Pure electric tug energy consumption prediction method based on genetic algorithm and deep learning

Abstract

The invention provides a pure electric tug energy consumption prediction method based on a genetic algorithm and deep learning, and the method comprises the steps: collecting the operation data of a pure electric tug in real time, enabling the operation data to comprise electric cabin data, navigation data and environment parameters, and carrying out the preprocessing of the operation data; constructing a training set containing a working condition label, including extracting a tug unit distance energy consumption rate according to the instantaneous energy consumption and the navigational speed of the electric warehouse, and judging a working condition type by mapping the tug unit distance energy consumption rate and combining the tug position information; constructing a pure electric tug energy consumption prediction model based on bidirectional LSTM, dynamically optimizing model hyper-parameters by using a genetic algorithm, and performing model training; and updating data in real time through a sliding time window, and synchronously outputting predicted values of energy consumption per minute and energy consumption per sea based on the trained pure electric tug energy consumption prediction model. The influence of different working conditions of the pure electric tug on the energy consumption of the tug is fully considered, and real-time prediction can be supported.

Description

TECHNICAL FIELD

[0001] The present application belongs to the technical field of intelligent prediction of ship energy consumption, and particularly relates to a pure electric tugboat energy consumption prediction technology based on a genetic algorithm and deep learning. BACKGROUND

[0002] In a port operation system, a tugboat, as a key operation equipment, has long been responsible for core tasks such as assisting large ships to berth and unberth and escorting maritime traffic. The power system of a tugboat is gradually transitioning from a diesel engine to a lithium battery. Pure electric tugboats, with their zero-pollution emission and high energy conversion efficiency, are gradually becoming important equipment in the fields of port towing and offshore operations. However, due to the limitation of the energy density of current energy storage units, such ships have the significant defect of insufficient endurance. Therefore, real-time prediction of energy consumption with high accuracy and strong robustness is the premise and basis for improving the energy efficiency of pure electric tugboats. Real-time and accurate prediction of tugboat energy consumption under different working conditions can provide a reference for the intelligent operation and management of tugboats.

[0003] The current existing technologies for ship energy consumption prediction mainly include white box models, black box models and gray box models. For example, Yang et al. combined a genetic algorithm with the physical principles of a propulsion system to construct a high-precision gray box model through global parameter optimization. N. Srivastava and Hinton et al. improved the neural network hidden layer connection mechanism to enhance the model generalization ability. S. Ioffe et al. designed an inter-layer standardization strategy to significantly improve the stability of the training process. These achievements provide effective tools for traditional ship energy efficiency analysis. However, the energy consumption of a pure electric tugboat is related not only to the sailing environment and sailing data but also to the working condition of the tugboat. The above-mentioned research methods cannot effectively predict the energy consumption of a pure electric tugboat under different working conditions. The method proposed in the present application can real-time predict the energy consumption per minute and per nautical mile of a pure electric tugboat under different working conditions.

[0004] Through research on related patent documents, no existing research has been found that can reach the technical level of the present application: CN118536408A provides a tugboat power prediction method, but its defects are: the power prediction relies on a physical model (wind wave flow resistance formula), and the data-driven deep learning technology is not fully combined, making it difficult to deal with complex nonlinear relationships; the LSTM model is not optimized for hyperparameters, and is only used for time series prediction, without being designed for energy consumption multi-target (energy consumption per minute and per nautical mile) output; the working condition recognition relies on cluster analysis, which is not real-time enough and cannot adapt to high-frequency data scenarios.

[0005] CN118839590A provides a tug operation energy consumption prediction method, but its defects are: feature engineering relies on manual design (such as interaction features, bin processing), and the automatic feature extraction capability of deep learning is not fully utilized; the model training does not consider hyperparameter optimization, and the fitting capability of traditional machine learning methods such as LightGBM for complex time series data is limited; the sensor data noise problem (such as abnormal values of electric warehouse current / voltage) is not solved, the data preprocessing is simple, and the model stability is affected.

[0006] It can be seen that the traditional energy consumption prediction method does not consider the energy consumption difference of the tug under different working conditions (such as towing, pushing, and parking), and relies on static physical models or manual feature design, which cannot adapt to real-time environmental changes (such as dynamic interference of wind and wave flow) and sensor data noise, resulting in large deviation between the prediction result and the actual demand. SUMMARY

[0007] The technical problem to be solved by the present application is to overcome the shortcomings of the prior art, and a pure electric tug energy consumption prediction scheme based on genetic algorithm and bidirectional long short-term memory network is proposed.

