Unmanned ship motion state and energy consumption combined prediction method based on deep learning

Abstract

The invention discloses an unmanned ship motion state and energy consumption joint prediction method based on deep learning, and the method comprises the steps: constructing an MSECP model based on an LSTM deep neural network, inputting the motion state of the current time step of an unmanned ship and a control signal into the MSECP model, predicting the motion state and energy consumption of the next time step of the unmanned ship through the MSECP model, and carrying out the prediction of the motion state and energy consumption of the unmanned ship. And based on the prediction result and the future control signal as input, predicting the motion state and energy consumption of the unmanned ship in a plurality of future time steps in an iteration mode. According to the method, the motion state and the energy consumption are integrated in the same prediction framework for joint modeling and prediction, so that the reliability and the consistency of prediction results are greatly improved, and deviation or inconsistency possibly caused by independent prediction of the motion state and the energy consumption is avoided. According to the method, the MSECP model is constructed based on the LSTM deep neural network, and the prediction precision of the motion state and the energy consumption of the unmanned ship is remarkably improved.

Description

Technical Field

[0001] The present invention relates to the field of unmanned boat control technology, and more specifically, to a method for jointly predicting the motion state and energy consumption of an unmanned boat based on deep learning. Background Art

[0002] Existing unmanned boats are mainly based on traditional control methods of physical modeling and some model-based machine learning algorithms (such as deep learning, regression models, etc.), which use known physical parameters and environmental data to predict motion states and energy consumption. These methods rely on preset models and environmental conditions to make predictions and evaluations through calculations.

[0003] The existing unmanned boat motion status and energy consumption prediction have the following deficiencies:

[0004] (1) Traditional physical modeling and control methods require highly accurate physical parameters and real-time environmental data, which makes them poorly adaptable in complex dynamic water environments and difficult to effectively capture the complex motion characteristics of unmanned surface vehicles (USVs) in a changing environment. In addition, the calculation process is cumbersome and cannot meet the requirements of real-time and high efficiency.

[0005] (2) Existing energy consumption prediction methods have accuracy issues, making it difficult to accurately predict the remaining power and range of unmanned surface vessels, especially in dynamic waters. Energy consumption changes are difficult to accurately control, affecting the safety and stability of mission execution.

[0006] (3) Existing technologies usually model and predict motion state and energy consumption separately, failing to fully consider the intrinsic relationship between the two, which may lead to inconsistencies in the results of independent predictions, thereby affecting the accuracy and reliability of the overall system.

[0007] Therefore, it is necessary to propose a new solution to solve the above problems. Summary of the invention

[0008] The purpose of the present invention is to overcome the deficiencies of the above-mentioned prior art and provide a method for jointly predicting the motion state and energy consumption of an unmanned boat based on deep learning.

[0009] In order to achieve the above object, the present invention adopts the following technical solution:

[0010] A method for jointly predicting the motion state and energy consumption of an unmanned boat based on deep learning comprises: constructing an MSECP model based on an LSTM deep neural network, inputting the motion state and control signal of the unmanned boat in the current time step into the MSECP model, the MSECP model predicting the motion state and energy consumption of the unmanned boat in the next time step, and predicting the motion state and energy consumption of the unmanned boat in multiple future time steps in an iterative manner based on the prediction result and the future control signal as input.

[0011] Furthermore, the method of predicting the motion state and energy consumption of the unmanned boat by the MSECP model is:

[0012] Step S1: at time step t, the motion state of the unmanned boat is and control signals Input into the established MSECP model; where, is the speed of the unmanned boat, is the heading angle of the unmanned boat, is the heading angle of the unmanned boat, is the energy consumption of the unmanned boat, and They are the speed control signals of the left and right motors of the unmanned boat respectively;

[0013] Step S2, to minimize the predicted value of the model output and the true value at the next time step The mean square error between them is used to update the MSECP model weights. The calculation formula is as follows:

[0014] (1)

[0015] In formula (1), N is the number of samples;

[0016] Step S3, repeating steps S1 to S2 until the MSECP model converges;

[0017] Step S4: at time step t, the motion state of the unmanned boat is and control signals Input into the trained MSECP model and obtain the predicted value at time step t+1 ;

[0018] Step S5: The predicted value and control signals Input into the trained MSECP model and obtain the predicted value at time step t+2 ;

[0019] Step S6, repeat step S5 until all predicted values ​​of N time steps are obtained.

