A method for transferring offshore island and reef elevation based on neural network

Abstract

The present invention discloses a method for transmitting the elevation of offshore islands and reefs based on a neural network. First, GNSS joint survey is performed on the control points of land and islands and reefs, and joint elevation survey is performed on the control points of land to obtain the GNSS / leveling elevation anomaly of land, and then the residual elevation anomaly is obtained by using the EGM2008 gravity field model, a BP neural network is constructed, and the residual elevation anomaly is used to train the neural network, and the residual elevation anomaly of islands and reefs is obtained according to the trained neural network, and finally the elevation of the control points of islands and reefs is obtained by using the GNSS geodetic height results of the joint survey of islands and reefs, the EMG2008 gravity field model and the residual elevation anomaly of islands and reefs. The leveling results obtained by the present invention are high in precision and simple in measurement, which not only solves the problem of short transmission distance of the cross-river leveling elevation method, but also solves the problem that the elevation cannot be transmitted when there is no long-term tide gauge station, and is of great significance for the transmission of offshore island and reef elevation.

Description

Technical Field

[0001] The invention belongs to the field of ocean surveying, and in particular relates to a method for transmitting offshore island and reef elevation based on a neural network. Background Art

[0002] Accurately measuring island and reef topography is one of the basic tasks for building a "digital ocean", and the elevation benchmark of islands and reefs is the basis for island and reef topography measurement. Establishing an island (reef) elevation benchmark that is unified with the land is a major problem facing the current island (reef) surveying and mapping in my country. The traditional land elevation transfer method cannot be directly applied to the practice of long-distance cross-sea elevation transfer, and the existing transfer methods such as cross-river leveling can only adapt to high-short distance transfer within 3.5km [National First and Second Class Leveling Measurement Specifications GB / T 12897-2006,16]. The synchronous tide gauge method requires a long-term tide gauge station to transfer elevation [Research on the Transmission Method of the Elevation Benchmark of Islands and Reefs Outside the Yangtze River Estuary, Wu Jingwen et al., People's Yangtze River, 2021, 52(7), 120-123]. Summary of the invention

[0003] In order to solve the technical problems existing in the known technology, the present invention provides a method for transmitting the elevation of offshore islands and reefs based on a neural network, which does not require a long-term tide gauge station, has a long transmission distance and a high transmission accuracy.

[0004] The technical solution adopted by the present invention to solve the technical problems existing in the known technology is: a method for transmitting the elevation of offshore islands and reefs based on a neural network, which adopts the following steps:

[0005] S1: Conduct GNSS joint survey at land and island control points, and perform leveling at land control points;

[0006] S2: Using the GNSS geodetic height results and level height results on land, obtain the land GNSS / level height anomaly;

[0007] S3: using the GNSS / leveling elevation anomaly of step S2 and the corresponding EGM2008 gravity field model elevation anomaly, calculate the land residual elevation anomaly, denoted as REA;

[0008] S4: construct a BP neural network model, and use the BP neural network model to establish the nonlinear relationship between the north coordinate X and east coordinate Y of the land control point and the residual elevation anomaly;

[0009] S5: using the north coordinate X and the east coordinate Y of the land control point in step S1 and the residual elevation anomaly REA in step S3, training the BP neural network model established in step S4;

[0010] S6: Take the north coordinate X and east coordinate Y of the island control point as input parameters, substitute them into the BP neural network model trained in step S5, and calculate the residual elevation anomaly REA of the point BP ;

[0011] S7: Residual elevation anomaly REA of the island and reef control points calculated using step S6 BP Elevation anomaly of EGM2008 gravity field model for compensating island and reef control points EGM , get the elevation anomaly of the island reef control point ξ=ξ EGM +REA BP ;

[0012] S8: Using the GNSS geodetic height H0 of the island control point obtained in step S1 and the elevation anomaly obtained in step S7, the horizontal height of the island control point can be obtained: h = H0-ξ = H0-ξ EGM -REA BP .

[0013] In the step S4, the BP neural network model is provided with 5 layers, namely, an input normalization layer, an input layer, a hidden layer, an output layer, and an output anti-normalization layer.

