A method, system, terminal and medium for predicting water depth

Abstract

The present invention provides a method, system, terminal and medium for predicting water depth, the method comprising: S1. constructing a relationship model between predicted water depth and geographical location and time, and defining a loss function; S2. constructing a training matrix of a training data set based on geographical location and time, and using the converted data set obtained after convolution, excitation and pooling of the training matrix as the input layer of a neural network; S3. constructing a neural network, and obtaining an output layer based on the input layer of the neural network, the output layer being a predicted water depth data set; S4. using the predicted water depth data set and the corresponding actual water depth data set to train the neural network, iteratively optimizing to obtain an optimal parameter matrix with the goal of minimizing the loss function, and predicting water depth based on the neural network with the optimal parameter matrix. The present invention applies a neural network model to an application scenario for predicting ocean water depth, and predicts the water depth corresponding to a given geographical location and time through a trained neural network, thereby realizing an evolutionary analysis of underwater terrain scouring conditions.

Description

Technical Field

[0001] The present invention relates to the field of ocean surveying and mapping technology, and in particular to a method, system, terminal and medium for predicting water depth. Background Art

[0002] The main purpose of the seabed topography and geomorphology survey and research and the comprehensive offshore marine survey is to conduct a comprehensive survey of the offshore seabed topography and geomorphology, obtain detailed basic data on the offshore topography and geomorphology, provide measured data, basic maps and scientific basis for comprehensive ocean evaluation, construction of the "digital ocean" information infrastructure, and national economic and national defense construction, and provide scientific data for marine economic development, marine development and utilization, marine disaster reduction and prevention, marine environmental protection, marine rights and interests maintenance, and sustainable development of the marine economy.

[0003] At present, multi-beam echo sounder is one of the main tools for seabed topography and landform survey and research. By analyzing and processing its measurement data, the seabed depth topography map can be obtained. However, in the actual investigation and research process, it is often necessary to conduct long-term measurements on the same area of ​​the seabed to achieve continuous observation of underwater topography scouring, predict its changing trend and analyze its evolution.

[0004] However, current multi-beam echo sounders can only provide static seabed depth data at a time. In order to achieve continuous observation of the same sea area, the seabed depth can only be measured again and again, which is time-consuming and labor-intensive, and cannot make effective predictions. It is also difficult to analyze the evolution of underwater terrain erosion. Summary of the invention

[0005] In view of the defects of the prior art, the present invention provides a method, system, terminal and medium for predicting water depth, so as to solve the problem that the prior art cannot predict water depth and then perform evolution analysis of underwater terrain scouring conditions, and realize the prediction of water depth and then perform evolution analysis of underwater terrain scouring conditions.

[0006] In order to achieve the above object, the present invention provides a method for predicting water depth, comprising:

[0007] S1. Construct a relationship model between predicted water depth, geographical location and time, and define the loss function;

[0008] S2. Construct a training matrix of the training data set based on geographic location and time, and use the transformed data set obtained after convolution, excitation and pooling of the training matrix as the input layer of the neural network;

[0009] S3. Construct a neural network, and obtain an output layer based on the input layer of the neural network, where the output layer is a predicted water depth data set;

[0010] S4. Use the predicted water depth data set and the corresponding actual water depth data set to train the neural network, use the genetic algorithm to iteratively optimize the optimal parameter matrix with the goal of minimizing the loss function, and predict the water depth based on the neural network with the optimal parameter matrix.

[0011] According to a method for predicting water depth provided by the present invention, step S1 specifically includes:

[0012] S11. The geographical location includes latitude and longitude. The latitude and longitude are converted into a plane coordinate system, with time as the sequence and depth as the label data, to construct the predicted water depth. Relationship model with latitude Y, longitude X, and time T;

[0013] S12. The root mean square error between the predicted water depth and the corresponding actual water depth is selected as the loss function to evaluate the effect of deep learning.

[0014] According to a method for predicting water depth provided by the present invention, step S2 specifically includes:

[0015] S21. Construct the training matrix based on latitude Y, longitude X, and time T:

[0016]

[0017] S22. Given a convolution kernel C, extract a data subset from the training matrix according to its own resolution and with a fixed step size, multiply it accordingly, and then calculate the sum until the retrieval of the training matrix is ​​completed;

[0018] S23. Process the convolution training matrix using the nonlinear activation function tanh;

[0019] S24. Use the maximum pooling method to pool the training matrix after excitation, and the obtained transformed data set is used as the input layer of the neural network.