[0020] Furthermore, the MSECP model uses stacking of three LSTM layers and one fully connected layer for time series learning, and the number of neurons in the three LSTM layers is set to 128, 64 and 32, respectively, to capture the long-term dependency between the motion state and energy consumption of the unmanned boat, and the final state and energy consumption are predicted through the fully connected layer.

[0021] Furthermore, the structure of the LSTM layer is:

[0022] (2)

[0023] In formula (2), and is the hidden state of the LSTM, is the input gate of LSTM, is the forget gate of LSTM, is a candidate memory unit of LSTM, is the output gate of LSTM; is the input vector at time step t, σ is the activation function, is the bias vector of the linear transformation, is the weight matrix of the linear transformation.

[0024] Furthermore, the unmanned boat adopts an electric under-driven type, and is equipped with a data acquisition module. The data acquisition module obtains the motion state and energy consumption information of the unmanned boat in real time, and transmits it to the processing unit of the onshore control center through wireless communication. The processing unit is responsible for constructing the MSECP model.

[0025] Furthermore, the data collected by the data acquisition module are preprocessed and then input into the MSECP model for training; during the training process, the weights are optimized using the back propagation algorithm and the loss function algorithm, and the prediction ability of the MSECP model is verified using cross validation.

[0026] The beneficial effects of the present invention are:

[0027] 1. The present invention integrates the motion state and energy consumption in the same prediction framework for joint modeling and prediction, which greatly improves the reliability and consistency of the prediction results, avoids the deviation or inconsistency that may be caused by independent prediction of the two, and improves the adaptability in complex water environments.

[0028] 2. The present invention constructs the MSECP model based on the LSTM deep neural network, which significantly improves the prediction accuracy of the motion state and energy consumption of the unmanned boat, especially in complex water environments, the prediction accuracy is better than the traditional method.

[0029] 3. The present invention optimizes the energy consumption planning of the unmanned boat and can perform real-time energy consumption control during mission execution, thereby reducing energy waste and improving operation safety.

[0030] 4. The present invention enhances the autonomous navigation and control capabilities of the unmanned boat in complex and changeable environments, and improves the stability and reliability of the system. BRIEF DESCRIPTION OF THE DRAWINGS

[0031] Figure 1 This is a flow chart of a method for jointly predicting the motion state and energy consumption of an unmanned boat based on deep learning in this embodiment;

[0032] Figure 2 This is an effect diagram of the 1-step speed prediction of the MSECP model in this embodiment;

[0033] Figure 3 This is an effect diagram of the 1-step heading angle prediction of the MSECP model in this embodiment;

[0034] Figure 4 This is an effect diagram of the heading angle prediction of the MSECP model in step 1 in this embodiment;

[0035] Figure 5 This is an effect diagram of the energy consumption prediction of the MSECP model in one step in this embodiment;

[0036] Figure 6 This is an effect diagram of the MSECP model predicting 10 consecutive steps of speed in this embodiment;

[0037] Figure 7 This is an effect diagram of the MSECP model predicting the heading angle for 10 consecutive steps in this embodiment;

[0038] Figure 8 This is an effect diagram of the MSECP model predicting the heading angle for 10 consecutive steps in this embodiment;

[0039] Fig. 9 This is an effect diagram of the MSECP model's 10-step continuous energy consumption prediction in this embodiment. DETAILED DESCRIPTION

[0040] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only 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.

