be understood as limiting the present invention. In addition, the terms "first" and "second" are used for descriptive purposes only and cannot be understood as indicating or implying relative importance.

[0031] In the description of the present invention, it should be noted that, unless otherwise clearly specified and limited, the terms "installed", "provided with", "connected", etc. should be understood in a broad sense. For example, "connected" can be a fixed connection, a detachable connection, or an integral connection; it can be a mechanical connection or an electrical connection; it can be a direct connection, or it can be indirectly connected through an intermediate medium, or it can be the internal communication of two components. For ordinary technicians in this field, the specific meanings of the above terms in the present invention can be understood according to specific circumstances.

[0032] In recent years, machine learning methods have developed rapidly and have been widely used in various disciplines. As a variant model of recurrent neural networks, LSTM (Long Short-Term Memory Network) can effectively utilize long-distance temporal information and is often used for time series data prediction, but it also has the disadvantage of easily ignoring low-weight signals. The present invention designs an IALSTM network (Information Aware Long Short-Term Memory Network) based on the LSTM deep learning network, and establishes an association model of tide data of neighboring ocean stations, thereby filling in missing data based on data from neighboring stations. The model has a mature structure, stable performance, simple implementation, and high accuracy, thereby achieving effective precision filling of missing tide data at ocean stations and improving the integrity and continuity of the data.

[0033] Reference Figure 1 As shown, the present invention provides a method for filling ocean station tide data based on the IALSTM network, comprising:

[0034] S10, obtaining tide level data of the target ocean station and associated ocean stations;

[0035] S20, using part of the tide level data of the associated ocean station as training data to train the IALSTM model;

[0036] S30. Filling in the missing tidal data of the target ocean station based on the IALSTM model.

[0037] In this embodiment, based on the tidal level data of the target ocean station and the tidal level data of related stations, an association relationship is established through the IALSTM deep learning neural network model, so that the missing tidal level data of the ocean station can be filled with high precision based on the neighboring stations, thereby improving the integrity and continuity of the data.

[0038] The scheme of the present invention is described in detail below with reference to the figures:

[0039] like Figure 2 This is a schematic diagram of the implementation of the method for filling the ocean station tide data based on the IALSTM network of this embodiment. For example, the tide data used in this embodiment comes from the two ocean stations of Chongwu and Jinjiang in Fujian Province, from January 1, 2017 to December 31, 2017. The tide data of Jinjiang station is set as output data, and the tide data of Chongwu station is set as input data, that is, the tide data of Jinjiang is predicted using the Chongwu tide data to establish a model. The length of the input data sequence is 8760 groups, of which 7008 groups (80%) are used for training, 876 groups (10%) are used for verification, and 876 groups (10%) are used for testing.

[0040] In this embodiment, the input and output sample lengths are 1:1, that is, the Chongwu tide data at the same time is used to predict the Jinjiang tide data, and the IALSTM deep learning model is input for training to form a training model. Using this model, when the Jinjiang tide data is missing and the Chongwu data is complete at the same time, this model can be used to predict the Jinjiang tide data at that moment, thereby filling in the Jinjiang tide data.

[0041] The deep learning neural network structure of IALSTM is as follows:

[0042] Based on the traditional LSTM (Long Short Term Memory Network), information perception functions are added, such as Figure 3 As shown in Figure 3, it nests multi-dimensional attention and introduces a fusion gate unit to filter the input features to minimize harmful noise.

[0043] IALSTM uses different functions to calculate the state of the hidden layer. Specifically, it introduces gate functions (three gating mechanisms, namely input gate i t , forget gate f t and output gate o t ), so that it can retain important information for a long time and mine the temporal variation patterns of relatively long intervals and delays in time series.

[0044] The functions of these three gates are:

[0045] ① Forget gate f t Control the internal state c at the previous moment t-1 How much information needs to be forgotten;

[0046] ② Input gate i t Control the candidate state g at the current moment t How much information needs to be saved?

[0047] ③Output gate o t Control the internal state c at the current moment t How much information needs to be output to the external state h t .

[0048] The specific calculation formula is as follows:

[0049] i t =σ(W i ·[h t-1 ,x t ]+b i )

[0050] f t =σ(W f ·[h t-1 ,x t ]+b f )

[0051] o t =σ(W o ·[h t-1 ,x t ]+b o )

[0052] cs t =tanh(W g ·[h t-1 ,x t ]+b g )

[0053]

[0054]

[0055] cst Indicates candidate status, hs t represents the external state, W and b are the corresponding weight coefficient matrix and bias term respectively; σ and tanh are the sigmoid and hyperbolic tangent activation functions respectively.

[0056] The attention mechanism can make the strong signals in the input data more prominent, making it easier to predict the target value. However, the disadvantage is that it is likely to introduce high-weight noise, and when the weight is too small, the entire input may be discarded. Based on this, this embodiment designs a multi-dimensional attention mechanism to optimize the weight vector in the traditional attention mechanism from a scalar to a vector. This modification provides a more refined feature extraction strategy. The data processed by the multi-dimensional attention mechanism is processed by the fusion gate and the candidate state cs t Fusion, generating external state hs t .

[0057] The specific implementation process of this embodiment is as follows:

[0058] First, the tide data of the target ocean station and the associated ocean station are normalized to ensure that the tide data of the target ocean station and the tide data of the associated ocean station are data at the same time; then confirm the correlation: the tide data of Chongwu Station and Jinjiang Station in this embodiment are tested by Pearson correlation, and the correlation coefficient is 0.9998 (greater than the threshold of 0.8, and the threshold can be autonomously adjusted according to the specific application scenario), indicating that the two tide time series are highly correlated, and the next step of model construction can be carried out. It should be noted here that the better the correlation, the better the model effect.