[0020] According to a method for predicting water depth provided by the present invention, the neural network in step S3 specifically includes an input layer, a first hidden layer, a second hidden layer and an output layer; constructing the neural network specifically includes the following steps:

[0021] S31. Construct the parameter matrix ω from the input layer to the first hidden layer, through p i ω=Q implements input layer neuron p i The connection with the first hidden layer neuron Q;

[0022] S32. construct a parameter matrix θ from the first hidden layer to the second hidden layer, and realize the connection between the first hidden layer neuron Q and the second hidden layer neuron Z through Q·θ=Z;

[0023] S33. Construct the parameter matrix μ from the second hidden layer to the output layer, through Z·μ=h i Realize the second hidden layer neurons Z and the output layer neurons h i The connection between them, the output layer outputs the predicted water depth dataset.

[0024] According to a method for predicting water depth provided by the present invention, step S4 specifically includes:

[0025] S41. Input the transformed data set obtained based on the training data set into the neural network to obtain the predicted water depth data set, and combine it with the corresponding actual water depth data set to obtain the loss function cost(ω,θ,μ) about ω, θ and μ;

[0026] S42. Using a genetic algorithm, the loss function is continuously iterated by the optimizer, while updating the parameters ω, θ and μ until the loss function is minimized and the optimal parameter matrix is ​​obtained;

[0027] S43. Predicting water depth based on the neural network that obtains the optimal parameter matrix.

[0028] According to a method for predicting water depth provided by the present invention, step S42 specifically includes:

[0029] S421. Encode the weights and thresholds of the neural network;

[0030] S422. Generate an initial population using a random number generation method;

[0031] S423. Calculate individual fitness;

[0032] S424. Perform selection, crossover, and mutation operations;

[0033] S425. If the loss function is minimized or the maximum number of training times is reached, the iteration is terminated, the optimal weights and thresholds of the neural network are output, and the optimal parameter matrix is ​​obtained, otherwise the iteration continues.

[0034] Another aspect of the present invention provides a system for predicting water depth, comprising:

[0035] Preparation module: used to build a relationship model between predicted water depth, geographical location and time, and define the loss function;

[0036] Conversion module: used to construct a training matrix of the training data set based on geographic location and time. The converted data set obtained after convolution, excitation and pooling of the training matrix is ​​used as the input layer of the neural network.

[0037] Training module: used to construct a neural network, obtain the output layer of the neural network based on the input layer of the neural network, and the output layer is the predicted water depth data set;

[0038] Prediction module: used to train the neural network using the predicted water depth dataset and the corresponding actual water depth dataset, iteratively optimize to obtain the optimal parameter matrix with the goal of minimizing the loss function, and predict the water depth based on the neural network with the optimal parameter matrix.

[0039] According to a third aspect of the present invention, a terminal is provided, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor can implement the steps of the above-mentioned method for predicting water depth when executing the computer program.

[0040] A fourth aspect of the present invention provides a non-transitory computer-readable storage medium having a computer program stored thereon, which can implement the steps of the above-mentioned method for predicting water depth when the computer program is executed by a processor.

[0041] The present invention provides a method, system, terminal and medium for predicting water depth. A deep learning relationship model between water depth value, geographical location and time in underwater terrain scouring scenario is established, and a corresponding loss function is given. A training matrix is ​​constructed with geographical location and time. After convolution, excitation and pooling, the matrix is ​​input into a neural network. A genetic algorithm is used for continuous iteration to minimize the loss function, thereby obtaining an optimal parameter matrix, realizing the training and final construction of the neural network. The neural network constructed in this way predicts the water depth value of a given geographical location at a certain moment in the future, thereby realizing the evolution analysis of the underwater terrain scouring situation. BRIEF DESCRIPTION OF THE DRAWINGS

[0042] In order to more clearly illustrate the technical solutions in the present invention or the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0043] Figure 1 is a flow chart of a method for predicting water depth provided by an embodiment of the present invention;

[0044] Figure 2 It is a structural diagram of performing convolution, excitation and pooling on a training matrix provided by an embodiment of the present invention;

[0045] Figure 3 is a schematic diagram of the structure of a neural network provided by an embodiment of the present invention;

[0046] Figure 4is a schematic diagram of a process for optimizing neural network parameters using a genetic algorithm provided by an embodiment of the present invention;