[0008] The present application provides a pure electric tug energy consumption prediction method based on genetic algorithm and deep learning, comprising the following processes: Real-time acquisition of the running data of the pure electric tug, the running data including electric warehouse data, navigation data and environmental parameters, and preprocessing of the running data; Constructing a training set containing working condition labels, including extracting the energy consumption rate per unit distance of the tug according to the instantaneous energy consumption of the electric warehouse and the sailing speed, and determining the working condition type by mapping the energy consumption rate per unit distance of the tug and combining the position information of the tug; Constructing a pure electric tug energy consumption prediction model based on bidirectional LSTM, dynamically optimizing the model hyperparameters using genetic algorithm, and training the model; Real-time updating of data through a sliding time window, based on the trained pure electric tug energy consumption prediction model, and simultaneously outputting the prediction values of energy consumption per minute and energy consumption per nautical mile.

[0009] Moreover, the electric warehouse data includes electric warehouse remaining capacity, electric warehouse current, and voltage, the navigation data includes propeller speed, power, ship position, ground speed, and heading, and the environmental parameters include wind speed, wind direction, flow rate, flow direction, and significant wave height.

[0010] Moreover, when preprocessing, the data segments with electric warehouse remaining capacity SOC exceeding the range of 0-100% are deleted, and the missing values of the heading and wind direction data are filled with sliding window mean values.

[0011] Moreover, the instantaneous energy consumption of the power bin is calculated, and it is judged whether it is greater than 0. If yes, it is judged that the tugboat is in the charging working condition, and if not, the energy consumption rate per unit distance is calculated. The energy consumption rate per unit distance of the tugboat is mapped by using a sigmoid function, and the working condition type is determined in combination with the port location information.

[0012] Moreover, the working condition type includes towing, pushing, sailing, port stopping and charging working conditions, wherein the corresponding label codes of the towing and pushing working conditions are 0, the corresponding label code of the sailing working condition is 1, the corresponding label code of the port stopping working condition is 2, and the corresponding label code of the charging working condition is 3.

[0013] Moreover, the pure electric tugboat energy consumption prediction model based on the bidirectional LSTM includes a bidirectional LSTM layer, a full connection layer, a reverse normalization layer and a multi-target prediction layer arranged in sequence. The multi-target prediction layer outputs the energy consumption prediction per nautical mile and the energy consumption prediction per minute through a white box model.

[0014] Moreover, the implementation of dynamically optimizing the model hyperparameters by using the genetic algorithm is to initialize a population and randomly generate a hyperparameter combination; train the model by taking the minimization of the validation set loss as the fitness function; and finally output the optimal hyperparameter combination by iterative optimization through selection, crossover and mutation operations.

[0015] On the other hand, the present application also provides an electronic device comprising a memory, a processor and a computer program stored on the memory and executable on the processor, wherein the processor implements the genetic algorithm and deep learning based pure electric tugboat energy consumption prediction method as described above when executing the program.

[0016] On the other hand, the present application also provides a non-transitory computer readable storage medium having a computer program stored thereon, wherein the computer program is executed by a processor to implement the genetic algorithm and deep learning based pure electric tugboat energy consumption prediction method as described above.

[0017] On the other hand, the present application also provides a computer program product comprising a computer program, wherein the computer program is executed by a processor to implement the genetic algorithm and deep learning based pure electric tugboat energy consumption prediction method as described above.

[0018] The pure electric tugboat energy consumption prediction scheme provided by the present application fully considers the influence of different working conditions of the tugboat on the energy consumption of the tugboat, and constructs a deep learning network capable of predicting the real-time energy consumption rate of the pure electric tugboat under different working conditions. Compared with the prior art, the difference and advantage of the present application are that: 1) Multi-modal data fusion: integrate working condition labels, environmental parameters and sensor data to construct multi-dimensional input features covering the whole scene of tugboat operation.

[0019] 2) A special deep learning model is built for pure electric tug energy consumption prediction, combining bidirectional LSTM and genetic algorithm dynamic optimization, taking into account time series dependence and hyperparameter optimization, the model convergence speed is improved by 30%, and the prediction accuracy is significantly higher than that of the comparative document.