[0014] In step S4, the hidden layer has 5 nodes.

[0015] The advantages and positive effects of the present invention are as follows: by selecting appropriate input and output parameters to establish a BP neural network model and training it, the trained BP neural network model is used to predict the residual elevation anomaly of offshore islands and reefs, and then the elevation anomaly under the gravity field and the geoid height are combined to calculate the level height of offshore islands and reefs. The leveling result has high accuracy and is easy to measure. It not only solves the problem of short transmission distance of the cross-river leveling elevation method, but also solves the problem that the elevation cannot be transmitted when there is no long-term tide gauge station, which is of great significance for the transmission of offshore island reef elevation. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] Figure 1 is a flow chart of the present invention;

[0017] Figure 2 This is a schematic diagram of the collected control point data;

[0018] Figure 3 This is a schematic diagram of GNSS / leveling height anomalies of land control points;

[0019] Figure 4 This is a schematic diagram of 1′×1′ grid elevation anomaly under the EGM2008 gravity field model in the measurement area;

[0020] Figure 5 This is a schematic diagram of residual elevation anomalies at land control points;

[0021] Figure 6 Schematic diagram of the constructed BP neural network model;

[0022] Figure 7 This is a schematic diagram of BP neural network model training samples;

[0023] Figure 8 This is a schematic diagram of the BP neural network model test sample;

[0024] Fig. 9 Schematic diagram of predicted elevation anomalies for island and reef control points. DETAILED DESCRIPTION

[0025] In order to further understand the content, features and effects of the present invention, the following embodiments are given as examples and described in detail with reference to the accompanying drawings:

[0026] A certain island reef and nearby land in the south is taken as the measurement research area. The island reef is more than 10 kilometers away from the nearby land and is connected to the land by a seawall. The offshore island reef elevation transfer method based on a neural network provided by the present invention is used to measure the level height of the island reef, and the following steps are adopted:

[0027] S1: Conduct GNSS joint survey and leveling joint survey at control points on land and islands and reefs.

[0028] Specifically, the island and its surrounding land are taken as the survey research area, and the control point data on the land near the island and reef, as well as the control point data on the island and reef, are collected, and GNSS joint survey and leveling joint survey are carried out on the control points on land and islands. Figure 2 , Figure 2 The data of land and island and reef control points are displayed, among which points 1, 2, 3 and 4 in the small boxes are island and reef control points.

[0029] S2: Use the GNSS geodetic height results and level height results on land to obtain the land GNSS / level elevation anomaly.

[0030] Specifically, the GNSS geodetic height H0 of the control point on land is subtracted from the level height h to calculate the GNSS / level height anomaly ξ GNSS / 水准 =H0-h. Please refer to Figure 3 , Figure 3 Land control point GNSS / leveling anomalies are shown.

[0031] S3: Using the GNSS / leveling elevation anomaly from step S2 and the corresponding EGM2008 gravity field model elevation anomaly, calculate the land residual elevation anomaly, denoted as REA.

[0032] The EGM2008 gravity field model is a global gravity field model developed and released by the National Geospatial-Intelligence Agency (NGA) of the United States based on the latest data. It is the global gravity field model with the highest resolution, best accuracy and most orders to date. Figure 4 , Figure 4 The 1′×1′ grid elevation anomaly diagram of the island and its nearby land under the EGM2008 gravity field model is shown.

[0033] First, the EGM2008 gravity field model is used to calculate the elevation anomaly of the land control point under the EGM2008 gravity field model. EGM , then subtract ξ from the GNSS / leveling height anomaly obtained in step S2 EGM , we can get the residual elevation anomaly, which is REA, and the calculation formula is as follows: REA = ξ GNSS / 水准 -ξ EGM . See Figure 5 , Figure 5 Residual elevation anomalies for land control points are shown.

[0034] S4: Construct a BP neural network model and use the BP neural network model to establish the nonlinear relationship between the north coordinate X and east coordinate Y of the land control point and the residual elevation anomaly.