[0047] Figure 5 It is a schematic diagram of the structure of a system for predicting water depth provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0048] In order to make the purpose, technical solution and advantages of the present invention clearer, the technical solution of the present invention will be clearly and completely described below in conjunction with the drawings of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0049] Figure 1 FIG1 is a flow chart of a method for predicting water depth provided by an embodiment of the present invention. As shown in FIG1 , the method for predicting water depth specifically includes the following steps:

[0050] S1. Construct a relationship model between predicted water depth, geographical location and time, and define the loss function;

[0051] S2. constructing a training matrix of the training data set based on the geographic location and time, and performing convolution, excitation and pooling on the training matrix to obtain a transformed data set as the input layer of the neural network;

[0052] S3. constructing the neural network, and obtaining the output layer of the neural network based on the input layer of the neural network, wherein the output layer is the predicted water depth data set;

[0053] S4. Use the predicted water depth data set and the corresponding actual water depth data set to train the neural network, iteratively optimize to obtain the optimal parameter matrix with the goal of minimizing the loss function, and predict the water depth based on the neural network that obtains the optimal parameter matrix.

[0054] The above method for predicting water depth first uses the training matrix composed of known quantities such as geographic location and time to convolve, excite and pool the converted data set obtained as the input layer of the neural network. The converted data set is processed by the neural network in a series of ways to obtain the predicted water depth data set of the output layer. The predicted water depth data set and the corresponding actual water depth data set are then used to train the neural network, and the optimal parameter matrix is ​​obtained by iterative optimization with the goal of minimizing the loss function. At this point, the neural network with the optimal parameter matrix can be used to predict water depth.

[0055] In a specific embodiment of the present invention, the above step S1 specifically includes the following steps:

[0056] S11. Underwater terrain data has four basic attributes: latitude (Y), longitude (X), depth (H), and time (T). By converting latitude and longitude into a plane coordinate system, with time as the sequence and depth as the label data, the prediction data can be constructed. Relationship model with training data Y, X, T:

[0057]

[0058] S12. In order to evaluate the effect of deep learning, that is, to measure The difference with H, the root mean square error is selected as the loss function, and the loss function is defined as follows:

[0059]

[0060] Where n represents the number of data sets. represents the predicted water depth data, and H represents the corresponding actual water depth data. In the iterative process of deep learning, the ultimate goal of training the machine is to minimize the value of cos t.

[0061] Figure 2 is a schematic diagram of a structure for performing convolution, excitation and pooling on a training matrix provided by an embodiment of the present invention, such as Figure 2 As shown, in a specific embodiment of the present invention, the above step S2 specifically includes the following steps:

[0062] S21. Construct a training matrix based on latitude Y, longitude X, and time T:

[0063]

[0064] S22. Given a convolution kernel C, its dimension is the same as the training matrix being convolved, but with a smaller horizontal resolution.

[0065]

[0066] This convolution kernel extracts a subset of data from the convolved data at a fixed step size according to its own resolution, multiplies it accordingly, and then calculates the sum until all data are retrieved. The training matrix after convolution becomes:

[0067]

[0068] S23. In order to consider the nonlinearity of the data, the activation function tanh is used to process the training matrix (K, S) after convolution to obtain:

[0069]

[0070] S24. Use the maximum pooling method to pool the training matrix after excitation, and the obtained data is used as the input layer of the neural network:

[0071]

[0072] Among them, m=n-2.

[0073] Figure 3 is a schematic diagram of the structure of a neural network provided by an embodiment of the present invention, such as Figure 3 As shown, the neural network in the above step S3 includes an input layer, two hidden layers and an output layer, wherein the input layer includes 1 neuron node, the first hidden layer and the second hidden layer include 4 and 5 neuron nodes respectively, and the output layer includes 1 neuron node. The neural network construction specifically includes the following steps:

[0074] S31. Input layer to the first hidden layer: For a neuron p in the input layer i , the first hidden layer contains 4 neurons, which can be expressed as a 1*4 matrix Q=[q1,q2,q3,q4]. i The mapping between the matrix Q and the matrix Q constructs a parameter matrix ω = [ω1, ω2, ω3, ω4], and the connection between the input layer neurons and the first hidden layer neurons is realized through the following relationship:

[0075] p i ω=Q

[0076] S32. From the first hidden layer to the second hidden layer: The second hidden layer has 5 neurons, which can be expressed as a 1*5 matrix Z=[z1,z2,z3,z4,z5]. To achieve the mapping between Q and Z, a new parameter matrix is ​​constructed:

[0077]

[0078] The connection between the first hidden layer neurons and the second hidden layer neurons is realized through Q·θ=Z.