[0041] Embodiment: A method for jointly predicting the motion state and energy consumption of an unmanned boat based on deep learning, such as Figure 1 As shown, it includes: constructing an MSECP model based on the LSTM deep neural network, inputting the motion state and control signal of the unmanned boat in the current time step into the MSECP model, and the MSECP model predicts and outputs the motion state and energy consumption of the unmanned boat in the next time step, and based on the prediction results and future control signals as input, predicting the motion state and energy consumption of the unmanned boat in multiple future time steps in an iterative manner. By integrating the motion state and energy consumption of the unmanned boat in the same prediction framework for joint modeling and prediction, the reliability and consistency of the prediction results are greatly improved; at the same time, by adopting the LSTM deep neural network, the prediction accuracy of the motion state and energy consumption of the unmanned boat is significantly improved.

[0042] Specifically, the method of MSECP model predicting the motion state and energy consumption of unmanned boat is:

[0043] Step S1: at time step t, the motion state of the unmanned boat is and control signals Input into the established MSECP model; where, is the speed of the unmanned boat, is the heading angle of the unmanned boat, is the heading angle of the unmanned boat, is the energy consumption of the unmanned boat, and They are the speed control signals of the left and right motors of the unmanned boat respectively;

[0044] Step S2, to minimize the predicted value of the model output and the true value at the next time step The mean square error between them is used to update the MSECP model weights. The calculation formula is as follows:

[0045] (1)

[0046] In formula (1), N is the number of samples;

[0047] Step S3, repeating steps S1 to S2 until the MSECP model converges;

[0048] Step S4: at time step t, the motion state of the unmanned boat is and control signals Input into the trained MSECP model and obtain the predicted value at time step t+1 ;

[0049] Step S5: The predicted value and control signals Input into the trained MSECP model and obtain the predicted value at time step t+2 ;

[0050] Step S6, repeat step S5 until all predicted values ​​of N time steps are obtained.

[0051] Among them, the MSECP model uses stacking 3 LSTM layers and 1 fully connected layer for time series learning, such as Figure 1 As shown in the figure, the number of neurons in the three LSTM layers is set to 128, 64 and 32 respectively, which is used to capture the long-term dependency between the motion state and energy consumption of the unmanned boat. The number of neurons in the fully connected layer is set to 16 for the final state and energy consumption prediction. The activation function of the LSTM layer is tanh.

[0052] Furthermore, the architecture of the LSTM layer is:

[0053] (2)

[0054] In formula (2), and is the hidden state of the LSTM, is the input gate of LSTM, is the forget gate of LSTM, is a candidate memory unit of LSTM, is the output gate of LSTM; is the input vector at time step t, σ is the activation function, is the bias vector of the linear transformation, is the weight matrix of the linear transformation.

[0055] Furthermore, the unmanned boat adopts the Wisdom-ⅠUSV platform, which includes a power system, a battery management system and a sensor system (such as GPS, an electronic compass, an accelerometer, etc.). The unmanned boat is an electric underdriven type, and it navigates through the differential motion of two brushless motors. The unmanned boat is equipped with a data acquisition module, which obtains the motion state and energy consumption information of the unmanned boat in real time, and transmits it to the processing unit of the shore control center through wireless communication. The processing unit is responsible for constructing the MSECP model to achieve high-precision prediction of the motion state and energy consumption of the unmanned boat.

[0056] Specifically, the unmanned boat is equipped with a GPS positioning system and an electronic compass for measuring the state of motion; the battery management system is used to monitor the battery voltage and output current, so as to record the energy consumption of the unmanned boat; the onshore control center controls the propeller of the unmanned boat by sending random signals, allowing the unmanned boat to travel freely and collect data. The collected data sets cover a variety of information such as speed, heading angle, heading angle, energy consumption, and motor control signals.

[0057] Before the collected data set is input into the MSECP model for training, it is necessary to preprocess the data, including normalization, standardization, processing missing values, removing outliers, feature selection and extraction, and other steps.