[0035] Specifically, a five-layer BP neural network model is constructed, which includes input normalization layer, input layer, hidden layer, output layer, and output anti-normalization layer. The north coordinate X and east coordinate Y of the land control point are used as input data, and the residual elevation anomaly of the control point is used as output data. The number of hidden layer nodes is set to 5. Figure 6 , Figure 6 The constructed five-layer BP neural network model is demonstrated.

[0036] S5: Using the north coordinate X and east coordinate Y of the land control point in step S1 and the residual elevation anomaly REA in step S3, the BP neural network model established in step S4 is trained.

[0037] Measurement of the study area location Figure 2 As shown in the figure, 24 points among the land control points are selected to construct the training sample set; the remaining 14 points among the land control points are selected to construct the test sample set. Figure 7 and Figure 8 , Figure 7 The BP neural network model training sample is shown. Figure 8 The BP neural network model verification sample is shown.

[0038] The BP neural network model is trained using training samples, and the learning rate, momentum coefficient, iteration generation, number of input normalization layers, number of input layers, number of hidden layers, number of output layers, and number of output anti-normalization layers are set.

[0039] Table 1 BP neural network model setting parameters

[0040] Learning Rate Momentum coefficient Iteration Generation Number of input normalization layers 0.4 0.7 5 2 Number of input layers Number of hidden layers Number of output layers Number of output denormalization layers 2 5 1 1

[0041] The north coordinate X and east coordinate Y in the training sample are used as the input of the BP neural network model. The obtained result is compared with the residual elevation anomaly REA of the output sample data of the training sample to calculate the error. The error is back-propagated to correct the parameters to be determined in the model. After iterative training, the trained BP neural network model is obtained.

[0042] The sample mean and mean error are used as the basic criteria for model analysis and verification. Its calculation formula is:

[0043] m 平 =∑Δ / n; (1)

[0044]

[0045] Among them, n is the number of samples involved in the test, and Δ is the difference between the output value of the BP neural network model and the sample residual elevation anomaly.

[0046] Substitute the test sample data into the trained BP neural network model for testing. After testing, the average value of the test sample is -2.6cm, and the mean error is ±3.9cm. This shows that the training of the BP neural network model meets the requirements.

[0047] S6: Substitute the north coordinate X and east coordinate Y of the island control point in step S1 as input parameters into the BP neural network model trained in step S5 to calculate the residual elevation anomaly REA of the point BP .

[0048] Specifically, four control points on the islands and reefs in the survey research area are selected as prediction points, and prediction samples are constructed. The north coordinates X and east coordinates Y of the control points in the sample data are used as input parameters and substituted into the BP neural network model trained in step S5 to calculate the residual elevation anomaly REA of the control points. BP . See Fig. 9 , Fig. 9 The residual elevation anomaly REA of the island and reef control points predicted by the trained BP neural network model is shown. BP .

[0049] S7: Residual elevation anomaly REA of the island and reef control points calculated using step S6 BPElevation anomaly of EGM2008 gravity field model for compensating island and reef control points EGM , get the elevation anomaly of the island reef control point ξ=ξ EGM +REA BP .

[0050] Specifically, the EGM2008 gravity field model is used to calculate the elevation anomaly of the island and reef control points under the EGM2008 gravity field model. EGM Then, the residual elevation anomaly REA calculated in step S6 is used BP Compensation is performed to obtain the elevation anomaly of the island and reef control point ξ=ξ EGM +REA BP .

[0051] S8: Using the GNSS geodetic height H0 of the island control point obtained in step S1 and the elevation anomaly ξ obtained in step S7, the level height of the island control point can be obtained: h = H0 - ξ = H0 - ξ EGM -REA BP .

[0052] Specifically, the GNSS geodetic height H0 of the island control point obtained in step S1 and the elevation anomaly ξ obtained in step S7 are subtracted to obtain the level height h = H0-ξ = H0-ξ of the island control point. EGM -REA BP .

[0053] The leveling elevation of the island and reef control points measured in step S1 is compared with the elevation calculated by the neural network. The analysis results are as follows:

[0054] Table 2 Comparative analysis results of control points (unit: m)

[0055] statistics |Δ|Maximum 0.034 |Δ|Minimum 0.002 average value -0.004 Mean error ±0.024

[0056] Among them, Δ is the difference between the elevation of the island and the elevation transmitted across the sea based on the neural network.