[0020] 3) A multi-objective prediction layer is proposed to fuse with the physical model, integrating the white box physical formula at the output end of the deep learning, calculating the energy consumption per minute and the energy consumption per mile through the real-time speed, realizing the fine expression of energy consumption prediction, and meeting the differentiated needs of port scheduling and energy management. And the inverse normalization layer is adapted, the model output is restored to the actual physical quantity through inverse standardization, directly connected to the tug control system, avoiding secondary conversion error. 4) Real-time and applicability: through the sliding time window and the real-time judgment of working condition, it supports the energy consumption prediction of every minute, which meets the high-frequency scheduling needs of port operation. The training set data is cleaned, and the model stability is improved.

[0021] 5) The model proposed in the present application can support lightweight deployment (such as edge computing devices) to realize real-time operation of shipboard terminals, which is significantly better than existing solutions that rely on cloud computing. The implementation is simple and convenient, and has strong practicality, which solves the problems of low practicality and inconvenience in actual application in related technologies, has significant technical advancement, and has important market value. BRIEF DESCRIPTION OF DRAWINGS

[0022] Figure 1 is the pure electric tug energy consumption prediction flowchart based on genetic algorithm and deep learning of the embodiment of the present application; Figure 2 is the pure electric tug working condition identification flowchart of the embodiment of the present application; Figure 3 is the model network structure diagram of the embodiment of the present application; Figure 4 is the multi-objective prediction unit structure diagram of the embodiment of the present application; Figure 5 is the verification set loss result diagram of model training of the embodiment of the present application; Figure 6 is the comparison diagram of model prediction value and true value of the embodiment of the present application; Figure 7 is the performance comparison diagram of the scheme and the prior art scheme of the embodiment of the present application. DETAILED DESCRIPTION

[0023] The concept, specific structure and technical effects of the present application will be further described below in combination with the drawings and embodiments, so as to fully understand the purpose, features and effects of the present application.

[0024] The pure electric tug energy consumption prediction method provided by the application mainly includes the design of the following parts: (1) Bidirectional time sequence capture: preferably based on a bidirectional long short-term memory network to construct a pure electric tug energy consumption prediction model to simultaneously process forward and reverse time sequence information, effectively capture long-term dependencies in the operation of the tug (such as the decay trend of the battery power, the fluctuation period of the speed), and avoid the problem of local feature loss caused by traditional LSTM unidirectional modeling; (2) Introducing genetic algorithm to optimize network hyperparameters, taking the minimum validation set loss as the target, dynamically adjusting the model structure, which can avoid the local optimal trap of manual parameter tuning, and make the model convergence speed increase by 30%; support dynamic adjustment of Dropout rate and learning rate decay strategy (such as exponential decay), enhance the generalization ability of the model to different data distributions (such as port operation and cross-regional scheduling navigation), and strong adaptability; (3) Designing a multi-objective prediction layer, combining the LSTM output with the speed to simultaneously predict the energy consumption per minute (ΔSOC) and the energy consumption per mile (EPM); (4) Mapping the unit distance energy consumption rate, combining the speed threshold to judge the working condition of the tug in real time.

[0025] Embodiment 1 As Figure 1 shown, the application provides a pure electric tug energy consumption prediction method based on genetic algorithm and deep learning, which contains a model based on genetic algorithm and long short-term memory neural network, and the overall training steps of the model are as Figure 3 shown, the training and prediction process of the model of the present embodiment will be further explained, the present embodiment includes the following steps: S1, real-time acquisition of various data of the pure electric tug, including battery data, navigation data and environmental parameters, and data preprocessing of the collected data, including outlier deletion, null value filling and standardization.

[0026] In the embodiment of the application, the preferred implementation of S1 contains the following sub-steps: S11, the specific data collected by the tug data acquisition module includes battery remaining capacity, battery current, voltage, propeller speed, power, ship position, ground speed, heading, wind speed, wind direction, flow rate, flow direction, significant wave height.

[0027] S12, data null value processing: the data segment where the value deviates from the reasonable range or is empty should be deleted. Specifically, the data whose battery capacity SOC is greater than 100% and less than 0% should be deleted, the data whose heading and wind direction are not in the range of 0-360° should be deleted, and other obvious outliers should be deleted.