[0059] Secondly, a period of complete data is found in the tidal data of Chongwu Station and Jinjiang Station for building the model. This embodiment selects the data of 2017, and the data of this period is complete and of good quality.

[0060] Then, the IALSTM model is trained and the model parameters are debugged. This model uses a single hidden layer neural network with 512 neurons and a learning rate of 0.0001 to build the IALSTM model.

[0061] Finally, based on this model, the Jinjiang tidal data is filled. The usage method is: find the missing data of Jinjiang tidal level. If the Chongwu data at this moment is complete, then input the Chongwu data at this moment into the IALSTM model to obtain the output result, which can be used to supplement the missing data of Jinjiang tidal level. It is worth mentioning that the missing rate of tidal data at Chongwu station is 1.72%, and the missing rate of tidal data at Jinjiang station is 0.2%. It is rare that the tidal data of both stations are missing at a certain moment.

[0062] To verify the effect of this model, an experiment was designed and compared with the traditional filling method. Figure 4a-4cAs shown. Experiments were set up with missing durations of Jinjiang tide data of 1 hour, 6 hours, and 24 hours, and the filling effects of (1) astronomical tide replacement method, (2) linear interpolation method, (3) spline interpolation method, and (4) IALSTM filling method were calculated respectively. In this embodiment, the root mean square error (RMSE) was used to evaluate the filling effect of the ocean station tide data, and the calculation formulas were as follows.

[0063]

[0064] Among them, x real is the actual tide level, x interp is the predicted value obtained by interpolation, and n is the sample size. RMSE measures the deviation between the filled value and the true value. The smaller the RMSE, the better the filling effect and the higher the accuracy.

[0065] Table 1 Root mean square error RMSE (cm) of different filling methods

[0066] Missed test duration (h) astronomical tide level Linear interpolation Spline interpolation IALSTM 1 37.65 3.46 2.29 2.83 5 37.65 16.20 6.23 2.83 24 37.65 24.10 43.50 2.83

[0067] As shown in Table 1, when the data is missing for a long time, IALSTM has excellent filling effect (RMSE = 2.83cm), which is much better than the conventional interpolation method (linear difference RMSE = 24.10cm, spline difference RMSE = 43.50cm). At this time, the general interpolation method is no longer applicable to data filling, but IALSTM is not affected by this.

[0068] In fact, tide data observation and recording are now completed automatically by instruments. The missing data is mainly caused by instrument damage. In this case, the data missing time interval is generally very long, even as long as a month. For such a long series of missing data, ordinary interpolation methods are no longer applicable, and the advantages of the method of using LSTM deep learning neural network to learn and predict time series are reflected.

[0069] In addition, for this embodiment, the Jinjiang tide data can be used as input and the Chongwu tide data can be used as output, and an IALSTM model can be established to fill the Chongwu tide data based on the Jinjiang tide data.

[0070] The positive effects of the method provided in this embodiment are:

[0071] (1) High-precision filling of missing tidal data at ocean stations is achieved, and it is independent of the duration of data missing. Especially when the duration of data missing is long, this aspect of the present invention is significantly superior to the traditional difference method and can reach the international advanced level.

[0072] (2) The present invention only needs to use the long-term series data of the tide level of the ocean station, and does not need to use other data.

[0073] (3) The present invention occupies less resources and has a fast calculation speed.

[0074] (4) The present invention can be used for, but is not limited to, filling in tidal data at ocean stations. It can also be used for filling in missing data at fixed points in other scenarios, such as seawater temperature, salinity, air temperature, air pressure, etc.

[0075] Obviously, those skilled in the art can make various changes and modifications to the present invention without departing from the spirit and scope of the present invention. Thus, if these modifications and variations of the present invention fall within the scope of the claims of the present invention and their equivalents, the present invention is also intended to include these modifications and variations.

Claims

1. A method for filling ocean station tide data based on IALSTM network, characterized in that: include: Obtain tide data of target ocean stations and associated ocean stations; And normalize them to ensure that the tidal data of the target ocean station and the tidal data of the associated ocean station are data at the same time; obtain the correlation coefficient of the tidal data of the target ocean station and the associated ocean station through a correlation test method; when the correlation coefficient meets the threshold, it is confirmed that the tidal data of the target ocean station and the associated ocean station are correlated; Using part of the tide level data of the associated ocean station as training data to train the IALSTM model; the IALSTM model includes a multi-dimensional attention module for optimizing the weight vector in the traditional attention mechanism from a scalar to a vector; The multi-dimensional attention module introduces a fusion gate unit to filter the input features to minimize harmful noise; The missing tidal data of the target ocean station are filled based on the IALSTM model.

2. A method for filling ocean station tide data based on IALSTM network according to claim 1, characterized in that: The IALSTM model adopts a single hidden layer neural network with 512 neurons and a learning rate of 0.0001.

3. A method for filling ocean station tide data based on IALSTM network according to claim 2, characterized in that: Filling the missing tidal data of the target ocean station based on the IALSTM model; including: Inputting the tide level data of the associated ocean stations at the same time into the IALSTM model to obtain output results; The output result is used as the filling result of the missing tide level data of the target ocean station at the same time.

Images