[0057] It can be seen from the table that the use of cross-sea elevation transmission based on neural networks can achieve a higher transmission accuracy and meet the needs of cross-sea elevation transmission of offshore islands and reefs.

[0058] In summary, the present invention provides a method for transmitting offshore island and reef elevation based on neural network learning, which adopts the following steps:

[0059] S1: Conduct GNSS joint survey at land and island control points, and perform leveling at land control points;

[0060] S2: Using the GNSS geodetic height results and level height results on land, obtain the land GNSS / level height anomaly;

[0061] S3: using the GNSS / leveling elevation anomaly of step S2 and the corresponding EGM2008 gravity field model elevation anomaly, calculate the land residual elevation anomaly, denoted as REA;

[0062] S4: construct a BP neural network model, and use the BP neural network model to establish the nonlinear relationship between the north coordinate X and east coordinate Y of the land control point and the residual elevation anomaly;

[0063] S5: using the north coordinate X and the east coordinate Y of the land control point in step S1 and the residual elevation anomaly REA in step S3, training the BP neural network model established in step S4;

[0064] S6: Take the north coordinate X and east coordinate Y of the island control point as input parameters, substitute them into the BP neural network model trained in step S5, and calculate the residual elevation anomaly REA of the point BP ;

[0065] S7: Calculate the residual elevation anomaly REA of the island and reef control points using step S6 BP Elevation anomaly of EGM2008 gravity field model for compensating island and reef control points EGM , get the elevation anomaly of the island reef control point ξ=ξ EGM +REA BP ;

[0066] S8: Using the GNSS geodetic height H0 of the island control point obtained in step S1 and the elevation anomaly obtained in step S7, the horizontal height of the island control point can be obtained: h = H0-ξ = H0-ξ EGM -REA BP .

[0067] Although the preferred embodiments of the present invention have been described above in conjunction with the accompanying drawings, the present invention is not limited to the above-mentioned specific embodiments, which are merely illustrative and not restrictive. Under the guidance of the present invention, ordinary technicians in this field can make many forms without departing from the scope of protection of the present invention and the claims, all of which fall within the scope of protection of the present invention.

Claims

1. A method for transmitting offshore island and reef elevation based on a neural network, characterized in that: Use the following steps: S1: Conduct GNSS joint survey at land and island control points, and perform leveling at land control points; S2: Using the GNSS geodetic height results and level height results on land, obtain the land GNSS / level height anomaly; S3: using the GNSS / leveling elevation anomaly of step S2 and the corresponding EGM2008 gravity field model elevation anomaly, calculate the land residual elevation anomaly, denoted as REA; S4: construct a BP neural network model, and use the BP neural network model to establish the nonlinear relationship between the north coordinate X and east coordinate Y of the land control point and the residual elevation anomaly; S5: using the north coordinate X and the east coordinate Y of the land control point in step S1 and the residual elevation anomaly REA in step S3, training the BP neural network model established in step S4; S6: Take the north coordinate X and east coordinate Y of the island control point as input parameters, substitute them into the BP neural network model trained in step S5, and calculate the residual elevation anomaly REA of the point BP ; S7: Residual elevation anomaly REA of the island and reef control points calculated using step S6 BP Elevation anomaly of EGM2008 gravity field model for compensating island and reef control points EGM , get the elevation anomaly of the island reef control point ξ=ξ EGM +REA BP ; S8: Using the GNSS geodetic height H0 of the island control point obtained in step S1 and the elevation anomaly obtained in step S7, the horizontal height of the island control point can be obtained: h = H0-ξ = H0-ξ EGM -REA BP .

2. The offshore island and reef elevation transmission method based on neural network according to claim 1 is characterized by: In the step S4, the BP neural network model is provided with 5 layers, namely, an input normalization layer, an input layer, a hidden layer, an output layer, and an output anti-normalization layer.

3. The offshore island and reef elevation transmission method based on neural network according to claim 2 is characterized by: In step S4, the hidden layer has 5 nodes.

Images