[0028] For example, for the SOC data of the electric warehouse, if it is detected that the instantaneous value exceeds 100%, the data segment of 10 seconds before and after this moment is deleted; for the heading data, the sliding window mean is used to fill in the missing values.

[0029] S13, calculate the SOC of the electric warehouse at each moment and the difference value of the last moment .

[0030]

[0031] wherein represents the moment, is the battery power at the moment, is the battery power at the moment, in the present application, the difference value of the electric warehouse power represents the instantaneous change value of the remaining power SOC of the electric warehouse, which is the target value of the model prediction.

[0032] S14, data standardization of the features and the target is performed to eliminate the dimensional difference, and the standardized time series data can be obtained. The standardization method is as follows:

[0033] wherein is the mean of all sample data, is the standard deviation of all sample data, is the standardized value.

[0034] S2, calculate the energy consumption per unit distance of the tugboat , and judge the current working condition of the tugboat based on the energy consumption per unit distance, and build a training set.

[0035] In the embodiment of the present application, the flow chart of the pure electric tugboat working condition recognition of S2 is as shown in Figure 2 , mainly including the following sub-steps: S21, calculate the instantaneous average current of the electric warehouse using the original data and the instantaneous average voltage , calculate the instantaneous energy consumption of the electric warehouse ; judge whether is greater than 0, if yes, it is judged that the tugboat is in the charging working condition, and the judgment process is ended, if not, calculate the energy consumption per unit distance , if the speed at this moment is 0, then , enter step S22. Wherein represents the absolute value of the instantaneous energy consumption, and INF represents the infinite value.

[0036] S22, in order to handle the infinite unit distance energy consumption, the application proposes to introduce a sigmoid activation function for mapping, and according to the sigmoid value obtained by mapping, if the sigmoid value is close to 0.5, it is judged that the tugboat is in the navigation working condition, if not, continue to judge whether close to 1, if yes, it is judged that the tugboat is in the towing or pushing working condition, if not, step S23 is entered.

[0037] The function can map any positive real number to the range of [0.5, 1], so when EPD is infinite, the sigmoid function returns a result of 1, and when EPD is 0, the function returns a result of 0.5. The sigmoid function formula used in the application is as follows.

[0038]

[0039] Where e is the natural base.

[0040] In the embodiment, the , when close to 1 (i.e. greater than 0.75 and not equal to 1), it is judged that the tugboat is in the towing or pushing working condition at this time, and when close to 0.5 (i.e. less than or equal to 0.75), it is judged that the tugboat is in the navigation state. S23, judge whether the tugboat positioning is near the port location (such as the tugboat base), if yes, judge that the tugboat is in the port, if not, judge that the tugboat is in the towing or pushing working condition.

[0041] According to and the tugboat position information, the working condition of the tugboat is comprehensively judged. Except is not 1, the working condition can be directly judged according to , when is 1, if the tugboat position is near the port location at this time, it is judged that the tugboat is in the port.

[0042] S24, label encoding is performed on the tugboat working condition feature column, and the two working conditions of towing and pushing are encoded as 0, the navigation working condition is encoded as 1, the port working condition is encoded as 2, and the charging working condition is encoded as 3; a data set containing the labeled working condition features is constructed, and the data set is divided according to the training set 80%, the verification set 10% and the test set 10%, and the processed part of the feature data set is shown in Table 1.

[0043] Table 1

[0044] S3, a deep learning network for pure electric tugboat energy consumption prediction is constructed, and the network hyperparameters are optimized by using a genetic algorithm, and a trained pure electric tugboat energy consumption prediction model is obtained.