[0058] In this embodiment, the data set contains 21,870 data groups, which are divided into training sets and test sets in a ratio of 8:2, which are approximately 17,493 training samples and 4,374 test samples respectively. This division method effectively provides a reasonable balance between model training and evaluation.

[0059] During the training process, the back propagation algorithm and loss function algorithm are used to optimize the weights to ensure that the MSECP model can accurately predict the motion state and energy consumption of the unmanned boat; cross-validation and other performance evaluation methods are used to verify the prediction ability of the MSECP model to ensure the generalization ability and reliability of the MSECP model in respective water environments. Specifically:

[0060] The Adam optimizer is used to adaptively adjust the learning rate of each parameter. Adam combines the advantages of stochastic gradient descent (SGD) and gradient descent algorithms with momentum and RMSprop, which helps to converge efficiently and can handle sparse gradients. At the same time, the learning rate is set to 0.001 and the model is trained for 500 rounds to ensure that the data set is traversed multiple times, thereby effectively minimizing the loss function.

[0061] MSE is used as the loss function to calculate the average of the square of the difference between the predicted value and the true value. During the training process, the value of the loss function is gradually reduced by continuously adjusting the weights and biases of the model, thereby improving the prediction accuracy of the model.

[0062] Cross-validation is used to evaluate the generalization ability of a model by dividing the training set into multiple subsets, with each subset taking turns as a validation set and the rest as a training set. This process is repeated multiple times and the average is taken as an estimate of the model performance.

[0063] When evaluating performance, mean absolute error (MAE) and root mean square error (RMSE) are used to evaluate the performance of the model. These indicators are standard indicators for regression tasks, among which MAE is less sensitive to outliers, while RMSE is more sensitive to outliers.

[0064] In this embodiment, the MSECP model is verified by single-step prediction and multi-step prediction, specifically:

[0065] (1) Single-step prediction

[0066] First, the data obtained from the unmanned boat is trained; after the training process, the 1-step prediction ability of the MSECP model is tested, including speed, heading angle, heading angle and energy consumption, such as Figure 2-Figure 5 The prediction results of 1600 data points are shown in the curve. The prediction is relatively poor only during the switch from 360° to 0°. There is only a small error between the overall prediction value and the true value. These results show that the designed MSECP model can predict the motion state and energy consumption of the unmanned boat through the trained data. The results are shown in Table 1:

[0067] Table 1: MAE and RMSE of MSECP model

[0068]

[0069] By inputting the motion state and control signal of the current time step, the motion state and energy consumption of the next time step can be directly predicted. This prediction method is suitable for scenarios with high real-time requirements, and can provide instant feedback for the unmanned boat to help the control system adjust the navigation strategy.

[0070] (2) Multi-step prediction

[0071] A two-layer feedforward neural network (FNN) and residual neural network (ResNet) model are introduced and compared. Figure 6-Figure 9 As shown in the figure, the prediction results of 10 consecutive steps are given. Experiments show that with the increase of the number of steps, the prediction errors of FNN and ResNet also increase significantly. Compared with FNN and ResNet, the MSECP model has the lowest MAE and RMSE in terms of motion state and energy consumption, and performs best in terms of multi-step motion and energy consumption. High accuracy can be achieved through multi-step prediction. The comparison is shown in Table 2:

[0072] Table 2: MAE and RMSE of different models

[0073]

[0074] Based on the prediction results of the previous moment and the new control signal, the motion state and energy consumption at multiple moments in the future are predicted iteratively. Multi-step prediction can provide planning and decision support over a longer time range, help optimize navigation routes and mission execution plans, thereby reducing energy consumption and improving voyage efficiency.