[0079] S33. From the second hidden layer to the output layer: The output layer has only one predicted value, so only a new parameter matrix needs to be constructed:

[0080]

[0081] By Z·μ=h i The connection between the second hidden layer neurons and the output layer neurons can be realized.

[0082] In a specific embodiment of the present invention, the above step S4 specifically includes the following steps:

[0083] S41. After completing the construction of the neural network according to steps S31 to S33, the transformed data set obtained after the transformation of the training data set obtained by measurement is input into the neural network to obtain the corresponding predicted water depth data set. Combined with the corresponding actual water depth data set, the loss function cost(ω,θ,μ) about ω, θ and μ can be obtained.

[0084] S42. Using the genetic algorithm, the optimizer continuously iterates to reduce the loss function until the loss function is minimized and the optimal parameter matrix is ​​obtained.

[0085] S43. For the water depth value at a given geographical location at a certain point in the future, its Y, X, and T are all known. After convolution, excitation, and pooling, they are substituted into the neural network structure to predict the water depth value, thereby realizing its evolution analysis.

[0086] Figure 4 FIG. 1 is a flow chart of optimizing neural network parameters using a genetic algorithm according to an embodiment of the present invention. Figure 4 As shown, step S42 specifically includes the following steps:

[0087] S421. According to step S3, the topology of the neural network is 4-5-1, so the number of weights is 25, the number of thresholds is 6, and the encoding length is 31. The weights and thresholds of the neural network are encoded as X 0 .

[0088] S422. Generate an initial population using a random number generation method.

[0089] S423. Use the following function to calculate individual fitness:

[0090]

[0091] Where n is the number of output nodes of the neural network, y i and i are the actual output value and predicted output value of the i-th node respectively.

[0092] S424. Perform selection, crossover, and mutation operations, as follows:

[0093] Selection: The roulette method is used to determine the selection operator, and the selection probability of each individual is:

[0094] f i = k / F i

[0095]

[0096] Among them, F i is the fitness value of individual i, and N is the number of individuals in the population.

[0097] Crossover: Use the crossover operator to generate new individuals. and The new individuals obtained after crossover recombination are:

[0098]

[0099]

[0100] Mutation: For the parent individual X=(x1,x2,…,x k ,…x l ), the value interval of each gene position is {a k ,b k}, then the kth gene position generates a new individual X=(x1,x2,…,x' k ,…x l ) is:

[0101]

[0102] f(g)=r2(1-g / G max ) 2

[0103] Where g is the number of iterations, G max is the maximum number of evolutions, r∈[0,1], and r2 is a random number.

[0104] S425. If the minimized loss function is obtained or the maximum number of training times is reached, the iteration is terminated, the optimal weights and thresholds of the neural network are output, and the optimal parameter matrix is ​​obtained, otherwise the iteration continues.

[0105] Figure 5 5 is a schematic diagram of the structure of a system for predicting water depth provided by an embodiment of the present invention. As shown in FIG5 , Embodiment 2 provides a system for predicting water depth, including:

[0106] Preparation module, used to build a relationship model between predicted water depth, geographical location and time, and define the loss function;

[0107] The conversion module is used to construct a training matrix of the training data set based on geographic location and time, and the converted data set obtained after convolution, excitation and pooling of the training matrix is ​​used as the input layer of the neural network;

[0108] The training module is used to construct a neural network and obtain the output layer of the neural network based on the input layer of the neural network. The output layer is a predicted water depth data set.

[0109] The prediction module is used to train a neural network using a predicted water depth data set and a corresponding actual water depth data set, iteratively optimize to obtain an optimal parameter matrix with the goal of minimizing the loss function, and predict the water depth based on the neural network with the optimal parameter matrix.

[0110] Embodiment 3 provides a terminal, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, the steps of the above-mentioned method for predicting water depth can be implemented.

[0111] Embodiment 4 provides a non-transitory computer-readable storage medium having a computer program stored thereon. When the computer program is executed by a processor, the steps of the above-mentioned method for predicting water depth can be implemented.

[0112] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions and improvements made within the spirit and principle of the present invention should be included in the protection scope of the present invention.