[0045] In the embodiment of the present application, S3 comprises the following sub-steps: S31, constructing a pure electric tug energy consumption prediction model based on bidirectional LSTM: the architecture of the model is as shown in Figure 3 The model input is the normalized time series data (electric warehouse current, voltage, speed, etc.), which is extracted by the bidirectional LSTM layer, mapped to the inverse normalization layer through the fully connected layer, and finally output by the multi-objective layer every minute and every nautical mile energy consumption. In the model, the first and second layers constitute the bidirectional LSTM layer, which is used to process sequence data and capture long-term and short-term dependencies in the sequence, i.e. a single layer of BiLSTM consists of two layers of LSTM, corresponding to the first and second layers. The first layer is a forward processing input sequence; the second layer is a reverse processing sequence. After processing, the outputs of the two LSTM are spliced to obtain the final output. Through this method, long-term and short-term dependencies in the tug operating condition data can be effectively captured. The third layer is a fully connected layer for linear transformation of the LSTM output. The fourth layer is an inverse normalization layer for outputting real prediction values. The fifth layer is a multi-objective prediction layer that outputs per nautical mile energy consumption prediction and per minute energy consumption prediction through a white box model, and finally maps to the output layer. In each training iteration process, the bidirectional LSTM model captures the time series relationship of the features in the training set, and processes the forward and reverse information of the sequence, which helps to increase the accuracy of the model prediction.

[0046] Further, the inverse normalization can output the normalized value output by the model through the inverse normalization function to the real prediction value. Its processing method is to inverse normalize the per-minute energy consumption prediction value output by the previous layer to obtain the real prediction value The inverse normalization formula is as follows:

[0047] Wherein is the mean of all sample data during training, is the standard deviation of all sample data during training, is the per-minute energy consumption prediction value output by the model.

[0048] Further, the multi-objective prediction layer can calculate the per nautical mile energy consumption prediction based on the per minute energy consumption prediction through the white box model. The structure of the multi-objective prediction layer is as shown in Figure 4 The multi-objective prediction layer combines deep learning model and white box physical formula to realize multi-dimensional prediction of pure electric tug energy consumption, which more comprehensively reflects the energy consumption characteristics of the tug under different operating conditions, and provides fine data support for optimizing energy distribution and formulating scheduling strategies. The specific steps are as follows: S311, the input of the multi-objective prediction layer is the model prediction value and the current speed, the distance D traveled per minute is calculated by dividing the current speed by 60.

[0049] S312, the EPM value is calculated by the E(x) unit, and the calculation formula is:

[0050] S313, the EPM value is calculated by the E(x) unit, and the calculation formula is: and the EPM value.

[0051] S32, genetic algorithm optimization of hyperparameters: the embodiment preferably uses the minimum validation set loss as the optimization target, sets the population size to 10, the crossover probability to 0.7, and the mutation probability to 0.2, and optimizes the model hyperparameters, including the hidden layer size, the LSTM layer number, the learning rate, and the Dropout rate.

[0052] The implementation steps of the genetic algorithm optimization of hyperparameters include: initializing the population and randomly generating hyperparameter combinations; training the model with the minimum validation set loss as the fitness function; and iteratively optimizing through selection, crossover, and mutation operations, and finally outputting the optimal hyperparameter combination.

[0053] Specifically, the genetic algorithm optimization of hyperparameters implemented in the embodiment includes: S321, population initialization: randomly generate 10 sets of hyperparameter combinations.

[0054] S322, fitness evaluation: train the model for 100 rounds for each set of parameters, and use the early stopping mechanism to monitor the validation loss.

[0055] S323, selection, crossover, and mutation are performed on the population for iterative optimization, a total of 10 generations, and the optimal hyperparameter combination is output.

[0056] S33, training of deep learning model: use the hyperparameters optimized by the genetic algorithm to train the model, with a training iteration number of 100, and add an early stopping mechanism to monitor the validation set loss. When the validation set loss does not decrease for 10 consecutive rounds, the training is stopped, the model is output, and the model prediction effect is tested using the test set. The validation set loss of the model training is shown in Figure 5 The results show that the model validation set loss is 0.1044, and the predicted results of the test set are compared with the true values as shown in Figure 6As shown in the figure, it can be seen from the figure that the predicted value and the true value curve fitting degree is high, the model prediction effect is good, the prediction variance of the model is reduced by about 27.91% compared with the LSTM model, and the prediction error variance of the model and other existing network models such as random forest (RF), extreme random forest (ET), extreme gradient boosting (XGBoost) and long short-term memory network (LSTM) is compared, including average relative error (MRE), mean absolute error (MAE) and root mean square error (RMSE) The comparison chart is as shown in Figure 7

[0057] S4, using the pure electric tug energy consumption prediction model to predict the energy consumption of the tug in real time.