[0075] The above experimental results confirm the feasibility of the MSECP algorithm in predicting the motion state and energy consumption of unmanned boats. The MSECP model performs well in both single-step prediction and multi-step prediction. Especially in multi-step prediction, it shows significant advantages in actual mission planning and control compared with FNN and ResNet. By using real ship data for experiments, the high-precision prediction ability of this method for motion state and energy consumption is fully demonstrated. At the same time, according to the predicted motion state and energy consumption data, the system can adjust the motion strategy and task execution plan in real time. Through the feedback of the prediction results, the navigation route and energy distribution of the unmanned boat are optimized, and an efficient energy consumption pipeline is realized to ensure the safe and stable completion of the mission.

[0076] The above is only a preferred embodiment of the present invention, and the protection scope of the present invention is not limited to the above embodiments. All technical solutions under the concept of the present invention belong to the protection scope of the present invention. It should be pointed out that for ordinary technicians in this technical field, some improvements and modifications without departing from the principle of the present invention should also be regarded as the protection scope of the present invention.

Claims

1. A method for jointly predicting the motion state and energy consumption of an unmanned boat based on deep learning, characterized in that: include: The MSECP model is constructed based on the LSTM deep neural network. The motion state and control signal of the unmanned boat in the current time step are input into the MSECP model. The MSECP model predicts the motion state and energy consumption of the unmanned boat in the next time step. Based on the prediction results and future control signals as input, the motion state and energy consumption of the unmanned boat in multiple future time steps are predicted in an iterative manner.

2. According to claim 1, a method for jointly predicting the motion state and energy consumption of an unmanned boat based on deep learning is characterized in that: The method of the MSECP model to predict the motion state and energy consumption of the unmanned boat is: Step S1: at time step t, the motion state of the unmanned boat is and control signals Input into the established MSECP model; where, is the speed of the unmanned boat, is the heading angle of the unmanned boat, is the heading angle of the unmanned boat, is the energy consumption of the unmanned boat, and They are the speed control signals of the left and right motors of the unmanned boat respectively; Step S2, to minimize the predicted value of the model output and the true value at the next time step The mean square error between them is used to update the MSECP model weights. The calculation formula is as follows: (1) In formula (1), N is the number of samples; Step S3, repeating steps S1 to S2 until the MSECP model converges; Step S4: at time step t, the motion state of the unmanned boat is and control signals Input into the trained MSECP model and obtain the predicted value at time step t+1 ; Step S5: The predicted value and control signals Input into the trained MSECP model and obtain the predicted value at time step t+2 ; Step S6, repeat step S5 until all predicted values ​​of N time steps are obtained.

3. According to claim 1, a method for jointly predicting the motion state and energy consumption of an unmanned boat based on deep learning, characterized in that: The MSECP model uses stacking of three LSTM layers and one fully connected layer for time series learning. The number of neurons in the three LSTM layers is set to 128, 64 and 32, respectively, to capture the long-term dependency between the motion state and energy consumption of the unmanned boat, and the final state and energy consumption are predicted through the fully connected layer.

4. According to claim 3, a method for jointly predicting the motion state and energy consumption of an unmanned boat based on deep learning is characterized in that: The architecture of the LSTM layer is: (2) In formula (2), and is the hidden state of the LSTM, is the input gate of LSTM, is the forget gate of LSTM, is a candidate memory unit of LSTM, is the output gate of LSTM; is the input vector at time step t, σ is the activation function, is the bias vector of the linear transformation, is the weight matrix of the linear transformation.

5. According to the method for jointly predicting the motion state and energy consumption of an unmanned boat based on deep learning in claim 1, it is characterized in that: The unmanned boat is of electric under-actuated type and is equipped with a data acquisition module. The data acquisition module acquires the motion state and energy consumption information of the unmanned boat in real time and transmits it to the processing unit of the onshore control center through wireless communication. The processing unit is responsible for constructing the MSECP model.

6. The method for jointly predicting the motion state and energy consumption of an unmanned boat based on deep learning according to claim 5 is characterized in that: The data collected by the data acquisition module are preprocessed and then input into the MSECP model for training. During the training process, the weights are optimized using the back propagation algorithm and the loss function algorithm, and the prediction ability of the MSECP model is verified using cross validation.

Images