Claims

1. A method for predicting water depth, characterized in that: The method comprises the following steps: S1. Construct a relationship model between the predicted water depth, geographical location and time, and define a loss function. The geographical location includes latitude and longitude. The relationship model is: in, To predict the water depth, X is longitude, Y is latitude, T is time, and the loss function is the root mean square error between the predicted water depth and the corresponding actual water depth; S2. Construct a training matrix of the training data set based on the geographic location and time, and perform convolution, excitation and pooling on the training matrix to obtain a transformed data set as the input layer of the neural network. The training matrix is: S3. constructing the neural network, wherein the neural network specifically includes an input layer, a first hidden layer, a second hidden layer and an output layer; obtaining the output layer of the neural network based on the input layer of the neural network, wherein the output layer is a predicted water depth data set; S4. Use the predicted water depth data set and the corresponding actual water depth data set to train the neural network, iteratively optimize to obtain the optimal parameter matrix with the goal of minimizing the loss function, and predict the water depth at a given geographical location at a certain moment in the future based on the neural network that has obtained the optimal parameter matrix, and then perform an evolutionary analysis on the underwater terrain scouring conditions.

2. The method for predicting water depth according to claim 1, characterized in that: The converted data set obtained after convolution, excitation and pooling of the training matrix in step S2 is used as the input layer of the neural network, which specifically includes: S22. Given a convolution kernel C, extract a data subset from the training matrix according to its own resolution and with a fixed step size, multiply it accordingly, and then calculate the sum until the retrieval of the training matrix is ​​completed; S23. Processing the training matrix after convolution using a nonlinear activation function tanh; S24. Use the maximum pooling method to pool the training matrix after excitation, and the obtained transformed data set is used as the input layer of the neural network.

3. The method for predicting water depth according to claim 1, characterized in that: Constructing the neural network specifically includes: S31. Construct a parameter matrix ω from the input layer to the first hidden layer, through p i ω = Q implements the input layer neuron p i The connection between the first hidden layer neuron Q; S32. construct a parameter matrix θ from the first hidden layer to the second hidden layer, and realize the connection between the first hidden layer neuron Q and the second hidden layer neuron Z through Q·θ=Z; S33. Construct a parameter matrix μ from the second hidden layer to the output layer, by Z·μ=h i Realize the second hidden layer neurons Z and the output layer neurons h i The connection between them, the output layer outputs the predicted water depth dataset.

4. The method for predicting water depth according to claim 1, characterized in that: The step S4 specifically includes: S41. Input the transformed data set obtained based on the training data set into the neural network to obtain the predicted water depth data set, and combine it with the corresponding actual water depth data set to obtain the loss function cost(ω, θ, μ) about ω, θ and μ; S42. Using a genetic algorithm, the loss function is continuously iterated by the optimizer, while updating the parameters ω, θ and μ until the loss function is minimized and the optimal parameter matrix is ​​obtained; S43. Based on the neural network that obtains the optimal parameter matrix, the water depth of a given geographical location is predicted at a certain moment in the future, and then an evolution analysis of the underwater terrain scouring is performed.

5. The method for predicting water depth according to claim 4, characterized in that: The step S42 specifically includes: S421. Encode the weights and thresholds of the neural network; S422. Generate an initial population using a random number generation method; S423. Calculate individual fitness; S424. Perform selection, crossover, and mutation operations; S425. If the loss function is minimized or the maximum number of training times is reached, the iteration is terminated, the optimal weights and thresholds of the neural network are output, and the optimal parameter matrix is ​​obtained, otherwise the iteration continues.

6. A system for predicting water depth, characterized in that: include: Preparation module: used to construct a relationship model between the predicted water depth, geographical location and time, and define a loss function. The geographical location includes latitude and longitude. The relationship model is: in, To predict the water depth, X is longitude, Y is latitude, T is time, and the loss function is the root mean square error between the predicted water depth and the corresponding actual water depth; Conversion module: used to construct a training matrix of the training data set based on the geographic location and time, and the converted data set obtained after convolution, excitation and pooling of the training matrix is ​​used as the input layer of the neural network. The training matrix is: Training module: used to construct the neural network, which specifically includes an input layer, a first hidden layer, a second hidden layer and an output layer; based on the input layer of the neural network, the output layer of the neural network is obtained, and the output layer is a predicted water depth data set; Prediction module: used to train the neural network using the predicted water depth data set and the corresponding actual water depth data set, iteratively optimize to obtain the optimal parameter matrix with the minimization of the loss function as the goal, and predict the water depth of a given geographical location at a certain moment in the future based on the neural network that obtains the optimal parameter matrix, and then perform evolutionary analysis on the underwater terrain scouring conditions.

7. A terminal comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the computer program, the steps of the method for predicting water depth according to any one of claims 1 to 5 are implemented.

8. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method for predicting water depth according to any one of claims 1 to 5 are implemented.

Images