[0058] In the embodiment of the application, in order to use the trained pure electric tug energy consumption prediction model to predict the pure electric tug energy consumption in real time, the trained model is deployed in the ship equipment, and the pure electric tug energy consumption is predicted in real time. Specifically, S4 comprises the following sub-steps: S41, obtaining real-time navigation data and environmental data of the pure electric tug and pre-processing and normalizing the data, and then judging the working condition of the tug.

[0059] S42, constructing time window data for prediction according to the pure electric tug characteristic data, which includes the running state of the pure electric tug, the navigation environment data and its working condition, and Table 2 is an example of part of the time window data. The pure electric tug energy consumption prediction model is used to predict the energy consumption per minute and the energy consumption per nautical mile.

[0060] Table 2

[0061] In order to realize real-time energy consumption prediction of the pure electric tug, the above steps will be repeatedly executed every minute, and new time data windows will be repeatedly constructed by time window real-time sliding, so as to realize real-time energy consumption prediction of the pure electric tug.

[0062] Embodiment 2 On the basis of the technical scheme of the pure electric tug energy consumption prediction method provided in embodiment 1, further, A pure electric tug energy consumption prediction system is provided, comprising a data acquisition module, a working condition judgment module, a genetic algorithm optimization module and a prediction model deployment application module.

[0063] Embodiment 3 A shipborne terminal device carrying the pure electric tug energy consumption prediction model is provided.

[0064] ​In order to facilitate understanding of the technical effects of the present application, see the results of the present application and the prior art provided in Table 3. In this example, the GA-BiLSTM-based tug energy consumption prediction model has the highest R2 value and the lowest MSE, RMSE, MRE and MAE values. Compared with the LSTM model, the prediction accuracy variance of GA-BiLSTM is reduced by 22.07%, 11.79%, 20.95% and 27.91% respectively. Therefore, it is not difficult to conclude that genetic algorithm optimization can effectively improve the prediction performance of the BiLSTM-based model. In addition, compared with the ET, RF and XGBoost-based tug energy consumption prediction model, the GA-LSTM tug energy consumption prediction model constructed has better prediction accuracy. Table 3

[0065] In specific implementation, the method proposed in the technical scheme of the present application can be automatically run by a person skilled in the art using computer software technology, and the system device of the method, such as a computer readable storage medium storing the corresponding computer program of the technical scheme of the present application and a computer device including the corresponding computer program, should also be within the protection scope of the present application.

[0066] The genetic algorithm and deep learning-based pure electric tug energy consumption prediction device provided by the present application is described below. The genetic algorithm and deep learning-based pure electric tug energy consumption prediction device described below can be correspondingly referred to each other with the genetic algorithm and deep learning-based pure electric tug energy consumption prediction method described above.

[0067] On the other hand, the present application also provides an electronic device, which can include a processor, a communications interface, a memory and a communications bus, wherein the processor, the communications interface and the memory complete mutual communication through the communications bus. The processor can call the logical instructions in the memory to execute the above-mentioned genetic algorithm and deep learning-based pure electric tug energy consumption prediction method.

[0068] In addition, the logic instructions in the above-mentioned memory can be realized in the form of a software function unit and sold or used as an independent product, and can be stored in a computer readable storage medium. Based on such understanding, the technical solutions of the present application essentially or the parts that contribute to the prior art or parts of the technical solutions can be embodied in the form of a software product. The computer software product is stored in a storage medium, and includes a plurality of instructions for causing a computer device (which can be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the methods described in various embodiments of the present application. The aforementioned storage medium includes: a U disk, a mobile hard disk, a read-only memory (ROM, Read-Only Memory), a random access memory (RAM, Random Access Memory), a magnetic disk or an optical disk, and various media that can store program codes.

[0069] In another aspect, the present application also provides a computer program product, which comprises a computer program, the computer program can be stored on a non-transitory computer readable storage medium, and the computer program can be executed by a processor to enable a computer to execute the genetic algorithm and deep learning based pure electric tug energy consumption prediction method provided by the above-mentioned methods.

[0070] In another aspect, the present application also provides a non-transitory computer readable storage medium, which stores a computer program, and the computer program is executed by a processor to implement the genetic algorithm and deep learning based pure electric tug energy consumption prediction method provided by the above-mentioned methods.

[0071] The device embodiments described above are only schematic, wherein the units illustrated as separate components can or can not be physically separate, and the components illustrated as units can or can not be physical units, i.e., can be located in one place, or can be distributed on a plurality of network units. Part or all of the modules can be selected according to actual needs to achieve the purpose of the present embodiment. Those skilled in the art can understand and implement without creative labor.

[0072] From the above description of the embodiments, those skilled in the art can clearly understand that the embodiments can be realized by means of software plus necessary general hardware platforms, and of course can also be realized by hardware. Based on such understanding, the above technical solutions essentially or the parts that contribute to the prior art can be embodied in the form of a software product, which can be stored in a computer readable storage medium, such as a ROM / RAM, a magnetic disk, an optical disk, etc., and includes a plurality of instructions for causing a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods described in various embodiments or some parts of the embodiments.

[0073] Finally, it should be noted that the above examples are only used to illustrate the technical solutions of the present application, and are not intended to limit the same; although the present application has been described in detail with reference to the foregoing examples, those of ordinary skill in the art should understand that they can still modify the technical solutions recorded in the foregoing examples, or make equivalent replacements for some of the technical features; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present application.

Claims

1. A pure electric tugboat energy consumption prediction method based on genetic algorithm and deep learning, characterized in that: The following processes are included: Real-time collection of pure electric tugboat operating data, including electrical storage data, navigation data, and environmental parameters, and pre-processing of the operating data; Construct a training set containing working condition labels, including extracting the tugboat's energy consumption rate per unit distance based on the instantaneous energy consumption of the power warehouse and the ship's speed, and determining the working condition type by mapping the tugboat's energy consumption rate per unit distance and combining it with the tugboat's position information; Build a pure electric tugboat energy consumption prediction model based on bidirectional LSTM, use genetic algorithm to dynamically optimize model hyperparameters, and perform model training; The data is updated in real time through a sliding time window, and based on the trained pure electric tugboat energy consumption prediction model, the predicted values ​​of energy consumption per minute and per nautical mile are output synchronously.

2. The method according to claim 1, wherein: The battery storage data includes the remaining power of the battery storage, the battery storage current, and the voltage; the navigation data includes the propeller speed, power, ship position, ship speed relative to the ground, and heading; and the environmental parameters include wind speed, wind direction, flow speed, flow direction, and significant wave height.

3. The method according to claim 2, wherein: During preprocessing, data segments whose remaining battery capacity (SOC) exceeds the range of 0%-100% are deleted, and the sliding window mean is used to fill missing values ​​in the heading and wind direction data.

4. The method according to claim 1, wherein: Calculate the instantaneous energy consumption of the battery compartment and determine whether it is greater than 0. If so, the tugboat is judged to be in charging condition. Otherwise, calculate the energy consumption rate per unit distance. Use the sigmoid function to map the energy consumption rate per unit distance of the tugboat, and determine the working condition type in combination with the port location information.

5. The method according to claim 1, wherein: The working condition types include towing, pushing, sailing, port parking and charging working conditions, wherein the corresponding label code of the towing and pushing working conditions is 0, the corresponding label code of the sailing working condition is 1, the corresponding label code of the port parking working condition is 2, and the corresponding label code of the charging working condition is 3.

6. The method according to claim 1, wherein: The pure electric tugboat energy consumption prediction model based on bidirectional LSTM includes a bidirectional LSTM layer, a fully connected layer, a denormalization layer and a multi-objective prediction layer arranged in sequence. The multi-objective prediction layer outputs energy consumption predictions per nautical mile and per minute through a white box model.

7. The method according to claim 1, wherein: The genetic algorithm is used to dynamically optimize the model hyperparameters, which is achieved by initializing the population and randomly generating hyperparameter combinations; training the model with minimizing the validation set loss as the fitness function; iteratively optimizing through selection, crossover and mutation operations, and finally outputting the optimal hyperparameter combination.

8. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein: When the processor executes the program, the pure electric tugboat energy consumption prediction method based on genetic algorithm and deep learning as described in any one of claims 1 to 7 is implemented.

9. 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 pure electric tugboat energy consumption prediction method based on genetic algorithm and deep learning as described in any one of claims 1 to 7 is implemented.

10. A computer program product comprising a computer program, characterized in that: When the computer program is executed by a processor, the pure electric tugboat energy consumption prediction method based on genetic algorithm and deep learning as described in any one of claims 1 to 7 is implemented.

Images